Top Tips for Using AI in Your Business

WEBVTT

NOTE
This file was generated by Descript

00:00:02.650 --> 00:00:06.279
Howard Bentham: Hello there, I'm Howard
Bentham, and this is OxTalks, the

00:00:06.280 --> 00:00:10.690
podcast powered by OxLEP, the local
enterprise partnership for Oxfordshire.

00:00:11.090 --> 00:00:14.929
If you're new to these podcasts, OxTalks
aims to discuss current issues in

00:00:14.929 --> 00:00:19.660
business and examine areas of interest
by hearing the thoughts of some genuinely

00:00:19.690 --> 00:00:22.340
remarkable and inspirational leaders.

00:00:22.805 --> 00:00:27.294
My guests are always first to acknowledge
the invaluable support that OxLEP can

00:00:27.294 --> 00:00:31.945
provide and the difference it could
make to your company or organisation.

00:00:32.775 --> 00:00:36.585
As you would expect, our focus is
on Oxfordshire's businesses in these

00:00:36.585 --> 00:00:40.865
podcasts, but naturally you may well
be enjoying OxTalks wherever you are.

00:00:41.414 --> 00:00:44.775
With that in mind, please feel free
to join in the conversation and

00:00:44.775 --> 00:00:46.735
share any thoughts and feedback.

00:00:47.480 --> 00:00:50.469
Our social media is the
easiest way to get in touch.

00:00:50.583 --> 00:00:54.343
We are at Oxfordshire LEP on X
and Oxfordshire Local Enterprise

00:00:54.343 --> 00:00:55.733
Partnership on LinkedIn.

00:00:56.343 --> 00:00:59.623
You can raise a question for future
discussions, or leave your thoughts

00:00:59.633 --> 00:01:01.303
on the topics of conversation.

00:01:01.883 --> 00:01:05.233
Please use the email address on
the podcast description and we

00:01:05.233 --> 00:01:06.563
look forward to hearing from you.

00:01:07.393 --> 00:01:11.423
In this edition, we'll explore practical
advice for businesses who are looking

00:01:11.423 --> 00:01:16.253
to use AI technology to streamline
their operations and grow successfully.

00:01:16.833 --> 00:01:21.543
Finding reliable statistics to context
exactly where we are in the UK with the

00:01:21.543 --> 00:01:26.663
number of businesses using AI throws up
many conflicting numbers and in such a

00:01:26.663 --> 00:01:31.073
fast moving field, the numbers will almost
certainly become out of date very quickly.

00:01:31.353 --> 00:01:33.903
I'll ask our guests to give
us their view on the state of

00:01:33.903 --> 00:01:35.933
play with AI in the UK shortly.

00:01:36.573 --> 00:01:40.883
What is clear though is that AI is
here to stay and it's becoming more

00:01:40.883 --> 00:01:42.833
widespread across the business landscape.

00:01:43.253 --> 00:01:46.003
A growing number of organisations
are wanting to explore

00:01:46.023 --> 00:01:47.813
using AI in the future too.

00:01:48.373 --> 00:01:52.843
Some sectors are using AI to great
effect, notably telecoms, legal

00:01:52.873 --> 00:01:56.623
and finance, whereas others are
yet to even begin looking at what

00:01:56.653 --> 00:01:58.693
difference this technology could bring.

00:01:59.030 --> 00:02:03.850
A consultation published by the UK
government in February 2024 said that

00:02:03.990 --> 00:02:09.810
the AI industry in the UK employs
over 50, 000 people and contributes 3.

00:02:09.810 --> 00:02:15.140
7 billion to the economy, adding that, and
I quote, "Our universities produce some of

00:02:15.140 --> 00:02:19.780
the best AI research and talent, and the
UK is home to the third largest number of

00:02:19.780 --> 00:02:22.580
AI unicorns and start ups in the world".

00:02:23.066 --> 00:02:27.496
So many questions still swirl around the
topic of AI, not least what it actually

00:02:27.496 --> 00:02:32.691
is and how it works, will it lead to mass
job losses as technology makes many roles

00:02:32.691 --> 00:02:35.981
redundant, and crucially, is AI safe?

00:02:36.321 --> 00:02:40.501
With computers now being able to think
for themselves and the slew of very

00:02:40.501 --> 00:02:44.931
negative press on the subject, has
Pandora's box finally been opened?

00:02:45.406 --> 00:02:49.466
As a business leader, how can you use
this technology most effectively to

00:02:49.466 --> 00:02:51.746
streamline or grow my organisation?

00:02:52.110 --> 00:02:56.819
With so much to explore on the topic, we
opted not to speak to ChatGPT to get the

00:02:56.820 --> 00:03:02.090
answers, but rather to two great leaders
in the field of AI in the OxTalks studio.

00:03:02.410 --> 00:03:05.960
I'm delighted to welcome Grae
Laws, Managing Director of Beyond

00:03:05.970 --> 00:03:10.610
Touch, and Guy Gadney, CEO of
Charisma AI based in Oxford.

00:03:13.730 --> 00:03:18.240
Can we start with the basics to ensure
that everyone listening or watching to

00:03:18.240 --> 00:03:21.320
this episode is completely across this?

00:03:22.040 --> 00:03:24.900
In everyday language, if you
would, give us your personal

00:03:24.900 --> 00:03:28.539
definition on what AI actually is.

00:03:28.850 --> 00:03:31.860
Guy Gadney: Well, I think the first thing
to say is it's a very broad umbrella.

00:03:31.860 --> 00:03:37.099
It's sort of a technology and this
is a technology where a machine, so a

00:03:37.099 --> 00:03:41.340
computer, has learned something in the
sense that you've given it a bunch of

00:03:41.340 --> 00:03:46.086
data and out of that data, it's created
something different, something new.

00:03:46.356 --> 00:03:49.056
I think that's the simplest
way, you know, you put something

00:03:49.056 --> 00:03:50.166
in and you get something out.

00:03:50.256 --> 00:03:54.451
Maybe that's overly simple, but I think
if you look at it in terms of, I don't

00:03:54.451 --> 00:03:56.501
know, let's take an example like ChatGPT.

00:03:56.711 --> 00:03:59.901
Actually, in its simplest sense, what
you might be putting into that or

00:04:00.091 --> 00:04:03.581
a high school student or something
might be putting in, write me an essay

00:04:03.601 --> 00:04:08.891
on this particular piece and what's
happening there is that ChatGPT is sort

00:04:08.891 --> 00:04:14.075
of understanding what that question
is, which is using a combination of

00:04:14.075 --> 00:04:16.315
technologies like natural language
processing and machine learning.

00:04:16.515 --> 00:04:20.845
It's understanding what that student has
said and then it is sort of thinking about

00:04:20.845 --> 00:04:24.945
that in the sense of wanting to understand
what the student is then saying and then

00:04:24.945 --> 00:04:28.628
kicking out the essay that the student
wants at the end and through that, in

00:04:28.628 --> 00:04:32.388
the middle of that process is a little
bit of magic that sits on the computers.

00:04:32.628 --> 00:04:35.698
But that is the data that I said
that sits behind the whole piece.

00:04:35.698 --> 00:04:38.653
Howard Bentham: Yeah, it's that bit
of magic, isn't it, Grae, that is the

00:04:38.683 --> 00:04:42.253
difference between working something
out on a calculator or something really

00:04:42.253 --> 00:04:46.483
basic to where we're getting this,
if you like, Artificial intelligence,

00:04:46.483 --> 00:04:48.343
what it is, it's thinking for itself.

00:04:48.533 --> 00:04:49.533
What's your definition?

00:04:49.533 --> 00:04:50.583
Grae Laws: Yeah, I tend to agree.

00:04:50.643 --> 00:04:55.083
I think this is computer programmes
that mimic the way human beings

00:04:55.083 --> 00:04:59.133
research information, understand
it and process it and reproduce it.

00:04:59.563 --> 00:05:04.603
If you imagine that language models of
this day and age have scraped the entire

00:05:04.603 --> 00:05:09.243
internet, have taken databases and
sources of information that are so vast

00:05:09.553 --> 00:05:13.493
and have simply compressed all of that
down and used an algorithm, started to

00:05:13.493 --> 00:05:21.233
understand and sort, sift it, understand
what it means to use an NLP to actually

00:05:21.243 --> 00:05:24.993
process that and it takes a period of
time for these language models to actually

00:05:24.993 --> 00:05:26.573
get to a point where they're ready.

00:05:26.623 --> 00:05:29.833
This foundation model, whether
it's very narrow and it might be

00:05:29.833 --> 00:05:33.553
specialised in something like biology
or science or videography, or it

00:05:33.553 --> 00:05:35.243
could be very broad and general.

00:05:35.463 --> 00:05:39.368
But once those models are in place,
then we build applications and those

00:05:39.368 --> 00:05:43.748
applications draw from those language
models, they make inferences and they

00:05:43.748 --> 00:05:48.688
present something back to you, whether
it's a video, a piece of audio, whether

00:05:48.688 --> 00:05:53.918
it's text, any one of those things and you
know, the future generations of artificial

00:05:53.968 --> 00:05:59.008
intelligence will see lots of language
models being used and agents coming and

00:05:59.008 --> 00:06:02.458
dipping in and saying, I need a little
bit of information from that one and more

00:06:02.458 --> 00:06:06.188
from that one and it builds up a bigger
picture and then presents that back to

00:06:06.188 --> 00:06:07.718
you in a mode that you've asked for it.

00:06:08.273 --> 00:06:10.023
Howard Bentham: Guy, can you
expand on that a bit more?

00:06:10.133 --> 00:06:14.443
And how, if you like, a computer
learns from its previous experience?

00:06:14.563 --> 00:06:17.503
Any sort of examples you can
bring to the table there?

00:06:18.058 --> 00:06:18.968
Guy Gadney: Well, yes.

00:06:18.998 --> 00:06:21.178
I mean, I think the
definition is really good.

00:06:21.238 --> 00:06:23.348
I think there are two
sides to this coin, really.

00:06:23.348 --> 00:06:27.988
One is a model, an AI model and
what that is, this programme,

00:06:28.038 --> 00:06:29.101
it's a piece of software.

00:06:29.101 --> 00:06:32.951
It's a programme, and that needs
to be trained and there's one thing

00:06:32.951 --> 00:06:36.241
that's interesting about this is that
the language that we use with AI is

00:06:36.241 --> 00:06:40.651
very human, you know, we talk about
Artificial intelligence, we talk about

00:06:40.671 --> 00:06:43.471
training, we talk about ingesting models.

00:06:43.581 --> 00:06:46.901
It's very human, but in essence
what's happening is that there is a

00:06:46.901 --> 00:06:51.398
computer programme which is taking
a bunch of data in the instance

00:06:51.398 --> 00:06:54.388
of the things we know about like
ChatGPT or those sorts of things.

00:06:54.758 --> 00:06:59.442
It is broadly information that is
available on the internet and probably

00:06:59.442 --> 00:07:02.302
other areas as well that we can get
into later and the ethics around that.

00:07:03.022 --> 00:07:08.122
That is then parsed, in other words,
the software sort of is categorising

00:07:08.122 --> 00:07:13.352
it and it's categorising it by meaning,
by sentiment, by topic and pattern

00:07:13.402 --> 00:07:17.262
matching that and one of the things,
the key things that AI is very good at

00:07:17.342 --> 00:07:21.492
is this sense of pattern matching, of
identifying a pattern of language, a

00:07:21.492 --> 00:07:26.602
pattern of mathematics, a pattern of sort
of anything that we know as information

00:07:27.117 --> 00:07:30.497
and in essence, what it then does is
it then predicts what's going to happen

00:07:30.507 --> 00:07:33.557
next when it is prompted by a user.

00:07:33.587 --> 00:07:38.077
So it sort of predicts the next
step of a sentence, it predicts

00:07:38.077 --> 00:07:40.027
that a verb follows a subject.

00:07:40.267 --> 00:07:44.567
So there's this whole process called
training where these AI models are

00:07:44.567 --> 00:07:48.927
trained on large amounts of data and
then that model is, it goes through

00:07:48.927 --> 00:07:52.954
a process of fine tuning, which is
then, I suppose, narrowing it down,

00:07:52.974 --> 00:07:56.764
whittling it to the point that it needs
to focus on, and that again may be a

00:07:56.764 --> 00:08:00.974
subject matter fine tuning around a
particular topic like mathematics, or

00:08:00.974 --> 00:08:04.104
indeed in the creative industries it
might be around a particular book or a

00:08:04.104 --> 00:08:10.464
particular brand, to make sure that what
the output is remains specific to that.

00:08:10.474 --> 00:08:15.404
So in other words, you don't want
Batman talking like the Princess Bride,

00:08:15.744 --> 00:08:18.604
you know, you've got to have Batman
talking like Batman and understanding

00:08:18.604 --> 00:08:23.014
the world of Batman and keeping on
topic in that space because otherwise

00:08:23.014 --> 00:08:26.084
it breaks the storyline and that's
what the fine tuning process does.

00:08:26.224 --> 00:08:27.054
Howard Bentham: It's fascinating.

00:08:27.154 --> 00:08:30.634
I alluded to this in the introduction,
gents, that it's really difficult

00:08:30.654 --> 00:08:35.184
to get accurate figures on the
uptake of AI, certainly in the UK,

00:08:35.274 --> 00:08:36.764
as far as businesses are concerned.

00:08:37.317 --> 00:08:40.747
Can you give us an idea in percentage
terms of where we are, Grae, with this?

00:08:41.102 --> 00:08:43.162
Grae Laws: It's over 50
percent of businesses these

00:08:43.162 --> 00:08:44.952
days are regularly using AI.

00:08:45.405 --> 00:08:48.895
For you as an individual, as an employee
of an organisation, you've probably

00:08:48.895 --> 00:08:52.295
been using AI for the last decade
and you just didn't even realise it.

00:08:52.875 --> 00:08:55.755
You would have been using it with YouTube
with its predictions as to what you should

00:08:55.755 --> 00:09:01.435
look at next, when you went shopping on
Amazon, when you used your SatNav, when

00:09:01.435 --> 00:09:05.930
you used your Fitbit, your smartwatch,
your iPad, your average smartphone

00:09:05.940 --> 00:09:10.867
today has probably got 10, 15 different
small AI tools that are running in the

00:09:10.867 --> 00:09:16.147
background that are totally invisible
and as we see the next gen of the Samsung

00:09:16.147 --> 00:09:21.617
phone and the next gen of the iPhone
bringing out their AI tools, they've

00:09:21.617 --> 00:09:23.587
actually already been in there for years.

00:09:24.047 --> 00:09:27.407
So for businesses, the uptake has been
significant and it's certainly the

00:09:27.407 --> 00:09:32.517
case that it required an innovation
trigger in the public eye for it

00:09:32.517 --> 00:09:35.757
really to be taken seriously and I
think that was obviously ChatGPT.

00:09:36.402 --> 00:09:40.402
Howard Bentham: And where is the UK in
comparison to the rest of the world?

00:09:40.662 --> 00:09:42.052
Have you got a handle on that?

00:09:42.127 --> 00:09:44.397
Grae Laws: If you spoke to IBM,
they would say we're in third place.

00:09:44.817 --> 00:09:47.897
If you looked at the Freshworks
report, that was published last year,

00:09:47.897 --> 00:09:49.687
they put us in about ninth place.

00:09:49.847 --> 00:09:53.117
We're always in the top 10 and it
really comes down to how do you

00:09:53.127 --> 00:09:55.447
measure the uptake and readiness?

00:09:55.447 --> 00:09:58.607
Because, you know, if people are
already working in the industry, then

00:09:58.607 --> 00:09:59.867
they're using it all of the time.

00:10:00.237 --> 00:10:03.327
If people are just working in business,
then they don't even realise they're

00:10:03.347 --> 00:10:07.937
using it, but they are and I guess
the readiness part of this, and in

00:10:07.937 --> 00:10:11.527
terms of preparing applications that
are practicable and have use cases in

00:10:11.527 --> 00:10:16.507
business, well, that grows every single
day and there are certain measures of

00:10:16.507 --> 00:10:21.347
enablement that some of these reports use
to say, what's the government interest

00:10:21.347 --> 00:10:23.377
and investment into AI and funding?

00:10:23.587 --> 00:10:24.817
What does that landscape look like?

00:10:25.297 --> 00:10:28.697
Who are the most innovative
organisations in AI and where do

00:10:28.697 --> 00:10:30.627
they, you know, kind of operate from?

00:10:30.977 --> 00:10:37.147
What is the capacity for human capital
and access to data and energy that will

00:10:37.167 --> 00:10:39.147
drive this innovation in the industry?

00:10:39.287 --> 00:10:42.637
And I think all of those measures
play their part in that ranking,

00:10:43.111 --> 00:10:45.671
and there's a certain amount of
subjectivity that goes with it as well.

00:10:45.671 --> 00:10:48.721
Howard Bentham: Yeah, it's a very
broad umbrella term, isn't it?

00:10:49.111 --> 00:10:49.941
Very much so.

00:10:50.661 --> 00:10:53.461
Just while we're on the sort
of generics of this, what about

00:10:53.461 --> 00:10:57.581
the doomsday scenarios, which
we do hear about in the press?

00:10:58.001 --> 00:11:01.681
Are we looking at the 21st
century's equivalent of

00:11:02.121 --> 00:11:03.751
Oppenheimer's atomic bomb here?

00:11:03.921 --> 00:11:05.611
Guy Gadney: The short
answer is probably not.

00:11:05.681 --> 00:11:09.454
I think what has happened is
that, in my sort of work that I've

00:11:09.454 --> 00:11:12.854
done over the years, working in
technology and creativity, and that's

00:11:12.854 --> 00:11:16.174
been building websites, it's been
building e commerce sites, it's been

00:11:16.174 --> 00:11:17.434
building mobile apps and all of this.

00:11:17.714 --> 00:11:23.114
I've never had a strand of philosophy
come into the work that I do day to

00:11:23.114 --> 00:11:24.924
day as much as it has done with AI.

00:11:25.124 --> 00:11:30.119
You know, it's really struck a chord
somewhere and I think in some ways what

00:11:30.119 --> 00:11:34.209
people are thinking is that they're
thinking that AI is somehow related

00:11:34.209 --> 00:11:38.399
to the movies, the sci fi movies that
they've seen in the past and for the

00:11:38.399 --> 00:11:42.809
sense of drama, those movies have
often had a punch to them, whether

00:11:42.809 --> 00:11:46.969
that's Terminator or Robocop or,
you know, AI taking over the world.

00:11:46.989 --> 00:11:51.469
It's a good story, there's got to be a
protagonist and antagonist and it's easy

00:11:51.469 --> 00:11:55.592
to cast that antagonist as AI because
it's a sense of attacking, you know.

00:11:56.132 --> 00:11:59.052
Also, I think there is a sense,
especially in the press at the moment

00:11:59.072 --> 00:12:04.352
of AI replacing the journalists,
and writing articles and so forth.

00:12:04.352 --> 00:12:09.202
So naturally, the people who are covering
that are worried for their own jobs that

00:12:09.202 --> 00:12:12.412
they may be replaced and I think that
is skewing the argument a little bit.

00:12:12.982 --> 00:12:18.767
The more, I suppose doomsday scenario
that folks like Elon Musk have said,

00:12:18.767 --> 00:12:22.037
you know, that it, there needs to
be a kill switch in this sense, and

00:12:22.037 --> 00:12:24.097
that it could wipe out humanity.

00:12:24.517 --> 00:12:27.827
I think it is a strand of logic that
people go down and it's a little

00:12:27.827 --> 00:12:29.637
bit fanciful to be able to do that.

00:12:29.717 --> 00:12:32.127
AI still is a piece of
software in a machine.

00:12:32.467 --> 00:12:33.867
Howard Bentham: this
stage, they would argue.

00:12:33.867 --> 00:12:37.827
Guy Gadney: Exactly at this stage and
there are two ways to look at that.

00:12:37.837 --> 00:12:42.677
How one is, It hasn't happened, but
it might, you know and that's the one

00:12:42.687 --> 00:12:45.247
which sows a seed of doubt in our minds.

00:12:45.247 --> 00:12:47.107
Like, oh my god, are
we in control of this?

00:12:47.197 --> 00:12:48.907
It's going to run out of control.

00:12:49.427 --> 00:12:53.537
The other way to look at it is, we
control our own future and if we are

00:12:53.537 --> 00:12:56.597
building this, and we're responsible
for it, and ultimately, at the end of

00:12:56.627 --> 00:13:01.187
the day, this code that you see and
often in the presentations I do about

00:13:01.207 --> 00:13:06.357
AI, I show code on screen and say,
look, this is AI, you know, look, it's

00:13:06.387 --> 00:13:10.177
numbers and it's letters and everything,
this has been created by people.

00:13:10.467 --> 00:13:14.767
So people, we are in control of this,
and we can control our own future and how

00:13:14.767 --> 00:13:16.587
we chart the next steps for the future.

00:13:16.827 --> 00:13:19.767
I'm sure we'll come on to things like
regulation, which is one way to do it.

00:13:20.727 --> 00:13:25.087
But there is a sense that, you know,
in an increasingly automated set of

00:13:25.087 --> 00:13:31.657
societies where banking systems and indeed
infrastructure is controlled by computers

00:13:31.657 --> 00:13:35.257
and is controlled by computers that are
connected to the internet, there could

00:13:35.257 --> 00:13:38.427
be some form of negative impact on it.

00:13:38.617 --> 00:13:40.167
I mean, like serious negative impact.

00:13:40.467 --> 00:13:44.657
Actually, probably it's more likely
to be someone, you know, as a human

00:13:44.657 --> 00:13:46.817
actor who's done that, in my opinion.

00:13:46.932 --> 00:13:47.372
Howard Bentham: Sure.

00:13:47.892 --> 00:13:51.092
Just before we leave this topic,
Grae, your thoughts on the open letter

00:13:51.392 --> 00:13:55.142
from the Future of Life Institute
just over a year ago, now calling for

00:13:55.142 --> 00:13:57.622
the pause on giant AI experiments.

00:13:57.622 --> 00:14:00.772
So it was something like a
thousand leading experts in the

00:14:00.772 --> 00:14:02.052
field that signed that letter.

00:14:02.127 --> 00:14:03.467
Grae Laws: Yeah, it was
all the great and the good.

00:14:03.937 --> 00:14:09.107
You know, it feeds into this well known
Silicon Valley conversation point, P Doom,

00:14:09.597 --> 00:14:15.697
the probability of doom and this is a zero
to 100 measure of like, do you think AI

00:14:15.717 --> 00:14:18.027
will be an existential threat to humanity?

00:14:18.227 --> 00:14:19.057
Where do you sit on that?

00:14:19.357 --> 00:14:19.967
Zero?

00:14:20.237 --> 00:14:21.922
Not worried at all, forget about it.

00:14:21.932 --> 00:14:25.012
One hundred, it's literally going
to eat my lunch, we're doomed.

00:14:25.352 --> 00:14:26.992
So everybody sits on that scale.

00:14:26.992 --> 00:14:31.312
It's a subjective scale, but in actual
fact, lots of those people who were

00:14:31.322 --> 00:14:35.362
part of that open letter sat on more
than 50 percent in terms of their PDoom.

00:14:35.722 --> 00:14:40.022
So their belief that there is a higher
probability of it not being for good,

00:14:40.292 --> 00:14:44.402
but actually in actual fact being
exploited, by bad actors and they could

00:14:44.412 --> 00:14:48.522
be individuals, but they could also be
state actors as well who actually use

00:14:48.522 --> 00:14:54.172
this technology for suppression, for
manipulation, as indeed as well as the

00:14:54.172 --> 00:14:57.082
environmental impacts that come with
it and you know, some of these people

00:14:57.082 --> 00:14:58.897
were big players and it is still are.

00:14:59.307 --> 00:15:04.687
Geoffrey Hinton, arguably the godfather
of AI at Google and Mo Gawdat as

00:15:04.687 --> 00:15:09.667
well, another huge figure, both of
those genuinely felt and indeed still

00:15:09.907 --> 00:15:14.387
feel the same as Elon Musk that we
need to be really careful with this.

00:15:14.737 --> 00:15:18.377
That in actual fact, if we don't
have guardrails, we could be in

00:15:18.377 --> 00:15:22.257
trouble and so that was a premise
for the security summit in Bletchley.

00:15:22.812 --> 00:15:24.662
Guy Gadney: And actually building
on that, that as well, I think

00:15:24.662 --> 00:15:28.342
there's a lens I like to look at, at
AI3, which is that AI is the cause

00:15:28.382 --> 00:15:30.052
and then automation is the effect.

00:15:30.592 --> 00:15:33.662
So if you're going to look at the
impact of what AI could be, look

00:15:33.702 --> 00:15:37.472
at automation and then you'll start
to see what the impact could be.

00:15:37.532 --> 00:15:42.232
So for example, electoral interference
with AI, right, that happened actually

00:15:42.232 --> 00:15:45.423
was driven out of social media to
begin with, with people who are

00:15:45.423 --> 00:15:47.393
writing false posts onto social media.

00:15:47.743 --> 00:15:51.553
The problem is with the introduction of
AI, that process was automated and that

00:15:51.553 --> 00:15:56.693
meant that it could go faster, wider,
deeper and so as we start to look at

00:15:56.693 --> 00:15:59.803
that sort of interference and political
interference, social interference

00:15:59.903 --> 00:16:03.753
that we're getting, at least we can
then identify where the problem is

00:16:03.823 --> 00:16:05.403
and start to do something about it.

00:16:05.783 --> 00:16:10.113
Clearly there are, as in any new
technology, there are people who use it

00:16:10.113 --> 00:16:14.843
for good and there are people who use it
for bad and the counter to all of this,

00:16:15.153 --> 00:16:19.243
to my mind, is education and it sounds,
again, very simple, but if we understand

00:16:19.253 --> 00:16:22.753
the landscape that we're stepping into,
we've got the map for what it looks

00:16:22.753 --> 00:16:24.943
like, then we can react to it properly.

00:16:25.133 --> 00:16:29.793
The danger comes where we've got, you
know, no map and we're running sort

00:16:29.793 --> 00:16:35.633
of Silicon Valley speed primed by
unbelievable amounts of money as fast as

00:16:35.633 --> 00:16:40.433
we possibly can into a space that we don't
understand and we don't know and there

00:16:40.433 --> 00:16:44.303
I think there are dangers as there would
be in any other fast moving environment.

00:16:44.703 --> 00:16:47.613
Howard Bentham: Just, I was just
going to flip the coin if we could,

00:16:47.713 --> 00:16:52.223
and tell us how AI could change the
world for the benefit of mankind here,

00:16:52.223 --> 00:16:54.223
we've sat around P Doom for a while.

00:16:54.253 --> 00:16:54.693
Grae Laws: We have.

00:16:54.793 --> 00:16:56.373
Howard Bentham: Let's look
on the brighter side of life.

00:16:56.553 --> 00:16:59.563
What are some of the things that
this could genuinely change?

00:16:59.638 --> 00:17:05.568
Grae Laws: Well, think about research into
clinical trials and the benefits of using

00:17:05.568 --> 00:17:08.678
it to research medicine for greater good.

00:17:09.058 --> 00:17:13.018
Doctors have been using AI through
Google Health for more than a decade.

00:17:13.018 --> 00:17:18.248
So they've got a huge language model
that's built upon records of treatments

00:17:18.498 --> 00:17:22.583
and think of pharmaceutical companies
have been spending decades, literally

00:17:22.623 --> 00:17:28.703
decades researching and doing clinical
trials on cures for seemingly incurable

00:17:28.703 --> 00:17:31.673
diseases and one of the great things
about artificial intelligence,

00:17:31.673 --> 00:17:35.533
particularly the deep learning aspects
of it, is that it can bring together

00:17:35.573 --> 00:17:40.113
unconnected points that a human being
might not necessarily be able to do.

00:17:40.113 --> 00:17:41.353
Howard Bentham: Spotting those patterns.

00:17:41.353 --> 00:17:42.033
Grae Laws: Spotting those patterns.

00:17:42.848 --> 00:17:45.998
And it doesn't have quite such the
same cognitive load that a doctor

00:17:45.998 --> 00:17:50.028
may have or a researcher may have
in trying to retain that information

00:17:50.028 --> 00:17:52.968
in their short term memory and then
be able to apply that insight to a

00:17:53.158 --> 00:17:55.278
second parallel piece of research.

00:17:55.578 --> 00:17:59.438
AI can just do that and not only does it
do it well, but because it's part of a

00:17:59.438 --> 00:18:03.958
larger language model, it can be happening
simultaneously in lots of different

00:18:03.968 --> 00:18:08.838
research centers and all of that is
informing that initial foundation model

00:18:09.268 --> 00:18:11.048
and therefore they all learn from that.

00:18:11.048 --> 00:18:15.688
So there's exponential opportunities
for health care, a longevity of

00:18:15.688 --> 00:18:19.618
life, reducing infant mortality
and that's just healthcare, that's

00:18:19.618 --> 00:18:21.128
before we go into everything else.

00:18:21.538 --> 00:18:24.218
Manufacturing, industry,
leisure, education.

00:18:24.343 --> 00:18:26.603
Howard Bentham: Well, it's not often
in the introduction to an OxTalks we've

00:18:26.833 --> 00:18:28.373
ended the world and then saved it.

00:18:28.663 --> 00:18:30.173
So how about that, boys, that's...

00:18:30.173 --> 00:18:31.163
Guy Gadney: All four minutes.

00:18:31.633 --> 00:18:33.313
Howard Bentham: Very
impressive start, I must say.

00:18:33.593 --> 00:18:36.743
Well, thanks for sharing those
foundations of AI in general terms.

00:18:36.863 --> 00:18:38.759
Let's find out a little
bit more about you chaps.

00:18:39.129 --> 00:18:41.969
Guy, tell us about Charisma
AI and what you do.

00:18:42.364 --> 00:18:46.234
Guy Gadney: So our sector is the creative
industries, which, you know, if you think

00:18:46.244 --> 00:18:49.934
about what we've just talked about with
sci fi movies and everything has had an

00:18:49.934 --> 00:18:55.284
interest in AI for decades, from a very
imaginative perspective and ironically,

00:18:55.294 --> 00:18:59.694
it's also the industry that I think has
the most fear at the moment about AI.

00:18:59.804 --> 00:19:03.194
Hence, the writer strikes in Hollywood
last year, the actor strikes, there's

00:19:03.194 --> 00:19:05.744
a real sense of concern around it.

00:19:06.044 --> 00:19:08.944
This is an area I've worked in now
for almost a couple of decades

00:19:08.944 --> 00:19:13.254
in terms of using technology for
storytelling and creative purposes and

00:19:13.254 --> 00:19:15.844
to Grae's point around the different
sectors that people use it in.

00:19:15.864 --> 00:19:18.954
For me, there's a wonderful space
within the creative industries,

00:19:18.964 --> 00:19:20.694
which is two sides of the coin.

00:19:20.744 --> 00:19:24.004
So the first one is the one that people
talk around productivity, you know,

00:19:24.224 --> 00:19:26.174
can you do the same thing faster?

00:19:26.474 --> 00:19:30.624
Could you get your idea that you've
got for a great story that you want and

00:19:30.674 --> 00:19:35.174
you just want to tell that story, can
you get it out to your audience faster?

00:19:35.484 --> 00:19:39.944
And the short answer is yes and that's
the sort of systems that we're working.

00:19:40.659 --> 00:19:45.129
The other side to the coin, which is even
more exciting to me is, okay, what new

00:19:45.169 --> 00:19:47.149
can it do that we couldn't do before?

00:19:47.509 --> 00:19:51.716
So, you know, much in the same way that
any form of technology that has been

00:19:51.716 --> 00:19:54.826
involved in the creative industries and
by the way, the creative industries have

00:19:54.966 --> 00:20:00.946
always used technology, you know, back
to the ponds in amphitheatres in Greek

00:20:00.946 --> 00:20:06.191
theatre to reflect the audio and bounce it
out and amplify it, always used technology

00:20:06.191 --> 00:20:09.141
up through Lucas and Spielberg and VFX.

00:20:09.351 --> 00:20:14.171
So actually there's no change here and it
enables new forms of storytelling and for

00:20:14.171 --> 00:20:18.681
us, those forms of storytelling, I can
articulate it very simply, creatively,

00:20:18.681 --> 00:20:23.471
which is that you as an audience member,
rather than just sitting at home,

00:20:23.701 --> 00:20:29.916
watching your TV series of Sherlock or
whatever it may be, we can bring you into

00:20:29.956 --> 00:20:32.346
that TV series to influence the story.

00:20:32.346 --> 00:20:35.696
You might have Benedict Cumberbatch
turn to you and say, okay, Howard,

00:20:35.726 --> 00:20:39.456
you know, you're now part of my gang
and I want you to prove yourself

00:20:39.456 --> 00:20:40.856
as a detective alongside me.

00:20:41.156 --> 00:20:43.346
Did you spot anything in the
last scene that was relevant?

00:20:43.356 --> 00:20:46.546
And suddenly you're speaking to
the main characters involved in it.

00:20:46.556 --> 00:20:51.306
So creatively, it's like a new form of
storytelling that is like a hybrid version

00:20:51.306 --> 00:20:55.156
of Hollywood and games, actually and
that's very much the technology that we've

00:20:55.156 --> 00:20:58.656
been building is to enable writers to
write this sort of thing and innovate it.

00:20:58.921 --> 00:21:01.721
Howard Bentham: And the specific
area of the entertainment industry

00:21:01.721 --> 00:21:03.141
that you're in, just describe that.

00:21:03.386 --> 00:21:05.266
Guy Gadney: Well, we've
worked a lot with film and TV.

00:21:05.326 --> 00:21:09.179
I think that's been our core area and
that includes working with the likes

00:21:09.179 --> 00:21:13.189
of Hollywood Studios, NBC, Universal,
DreamWorks, Warner Brothers, you know,

00:21:13.189 --> 00:21:17.214
in LA, all of those guys who are looking
at bringing their characters to life,

00:21:17.424 --> 00:21:21.894
making them pop out of the silver screen
in new ways, as well as here with Sky TV

00:21:21.894 --> 00:21:26.264
we did an interactive drama for one of
their shows, the BBC extending tomorrow's

00:21:26.264 --> 00:21:27.608
world, actually as, as that show.

00:21:27.608 --> 00:21:29.948
we did a sort of interactive version
of that going into the future

00:21:29.948 --> 00:21:34.028
a while back, and with OxLEPs
support around lockdown period.

00:21:34.028 --> 00:21:38.643
We were fortunate enough to buy the
rights to John Wyndham's classic sci

00:21:38.833 --> 00:21:43.663
fi novel, The Kraken Wakes, you know,
of these wonderful 1950s classics,

00:21:43.683 --> 00:21:46.683
which are very organic, actually,
they're not technically focused.

00:21:46.963 --> 00:21:50.543
So we adapted that into a video game
environment where you are cast as

00:21:50.543 --> 00:21:54.668
a journalist in inside the story,
chronicling, I won't spoil it for

00:21:54.668 --> 00:21:58.408
people, but, the alien invasion and
how they can influence the story.

00:21:58.408 --> 00:22:02.544
what I love about what we do is that
it is a completely seamless hybrid

00:22:02.544 --> 00:22:04.218
of technology and storytelling.

00:22:04.448 --> 00:22:07.754
Howard Bentham: And people that have
the stories, the writers, how are

00:22:07.764 --> 00:22:10.894
they, when they come to you with
this, saying, I've got this great

00:22:10.894 --> 00:22:15.094
idea, it's an amazing story, how does
it marry with the technology, with

00:22:15.094 --> 00:22:16.684
the worries that we've talked about?

00:22:16.744 --> 00:22:19.924
Guy Gadney: What's so interesting is it
depends on the writer and we did, when

00:22:19.924 --> 00:22:23.944
we first started building Charisma AI as
a technology, we ran writer's labs with

00:22:24.004 --> 00:22:27.984
Story Futures at the National Film and
Television School and we had two days

00:22:28.004 --> 00:22:32.679
workshop with 25 really diverse writers
across, from different backgrounds,

00:22:32.679 --> 00:22:35.489
different skill sets, different parts of
the industry, they could have been film,

00:22:35.489 --> 00:22:37.539
they could have been video games, theatre.

00:22:38.259 --> 00:22:43.489
The writers who most were drawn
to this were theatre and immersive

00:22:43.489 --> 00:22:46.829
theatre writers, because of
their proximity to audience.

00:22:47.099 --> 00:22:51.339
So, you know, if you go to a theatre
show or any live show, there's the bit

00:22:51.339 --> 00:22:53.718
at the end where they say, thank you
very much, you've been a great audience.

00:22:53.988 --> 00:22:55.248
Well that means something.

00:22:55.258 --> 00:22:58.468
It means that they've felt there's
an energy, there's a proximity to it.

00:22:58.868 --> 00:23:02.868
So as writers are writing, you know, what
they want to do, they're thinking about

00:23:02.878 --> 00:23:07.868
the audience, maybe much more than say a
movie writer or a TV writer who's thinking

00:23:07.868 --> 00:23:10.218
more around what it looks like on screen.

00:23:10.768 --> 00:23:15.208
So for them, I think it was this
way to explore, you know, different

00:23:15.208 --> 00:23:18.738
forms of telling the same story and
thinking, Oh gosh, I can do this now.

00:23:18.938 --> 00:23:22.238
I can now actually bring this character to
life and play with them in different ways.

00:23:22.538 --> 00:23:27.948
So actually, I sometimes say there's an
irony in the further we look ahead to the

00:23:27.958 --> 00:23:31.918
future of technology and storytelling,
actually, the more we are influenced

00:23:31.918 --> 00:23:36.628
and inspired by the older, further back
media like theatre and immersive theatre.

00:23:37.168 --> 00:23:40.298
Howard Bentham: But there is a danger
that creatives, and hence the strike you

00:23:40.298 --> 00:23:45.018
talked about in the States, that AI could
potentially take away your livelihood.

00:23:45.108 --> 00:23:49.208
Voice Over actors, a story just recently,
the person was actually driving his car

00:23:49.208 --> 00:23:53.988
and heard his voice come on an ad and
it was his voice without his permission.

00:23:54.228 --> 00:23:59.594
So there are seemingly some worries around
that lack of safeguards, and you get to

00:23:59.594 --> 00:24:03.974
the situation where you're going to lose
any control over your own IP really.

00:24:04.404 --> 00:24:06.754
Guy Gadney: Yes and you know, there
was another example with Scarlett

00:24:06.754 --> 00:24:11.031
Johansson as well, her voice being used
by OpenAI and having to resort to legal

00:24:11.031 --> 00:24:16.471
means to stop that and it is a concern
and my view on that is that if we let

00:24:16.491 --> 00:24:21.591
things run unchecked, at the moment
I think there is a mismatch of values

00:24:22.021 --> 00:24:25.471
between the technology industries and
the creative industries, where they're

00:24:25.471 --> 00:24:29.301
looking at each other with a lack of
understanding, and it's the first time

00:24:29.331 --> 00:24:33.301
I think that there's been this sort of
break between the two industries and

00:24:33.301 --> 00:24:37.551
the technology companies are sort of
running ahead without understanding

00:24:37.561 --> 00:24:39.371
the values of the creative industries.

00:24:39.641 --> 00:24:44.201
They're also not understanding the
value of the work that is being

00:24:44.201 --> 00:24:46.481
produced and the people who produce it.

00:24:46.881 --> 00:24:48.511
So there is work that we need to do there.

00:24:48.831 --> 00:24:52.251
Our view is that we are, you
know, really strongly trying

00:24:52.251 --> 00:24:53.611
to bridge these two industries.

00:24:53.631 --> 00:24:57.761
We're trying to create the bridge and
funnily enough, I sit on an advisory

00:24:57.761 --> 00:25:03.321
group run by Innovate UK, called Bridge
AI, representing the creative industries

00:25:03.321 --> 00:25:09.281
so that our voice is heard, so that
people do not rip off copyright or real

00:25:09.281 --> 00:25:13.881
people's work and that the influence
of the creative industries, as humans,

00:25:13.881 --> 00:25:17.971
as practitioners and creatives, is
heard and is felt as we move forward.

00:25:18.241 --> 00:25:20.790
You know, let's not make any mistake
about it, the creative industry is

00:25:20.790 --> 00:25:23.562
responsible for huge amounts of GDP
within the UK, and that's not to say that

00:25:23.562 --> 00:25:25.741
we don't have a lot of and worldwide.

00:25:25.991 --> 00:25:28.861
Without it, ChatGPT
simply would not exist.

00:25:29.021 --> 00:25:32.371
A lot of the AI that we see has
been trained on creative work.

00:25:32.701 --> 00:25:37.661
So there needs to be that value equation
that is more balanced than it is now

00:25:37.981 --> 00:25:41.371
and my sense is that, you know, we've
seen this writer strikes, we've seen

00:25:41.651 --> 00:25:45.351
the passion coming out of it and now
there needs to be, as Greg was talking

00:25:45.361 --> 00:25:49.711
about guardrails, there needs to be
a set of rules, but also the creative

00:25:49.711 --> 00:25:55.031
industries needs to engage and I don't
think that the answer is to block AI

00:25:55.031 --> 00:25:59.101
and tried to block things that generally
over the last few decades has not

00:25:59.101 --> 00:26:00.731
worked in technology advancements.

00:26:00.751 --> 00:26:06.961
It's very difficult and is not a strategy,
but to engage and to engage in the

00:26:06.961 --> 00:26:11.551
creation of these new technologies from
the creative industries is the way that

00:26:11.571 --> 00:26:15.831
we see the world and we hope that message
is going to expand over the coming years.

00:26:15.901 --> 00:26:19.881
Howard Bentham: Grae, you started your
own business, Beyond Touch in 2013.

00:26:19.881 --> 00:26:22.571
You're helping business owners
transform their organisation by,

00:26:22.911 --> 00:26:25.971
and I quote here, thinking lean,
staying energised and adapting to

00:26:25.971 --> 00:26:28.801
change, to quote a mantra of yours.

00:26:29.411 --> 00:26:30.936
Tell us a bit more about what you do.

00:26:31.346 --> 00:26:36.316
Grae Laws: So I do business mentoring
and group training and in essence,

00:26:36.316 --> 00:26:40.206
we're trying to help people unpack
their ideas and realise the true

00:26:40.206 --> 00:26:41.456
value of what they're trying to do.

00:26:41.456 --> 00:26:45.636
Whether it's, I've got an initial idea
and I want to start a business, or I'm

00:26:45.636 --> 00:26:50.396
already in business, but I've identified a
barrier, or I've got an issue that exists

00:26:50.406 --> 00:26:54.906
within my organisation and I just need to
work out how to overcome that challenge.

00:26:55.456 --> 00:27:00.146
So we work with OxLEP, other growth
hub organisations, universities,

00:27:00.176 --> 00:27:04.676
government agencies, to help people
with their business challenges.

00:27:05.121 --> 00:27:09.291
We do that through sessions that
involve artificial intelligence,

00:27:09.641 --> 00:27:13.461
as well as project management and
operations process management, and

00:27:13.461 --> 00:27:15.091
really getting into the detail of it.

00:27:15.101 --> 00:27:19.441
So when we do mentoring, and when we
sit down with individuals, it's an

00:27:19.521 --> 00:27:23.681
unpacking exercise, we're a trusted
friend, we're probing into the

00:27:23.681 --> 00:27:25.121
real challenge that sits behind it.

00:27:25.121 --> 00:27:29.101
We're asking why and get into the root
cause of it and then we're helping

00:27:29.101 --> 00:27:31.161
them to build action plans for change.

00:27:31.391 --> 00:27:35.231
So it's all very action oriented
and so for some business owners, one

00:27:35.231 --> 00:27:38.541
of the biggest challenges can be, I
don't have anybody that I can talk to.

00:27:38.878 --> 00:27:42.338
I can't admit that it's not
going right, or I can't admit

00:27:42.338 --> 00:27:43.438
that I don't have the answer.

00:27:43.448 --> 00:27:46.478
I can't talk to my family about it,
I don't want to talk to my colleagues

00:27:46.478 --> 00:27:50.248
about it, I most certainly can't mention
it to my employees, where do I go?

00:27:51.158 --> 00:27:55.218
That's the value of a good business mentor
and through organisations like OxLEP who

00:27:55.218 --> 00:28:01.718
provide those access to those mentors,
it can provide a sense of relief to those

00:28:01.718 --> 00:28:07.418
business owners and more importantly,
an opportunity to refocus them, energise

00:28:07.438 --> 00:28:10.898
them and send them off in the right
direction for them to achieve their goals.

00:28:10.943 --> 00:28:13.503
Howard Bentham: And focusing
specifically on AI here, we said

00:28:13.503 --> 00:28:17.133
with those stats, 50 percent uptake,
broadly speaking, are still 50

00:28:17.133 --> 00:28:19.293
percent that are not taking it up.

00:28:19.909 --> 00:28:23.629
What sort of noises are you
hearing back on the technology?

00:28:24.319 --> 00:28:26.559
Grae Laws: What I'm hearing is people
want to try it, but they don't know how.

00:28:27.209 --> 00:28:30.669
So when I do my mentoring, I
make a conscious decision to

00:28:30.669 --> 00:28:32.259
use AI during those meetings.

00:28:32.259 --> 00:28:35.839
They get advanced warning of
that, but I use various tools.

00:28:35.906 --> 00:28:39.036
My two favorite tools that I'll use
on many of my sessions is Fathom.

00:28:39.526 --> 00:28:43.156
So Fathom is an automated
meeting assistant.

00:28:43.346 --> 00:28:48.001
It comes into my meetings, it takes
notes, it records the meeting, it builds

00:28:48.041 --> 00:28:51.581
action plans, it has an AI tool built
into the back end of it, so when the

00:28:51.581 --> 00:28:56.161
meeting's finished, we can unpack more
information, gain insight from that,

00:28:56.401 --> 00:28:59.821
develop a plan, all of that information
goes on to the organisation that I've

00:28:59.821 --> 00:29:03.251
been working with, so they get a copy
of it, when we meet up a second time,

00:29:03.251 --> 00:29:07.341
we'll go back to that tool and say, okay,
let's go back to our previous meeting.

00:29:07.571 --> 00:29:09.001
Let's look at the actions that we agree.

00:29:09.021 --> 00:29:10.441
Let's gain some more insight.

00:29:10.451 --> 00:29:14.621
Let's talk, let's use the AI tool from
our previous session to help drive it

00:29:14.621 --> 00:29:19.511
forward and the other tool that I use
in all of my sessions, which again is a

00:29:19.731 --> 00:29:24.414
way to a soft entry point, really, for
many people who want to give it a go, but

00:29:24.414 --> 00:29:27.024
just don't understand the relevance of
it and why that would make any difference

00:29:27.364 --> 00:29:29.064
and that's a tool called Perplexity.

00:29:29.584 --> 00:29:32.624
Which is quite similar to
ChatGPT or Claude and all the

00:29:32.624 --> 00:29:33.444
other ones that are out there.

00:29:33.814 --> 00:29:40.094
My preference for using Perplexity is that
it feels more reliable, it's more fact

00:29:40.124 --> 00:29:45.104
based, it has less hallucinations, which
is a tendency of some AI models to come

00:29:45.104 --> 00:29:49.154
up with an answer if it can't actually
retrieve something that statistically is

00:29:49.154 --> 00:29:50.534
accurate, it will just make something up.

00:29:50.924 --> 00:29:52.574
So Perplexity tends not to do that.

00:29:53.014 --> 00:29:55.614
It always provides you
with reliable links.

00:29:55.614 --> 00:29:59.575
Howard Bentham: The thought of a computer
having an hallucination is quite worrying.

00:29:59.734 --> 00:30:01.814
Grae Laws: Well, it's part
of the P Doom measurement.

00:30:01.824 --> 00:30:02.144
Exactly.

00:30:02.214 --> 00:30:04.174
It really is, and that's
one of the big concerns.

00:30:04.174 --> 00:30:06.244
I mean, all of the AI companies
now are working on that.

00:30:06.524 --> 00:30:10.004
You think about the new artificial
intelligence model that's been developed,

00:30:10.044 --> 00:30:12.624
Apple Intelligence that's gone on
to the next generation of iPhones.

00:30:12.894 --> 00:30:17.134
They've got built into the core
of that to only use information

00:30:17.134 --> 00:30:21.174
from the language model and that
limits the ability for it to just

00:30:21.174 --> 00:30:22.934
hallucinate and just make things up.

00:30:23.329 --> 00:30:26.929
But anyway, coming back to perplexity,
why I love this tool is that, you know,

00:30:26.929 --> 00:30:28.209
I'll give you a really good example.

00:30:28.419 --> 00:30:32.259
I was doing a one to one session with
an organisation recently and this was

00:30:32.259 --> 00:30:36.526
to develop a matchmaking application
for B2B, business to business and

00:30:36.646 --> 00:30:39.226
this person had two challenges
that they brought to that session.

00:30:39.536 --> 00:30:42.356
The first challenge was, I've
done, I've written a website, but

00:30:42.366 --> 00:30:43.436
I just don't think it's any good.

00:30:43.446 --> 00:30:47.906
I think it might be a bit rubbish, I just
don't know and the other one was, I've got

00:30:47.906 --> 00:30:52.691
an app, I've got a whole lot of processes,
I've got a design, I feel like there's

00:30:52.691 --> 00:30:55.211
IP here and I just don't know what IP.

00:30:55.211 --> 00:30:57.791
I've had a look online, you
know, at the IP office and I

00:30:57.791 --> 00:30:58.911
just don't know what to do.

00:30:59.201 --> 00:31:01.511
So I said, I'll tell you what,
let's jump into Perplexity

00:31:01.561 --> 00:31:03.491
and we'll nail both of those.

00:31:03.771 --> 00:31:06.291
So we did, we jumped into
Perplexity, we did the IP one.

00:31:06.291 --> 00:31:10.221
In actual fact, because Perplexity has its
own language model, but most importantly,

00:31:10.221 --> 00:31:15.351
it seeks to find live examples,
real world sources of information,

00:31:15.371 --> 00:31:17.311
reliable sources of information.

00:31:17.741 --> 00:31:21.911
The natural factor becomes an extremely
useful tool for analysis and because

00:31:21.911 --> 00:31:26.351
you're having a dialogue and I do
encourage people, by demonstration,

00:31:26.771 --> 00:31:29.381
to start the conversation and
keep that conversation going.

00:31:29.381 --> 00:31:33.601
We were able to identify which
clauses within which types of IP were

00:31:33.601 --> 00:31:37.661
appropriate and the reasons why, we're
able to then say, condense all of

00:31:37.661 --> 00:31:41.451
this into a table, give me the links
to the IP office so I can do more

00:31:41.451 --> 00:31:45.051
reading about that and then we repeated
that same process for the website.

00:31:45.391 --> 00:31:46.931
Which was, who is best in class?

00:31:46.931 --> 00:31:47.971
Who are your competitors?

00:31:48.231 --> 00:31:49.841
Well, there are quite a few,
I don't really know who The

00:31:49.841 --> 00:31:52.431
most obvious competitors are,
okay, let's ask perplexity.

00:31:52.701 --> 00:31:56.731
Give me the top 10 UK companies
that do this, this and this.

00:31:57.281 --> 00:31:57.961
It provided them.

00:31:57.961 --> 00:32:01.151
So now we need to understand more the
value proposition of each of these

00:32:01.151 --> 00:32:07.441
organisations and how that will be
of benefit to this person, this user

00:32:07.441 --> 00:32:10.191
group, put it into a table and it did.

00:32:10.471 --> 00:32:13.181
So there you go, there's your
first bit of competitive analysis.

00:32:13.651 --> 00:32:17.001
So now, and it's got the links to their
websites, why don't you, maybe as an

00:32:17.011 --> 00:32:20.191
action out the end of this, go away
and just have a look at some of these

00:32:20.191 --> 00:32:24.831
websites and then reflect on what you've
put across in your website and let's

00:32:24.831 --> 00:32:28.511
use best practice to actually make sure
that we can give you a fighting chance

00:32:28.511 --> 00:32:33.481
to be a comparator or exceed expectation
when people are looking for you and may

00:32:33.481 --> 00:32:35.161
have found another one to compare with.

00:32:35.561 --> 00:32:37.531
Howard Bentham: So clearly,
the tools are out there.

00:32:37.651 --> 00:32:38.141
Grae Laws: They're there.

00:32:38.261 --> 00:32:42.576
Howard Bentham: It's that leap
from, okay, I'm the business owner

00:32:42.746 --> 00:32:45.970
thinking, right, yeah, I need to
do this, but who do I speak to?

00:32:45.970 --> 00:32:47.750
What do I do now?

00:32:48.046 --> 00:32:49.586
Guy, what would be your advice?

00:32:49.756 --> 00:32:51.846
Guy Gadney: I'd speak to Grae.

00:32:51.846 --> 00:32:53.426
Howard Bentham: Well, let's do that then!

00:32:53.766 --> 00:32:57.333
Guy Gadney: I think, I think
the first thing is to work out

00:32:57.363 --> 00:32:58.763
actually what your business does.

00:32:59.213 --> 00:33:02.603
With any new technology or anything
like this, you can be seduced by

00:33:02.623 --> 00:33:07.613
bright, shiny things and oh, look, I
should be using this and it can deviate

00:33:07.743 --> 00:33:10.723
away from what your core purpose is.

00:33:11.393 --> 00:33:14.828
Now the flip side to that coin is
maybe you do need to change your core

00:33:14.828 --> 00:33:19.038
business if it's not working, and AI can
give you that pattern matching tools.

00:33:19.038 --> 00:33:23.168
I mean, Grey's example was brilliant,
which is the big power struggle at

00:33:23.168 --> 00:33:24.788
the moment is around search engines.

00:33:25.118 --> 00:33:29.978
So Google has its own AI and
then ChatGPT has SearchGPT.

00:33:30.258 --> 00:33:34.268
Now the reason for doing that is that
if you think about how you search

00:33:34.268 --> 00:33:38.718
or have done to date, you get a list
of responses on that first page.

00:33:39.278 --> 00:33:42.188
What AI is going to do is going to
do exactly as Grae said, which is

00:33:42.188 --> 00:33:46.768
synthesise all of that, work out what
actually the key insights you want out

00:33:46.768 --> 00:33:48.538
of that and then write it up for you.

00:33:48.858 --> 00:33:52.818
So suddenly you're saving shed loads
of time of moving between different

00:33:52.818 --> 00:33:55.598
pages, pages one, two, three, and
all of that and you're getting

00:33:55.598 --> 00:33:56.988
that information in a heartbeat.

00:33:56.988 --> 00:33:58.573
Howard Bentham: It's doing the
heavy lifting, isn't it, really?

00:33:58.603 --> 00:34:01.603
Guy Gadney: Absolutely, and providing
that little bit extra as well, I think

00:34:01.603 --> 00:34:05.573
is providing you something that you may
not have thought about an angle on your

00:34:05.573 --> 00:34:07.063
business, which you may not have seen.

00:34:07.383 --> 00:34:11.513
So I think to know what your own strengths
are as a business owner, what you are

00:34:11.513 --> 00:34:17.475
in business to do is the core and then
use this as a way to provide further

00:34:17.475 --> 00:34:21.665
insights into that and the pattern
matching piece, I cannot overemphasise

00:34:21.685 --> 00:34:26.325
enough, this is the core of AI, the
ability to spot patterns and if you can

00:34:26.325 --> 00:34:31.905
spot patterns, you can start to see where
opportunities are in the future and where

00:34:31.945 --> 00:34:36.385
issues may be arising and if you can
do that, then you're going to de risk

00:34:36.385 --> 00:34:37.880
your business and that's a good thing.

00:34:37.995 --> 00:34:40.775
Howard Bentham: I can hear the audience
shouting that how much does it cost?

00:34:41.195 --> 00:34:42.005
Is this expensive?

00:34:42.005 --> 00:34:43.015
Grae Laws: No, it's free!

00:34:44.635 --> 00:34:48.135
And that's one of the greatest challenges
for all of these kind of developers

00:34:48.135 --> 00:34:51.945
of software and those who are spending
millions and billions on their large

00:34:51.945 --> 00:34:54.945
language models and things like that
is actually the business case is one

00:34:54.945 --> 00:34:59.605
where in which it's almost all free with
literally every application that I've

00:34:59.605 --> 00:35:04.205
ever tried, there's a free version and
it might limit you to how many iterations

00:35:04.215 --> 00:35:08.715
you can do of whatever this tool is and
that might be per day, per hour, per week,

00:35:08.715 --> 00:35:12.315
per month, whatever it is, but there's
always a free trial version of it and

00:35:12.315 --> 00:35:17.945
there's usually a lite version of it,
so the classic ones of Claude and Gemini

00:35:17.995 --> 00:35:24.620
and OpenAI's ChatGPT and Perplexity,
all of those have a free version, a

00:35:24.620 --> 00:35:28.090
lite version, but you can pay and I
wouldn't want to go on camera because

00:35:28.090 --> 00:35:31.760
my business partners listening, I'm sure
and they would say, you're spending that

00:35:31.760 --> 00:35:33.960
much on your AI tools, are you Grae?

00:35:33.980 --> 00:35:34.520
Okay.

00:35:34.910 --> 00:35:38.370
But I do subscribe and have
paid for accounts because I see

00:35:38.370 --> 00:35:39.940
the value in them, I really do.

00:35:40.340 --> 00:35:42.060
But we're talking tens
of pounds per month.

00:35:42.630 --> 00:35:47.530
Guy Gadney: And this presents a problem
for the UK, you know, taking it suddenly

00:35:47.530 --> 00:35:52.160
very high up the level, which is that
the reason that those companies can

00:35:52.160 --> 00:35:57.000
afford to do that, especially OpenAI, is
because they've got the funding to do it.

00:35:57.320 --> 00:36:00.740
So the cash burn of
OpenAI is eye watering.

00:36:01.050 --> 00:36:06.010
I believe it's something like five billion
a year that they are spending and not

00:36:06.050 --> 00:36:09.670
getting back in revenue, but they can
afford to do it because they have the

00:36:09.670 --> 00:36:14.190
venture capital backing to do it and in
fact, you know, recently Apple came in

00:36:14.500 --> 00:36:19.100
alongside Microsoft to help out with that
cash burn because it was looking wobbly

00:36:19.415 --> 00:36:22.915
for a moment, which is interesting,
given, you know, how much we're starting

00:36:22.915 --> 00:36:27.415
to rely on these tools and for the UK,
because we don't have that level of

00:36:27.425 --> 00:36:32.445
financial support, we don't have that
gearing in our venture capital, it means

00:36:32.445 --> 00:36:37.460
we struggle as a nation to create our
own Open AI, or our own perplexity, or

00:36:37.460 --> 00:36:39.110
our own Anthropic, or whatever it may be.

00:36:39.560 --> 00:36:45.600
So, I think as we start to look at how
the government looks at AI, what its

00:36:45.610 --> 00:36:49.030
strategies should be, we need to be
able to create an environment where

00:36:49.030 --> 00:36:55.125
we can form a company that can give
away their software, because they've

00:36:55.125 --> 00:36:58.345
got the cash runway to do it and, you
know, remember that every time you put

00:36:58.345 --> 00:37:02.165
something into ChatGPT and you're not
paying for it, it's doing two things.

00:37:02.195 --> 00:37:05.945
One, it's costing OpenAI and
the other one is that it's using

00:37:05.945 --> 00:37:07.225
your data that you put into it.

00:37:07.235 --> 00:37:09.195
So there's a risk reward piece here.

00:37:09.845 --> 00:37:14.365
But I think as a country for us to
be able to compete on that level,

00:37:14.915 --> 00:37:18.585
internationally and create those
sorts of competing companies, we need

00:37:18.585 --> 00:37:22.915
to slightly reframe, I think, our
investment to enable that sort of user

00:37:22.915 --> 00:37:24.475
acquisition, which is what it's doing.

00:37:24.525 --> 00:37:27.415
The reason that these things are
being given away for free is to

00:37:27.415 --> 00:37:31.615
acquire users that you can then
flip over into paying customers and

00:37:31.615 --> 00:37:34.305
at the moment, we don't have that
environment in this country to do that.

00:37:35.115 --> 00:37:38.175
Howard Bentham: That's really interesting,
just the macro economics of it, how

00:37:38.585 --> 00:37:42.115
these organisations are actually going
to make money as probably a conversation

00:37:42.115 --> 00:37:45.895
for another podcast I'd say, but
again, focusing on our audience here

00:37:46.005 --> 00:37:51.235
at OxLEP as well with trying to get
that information and indeed the sort

00:37:51.725 --> 00:37:57.350
of avenue to the expertise and the
funding, a conversation with OxLEP is a

00:37:57.350 --> 00:37:58.810
really good place to start, you'd say.

00:37:58.935 --> 00:37:59.485
Grae Laws: A hundred percent.

00:37:59.685 --> 00:38:03.570
I would absolutely agree and we've
got great organisations that are in

00:38:03.570 --> 00:38:06.990
Oxfordshire who have a background in AI.

00:38:07.020 --> 00:38:11.510
We've got some excellent speakers, some
incredible minds, you know, some of

00:38:11.510 --> 00:38:15.110
the greatest innovation that's actually
come out and certainly in terms of where

00:38:15.110 --> 00:38:19.510
education crosses over with industry and
AI has come out of Oxfordshire and, you

00:38:19.510 --> 00:38:24.094
know, really we build on experience and
you know, in a way we are all language

00:38:24.114 --> 00:38:28.614
models as human beings and we build
and we soak up through us osmosis our

00:38:28.614 --> 00:38:32.804
understanding of the industry that we're
in and it moves so quickly that you need

00:38:32.804 --> 00:38:36.884
to speak to people who have their finger
on the pulse, because I can recall the

00:38:36.884 --> 00:38:42.654
very first OxLEP workshop that I did on
Artificial Intelligence coincided with Sam

00:38:42.654 --> 00:38:47.864
Altman being kicked out of OpenAI and then
being brought back in and then leaving and

00:38:47.864 --> 00:38:51.554
then coming back and it was literally days
before I was doing a workshop and I had

00:38:51.624 --> 00:38:54.944
the whole thing set up, all of my slide
deck ready, everything was there, they

00:38:54.944 --> 00:38:58.684
had the whole history of AI, where the
investment was, who the big players were,

00:38:58.684 --> 00:38:59.964
and I had to keep going back into it.

00:38:59.964 --> 00:39:00.894
So right, okay, he's out.

00:39:00.914 --> 00:39:04.754
No, I got back into it, he's back
in again, but it moves that quickly

00:39:05.074 --> 00:39:07.804
and it's certainly the case that I
would say to any business that, you

00:39:07.804 --> 00:39:11.824
know, if you want to do your research
into this, maybe go to YouTube.

00:39:12.334 --> 00:39:15.294
That is one of the biggest
sources of information that I use.

00:39:15.294 --> 00:39:22.074
I follow people like Peter Diomedes and
Matt Vidpro, Matt Wolf, to name but three

00:39:22.074 --> 00:39:26.464
organisers or individuals who really
have their finger on the pulse and as you

00:39:26.464 --> 00:39:29.870
were saying, you know, a little earlier
about how people consume podcasts, you

00:39:29.870 --> 00:39:34.244
know, I listen, I don't necessarily
watch, but I actually play along because

00:39:34.244 --> 00:39:37.874
they actually say that the next five
business applications that have just

00:39:37.874 --> 00:39:39.394
been released, it's this, this, and this.

00:39:39.554 --> 00:39:43.084
I'm going to go through each one in turn
and I am there with a browser open and

00:39:43.084 --> 00:39:46.694
I am following along and trying it out
for myself and there's always a free

00:39:46.714 --> 00:39:49.587
version of it and some of them I just
bash them up and say, nah, rubbish, don't

00:39:49.587 --> 00:39:52.277
want to use that one, others I think,
ooh, actually that could be a keeper

00:39:52.537 --> 00:39:55.667
and then I can bring that information
then in to conversations that we're

00:39:55.667 --> 00:39:59.042
having with business leaders, you
never stare into infinity when somebody

00:39:59.042 --> 00:40:00.882
says, which AI tool should I use.

00:40:01.272 --> 00:40:05.102
But it's certainly based on
very, very recent information.

00:40:05.794 --> 00:40:08.064
Howard Bentham: It's a good time to
bring in OxLEP's Communications Manager,

00:40:08.064 --> 00:40:09.624
Rob Panting, into the conversation.

00:40:09.624 --> 00:40:13.414
Rob, what are businesses telling
you about incorporating AI

00:40:13.424 --> 00:40:15.834
into their everyday operations?

00:40:15.904 --> 00:40:18.764
What about the demand to understand
and implement here in Oxfordshire?

00:40:18.984 --> 00:40:19.544
Rob Panting: Thanks, Howard.

00:40:19.554 --> 00:40:21.344
Fascinating discussion already.

00:40:21.344 --> 00:40:25.806
I mean, it personally picked up a great
deal of knowledge and understanding

00:40:25.806 --> 00:40:29.366
as to what AI can offer businesses and
can offer individuals this morning.

00:40:29.746 --> 00:40:35.859
I think from a small business perspective,
which obviously generally is the audience

00:40:35.869 --> 00:40:40.909
that OxLEP works with, I go back to the
Business Support Tool, our online triage

00:40:40.919 --> 00:40:45.115
tool that we oversee, which gives us the
understanding and the finger on the pulse

00:40:45.125 --> 00:40:46.915
in terms of what businesses are asking.

00:40:47.385 --> 00:40:52.925
In terms of AI as a individual topic, we
don't tend to have a great deal of people

00:40:52.925 --> 00:40:58.415
coming forward saying, you know, we wish
to understand AI, more efficiently per se.

00:40:58.980 --> 00:41:03.163
But I think some of the topics that we
mentioned earlier on, things like growing

00:41:03.163 --> 00:41:07.863
customer base, things like managing
finance, things like developing better

00:41:07.923 --> 00:41:09.573
marketing communication strategies.

00:41:09.763 --> 00:41:12.826
AI is very much embedded in
all of those desires for small

00:41:12.826 --> 00:41:14.861
businesses to achieve those goals.

00:41:15.101 --> 00:41:18.531
So I think people are asking about
AI and asking how they can implement

00:41:18.531 --> 00:41:22.584
it in their business, not necessarily
coming forward and saying, we need to

00:41:22.584 --> 00:41:26.094
find out more about what artificial
intelligence can offer our business.

00:41:26.114 --> 00:41:31.764
But in the context of growing customer
base, managing our finance better in the

00:41:31.764 --> 00:41:34.634
context of that and implementing AI from
that point of view, I think there's a

00:41:34.634 --> 00:41:38.977
huge amount of interest and Grae is very
much sort of delivering that support

00:41:38.997 --> 00:41:42.572
from our perspective on the ground and
I think probably the information that

00:41:42.582 --> 00:41:45.732
you're receiving from small businesses
is perhaps positioned in that sense.

00:41:45.742 --> 00:41:46.002
Howard Bentham: Yeah.

00:41:46.342 --> 00:41:48.212
You hear the word streamline, Guy.

00:41:48.232 --> 00:41:52.242
How can AI help streamline
business processes?

00:41:52.767 --> 00:41:57.720
Guy Gadney: So the streamlining question
often attracts words like productivity

00:41:57.720 --> 00:42:04.165
and efficiency and those sorts of I
think that as long as there are barriers

00:42:04.835 --> 00:42:08.785
to your business, whatever barriers you
identify, there is probably a way that

00:42:08.785 --> 00:42:14.895
AI can knock down those barriers and
if that is in marketing communications,

00:42:14.895 --> 00:42:18.865
you know, you can't reach a particular
audience or you're not getting traction

00:42:18.865 --> 00:42:23.230
or not getting customer take up in
certain areas, then what we find

00:42:23.230 --> 00:42:28.450
certainly is that you can recraft the
tone of a marketing piece very, very

00:42:28.450 --> 00:42:33.010
quickly with AI to match communications
into a very specific demographic.

00:42:33.010 --> 00:42:36.777
So let me give you an example, with
OxLEP support, we produced this

00:42:36.777 --> 00:42:41.077
wonderful game, for The Kraken Wakes
and as we are marketing that, we are

00:42:41.077 --> 00:42:44.557
marketing it, not only as a game in
its own right, but we have videos

00:42:44.647 --> 00:42:46.087
of the game that go up on YouTube.

00:42:46.487 --> 00:42:49.827
So, to do that, when you upload a
piece of video to YouTube, you have a

00:42:49.827 --> 00:42:51.627
bit of text that describes the video.

00:42:51.937 --> 00:42:55.247
Now, normally what I would write
is something like, this is a

00:42:55.257 --> 00:42:58.317
wonderful adaptation of John
Wyndham's novel, The Kraken Wakes.

00:42:58.347 --> 00:42:59.937
Okay, so that's me saying that.

00:43:00.137 --> 00:43:04.457
However, the YouTube algorithm,
reads it in a different way.

00:43:04.777 --> 00:43:09.057
It reads it in a way that is either
going to be a way that puts it up

00:43:09.097 --> 00:43:10.797
the rankings or knocks it down.

00:43:11.087 --> 00:43:14.703
So now what I do is I go over
to ChatGPT and I say, here's

00:43:14.703 --> 00:43:15.963
my description of this game.

00:43:16.163 --> 00:43:18.553
Now rewrite that in such a
way that it's going to be more

00:43:18.553 --> 00:43:20.343
attractive to YouTube's algorithm.

00:43:20.623 --> 00:43:24.403
It then gives me another piece of text
that I put into the comments field and

00:43:24.403 --> 00:43:29.367
that increases the viewers tangibly, we
noticed an almost immediate difference,

00:43:29.507 --> 00:43:31.967
when we recraft that in a particular way.

00:43:32.327 --> 00:43:35.930
Now what's slightly scary is that we're
now getting the machine to write text

00:43:35.940 --> 00:43:37.610
so that another machine can read it.

00:43:38.020 --> 00:43:43.170
But acknowledging, as in any good piece
of marketing, who our audience is and

00:43:43.170 --> 00:43:46.930
that our audience actually is the YouTube
algorithm to increase the visibility of

00:43:46.930 --> 00:43:49.020
that, was a key part of that campaign.

00:43:49.330 --> 00:43:50.760
So that to me is streamlining.

00:43:50.770 --> 00:43:55.823
It's the ability to reduce the barriers to
reach certain markets, be more flexible,

00:43:55.833 --> 00:44:01.153
more agile and that an individual
marketeer within our team can now do a

00:44:01.153 --> 00:44:03.193
lot more and reach a broader audience.

00:44:03.608 --> 00:44:08.218
You know, as a specific example, I
think there are many ways that processes

00:44:08.218 --> 00:44:12.698
can be made easier in this, but we're
at the beginning, you know, we're at

00:44:12.708 --> 00:44:15.988
the beginning of a long journey with
AI and I think part of the things

00:44:15.988 --> 00:44:20.298
that we're hearing today and part of
the problem actually and to viewers

00:44:20.298 --> 00:44:24.798
and listeners of this podcast is, you
know, the question is, but what is it?

00:44:25.488 --> 00:44:28.198
And the problem is that
it's like a tsunami, it's

00:44:28.218 --> 00:44:30.748
really broad and really deep.

00:44:30.778 --> 00:44:33.878
There's not one nice little wave that
you can go, Oh, look, there it is.

00:44:34.108 --> 00:44:38.978
It's not like a metaverse or
text messaging or websites,

00:44:39.198 --> 00:44:40.698
this is really broad.

00:44:40.928 --> 00:44:44.918
So trying to say, How does it
actually fit for my business?

00:44:45.368 --> 00:44:47.948
It's a challenging question to
answer, and I go back to it.

00:44:47.998 --> 00:44:51.518
I think Grae was saying similar, which is
first understand your business and where

00:44:51.518 --> 00:44:55.518
your barriers are to that business and
then look at what the solution may be.

00:44:55.748 --> 00:44:59.958
Chances are there's an AI for that,
but that's probably the first step.

00:45:00.258 --> 00:45:04.433
Howard Bentham: Grae, let me put
my PDoom, stroke cynic hat back on.

00:45:04.673 --> 00:45:08.303
Is streamline another word for
redundancy in this context?

00:45:08.303 --> 00:45:12.397
I mean, there is a genuine fear amongst
unions and workers that AI is going

00:45:12.397 --> 00:45:14.047
to make thousands of people redundant.

00:45:14.307 --> 00:45:15.557
Grae Laws: I think that's inevitable.

00:45:16.407 --> 00:45:21.267
I think the way that most modern
thinkers look at the change in

00:45:21.567 --> 00:45:25.997
employment as a direct consequence
of AI is basically it's three to one.

00:45:26.437 --> 00:45:28.027
Three jobs go, one job comes.

00:45:28.347 --> 00:45:32.197
But that's over a long period of
time and don't forget, we have the

00:45:32.207 --> 00:45:35.812
whole, you know, kind of global
population is actually now in decline.

00:45:36.232 --> 00:45:38.402
You know, we have birth
gaps in almost every nation.

00:45:38.402 --> 00:45:41.822
So in actual fact, in the future,
there will be less people and in

00:45:41.822 --> 00:45:44.792
actual fact, you'll still have the same
infrastructure that needs to be supported.

00:45:44.792 --> 00:45:46.832
So how do you keep the systems running?

00:45:47.222 --> 00:45:50.592
How do you do all those mundane
tasks that nobody wants to do

00:45:50.592 --> 00:45:53.232
because there aren't enough people
to actually take those roles?

00:45:53.662 --> 00:45:59.122
Now, because we, this is something
that plays out over decades, it was

00:45:59.132 --> 00:46:03.542
hoped by some of the great leaders who
were driving forward with Artificial

00:46:03.542 --> 00:46:08.632
Intelligence, machine learning, and
also embodied artificial intelligence.

00:46:08.632 --> 00:46:12.712
I put it into robotics was that it
will be a gradual transfer, whereas

00:46:12.942 --> 00:46:17.382
there were less people to do the jobs,
suddenly our robots who were, you know,

00:46:17.392 --> 00:46:21.262
had their artificial intelligence built
in would replace them and it would be

00:46:21.262 --> 00:46:24.582
a nice, smooth transition and actually,
that's not what really what's happening.

00:46:24.582 --> 00:46:28.082
Actually, it's an exponential
curve in terms of the growth of AI

00:46:28.122 --> 00:46:32.852
and also the implementation of AI
within robotics, so it is a threat.

00:46:33.082 --> 00:46:33.592
PDoom.

00:46:33.602 --> 00:46:35.012
Yes, I get that.

00:46:35.042 --> 00:46:38.882
But, you know, I always encourage
people to embrace things and

00:46:38.882 --> 00:46:40.112
have a lived experience of it.

00:46:40.592 --> 00:46:44.902
You know, when I was back in the 1990s,
I was a printer running printing presses

00:46:44.902 --> 00:46:48.312
and I remember a Mac operator coming
through to me saying, Grae, when you

00:46:48.312 --> 00:46:51.142
finish that print run, come through
and I'm going to show you something.

00:46:51.452 --> 00:46:55.192
I stopped the press, went through, pointed
at his screen and said, what's that?

00:46:55.192 --> 00:46:56.952
And I said, it looks
like a bit of artwork.

00:46:57.282 --> 00:46:57.582
Because.

00:46:57.962 --> 00:46:58.452
It's not.

00:46:58.792 --> 00:47:03.002
It's called a PDF, and it's been sent
to me down a telephone line on something

00:47:03.002 --> 00:47:06.992
called email, and I looked at him like
he'd just used words I'd never ever heard

00:47:07.022 --> 00:47:10.722
of, but when he actually explained the
significance of that and how it would

00:47:10.722 --> 00:47:14.862
change the way I did my job, it gave
me food for thought, and I thought,

00:47:14.882 --> 00:47:21.022
actually, if you were to suspend disbelief
that this PDF technology that means I

00:47:21.032 --> 00:47:26.052
don't need to print things anymore will
happen and that everybody will have

00:47:26.102 --> 00:47:30.932
emails and can send them things between
them, then hey, I might be out of a job.

00:47:31.362 --> 00:47:33.652
So when I went home to my
wife, first thing I said was, I

00:47:33.662 --> 00:47:35.092
think I need to buy a computer.

00:47:36.112 --> 00:47:40.002
I think I need to understand what this
is all about and I would say, you know,

00:47:40.002 --> 00:47:44.652
that is just an example, I would encourage
everybody today to be thinking about

00:47:44.652 --> 00:47:46.632
how could I learn something about AI?

00:47:46.932 --> 00:47:50.772
How can I use some of those
simple tools to augment what I do?

00:47:51.072 --> 00:47:54.472
How can I demonstrate through
the application of artificial

00:47:54.512 --> 00:47:58.872
intelligence in my day to day that
I can create additional value for

00:47:58.872 --> 00:48:00.472
the organisation that I work for?

00:48:01.052 --> 00:48:06.032
And embrace some of those things that
will remove those boring, repetitive tasks

00:48:06.042 --> 00:48:11.032
that you don't like doing, I don't like
doing, but AI loves and in doing so, allow

00:48:11.032 --> 00:48:15.788
you to maybe get a better sense of self
worth, improve the value that you create

00:48:15.788 --> 00:48:19.858
for the organisation and give you that
longevity in your career that you deserve.

00:48:19.953 --> 00:48:22.633
Howard Bentham: Rob, is there a risk
that as AI becomes more prevalent,

00:48:22.643 --> 00:48:26.883
current working practices will change
or as we're discussing, disappear?

00:48:27.423 --> 00:48:30.353
Rob Panting: I guess I can come a
tip from my own personal perspective.

00:48:30.353 --> 00:48:34.133
So from a communications and
marketing point of view, like

00:48:34.133 --> 00:48:38.883
Grae has just reiterated, I've
had to embrace AI within my job.

00:48:39.238 --> 00:48:43.508
Guy made a great example earlier
on in terms of the use of YouTube

00:48:43.808 --> 00:48:47.278
and how the algorithms that have
been built into YouTube can support

00:48:47.298 --> 00:48:52.248
your word to reach a much wider
audience than perhaps a typically

00:48:52.348 --> 00:48:54.288
organic approach to these things.

00:48:54.542 --> 00:48:57.822
Again, speaking to businesses in
Oxfordshire, there's an attitude

00:48:57.822 --> 00:49:02.382
that people look to embrace AI and
perhaps approach it from a positive

00:49:02.382 --> 00:49:06.862
perspective and not be fearful as to
what it might do to you personally.

00:49:06.862 --> 00:49:09.142
I think if you approach it from
that perspective, come at it from

00:49:09.392 --> 00:49:12.362
a position of strength, come at it
from a position of knowledge, which

00:49:12.602 --> 00:49:16.752
Grae has just emphasised then I think
you're in a much better position to

00:49:16.752 --> 00:49:19.012
understand the potential benefits of AI.

00:49:19.052 --> 00:49:24.602
I think there's no doubting that it will
change our working culture and the way

00:49:24.602 --> 00:49:29.092
in which businesses operate, perhaps do
things differently, but you're better

00:49:29.092 --> 00:49:33.152
approaching it from a position of being
informed and understanding rather than

00:49:33.152 --> 00:49:37.762
a pessimistic position where actually if
you bring that negative approach to it,

00:49:38.062 --> 00:49:42.172
you're probably not going to see the best
of AI and certainly from the cohort of

00:49:42.172 --> 00:49:45.492
businesses that we speak to on a day to
day basis, I think the majority of them

00:49:45.492 --> 00:49:47.452
are approaching it from that position.

00:49:47.792 --> 00:49:51.582
I think the danger with small
businesses in particular, so those

00:49:51.582 --> 00:49:55.647
with staff numbers sort of 10 and
below is having the thinking space.

00:49:55.647 --> 00:49:59.587
Greg reiterated that right at
the start in terms of how small

00:49:59.587 --> 00:50:02.817
business owners find the headspace
to think about AI and understand it.

00:50:02.827 --> 00:50:07.489
So I think coming at it from a position
of positivity is always a good thing.

00:50:07.859 --> 00:50:12.109
Howard Bentham: Guy, Grae and Rob, thank
you for the moment, fascinating stuff.

00:50:12.159 --> 00:50:13.579
We'll chat again shortly.

00:50:13.949 --> 00:50:17.349
You're listening to OxTalks the
podcast powered by the Oxfordshire

00:50:17.379 --> 00:50:18.859
Local Enterprise Partnership.

00:50:19.259 --> 00:50:22.969
Please get in touch with the team at OxLEP
to comment on what you've been hearing.

00:50:23.189 --> 00:50:27.669
Find us on social media, we're
on X at Oxfordshire LEP or via

00:50:27.669 --> 00:50:31.029
LinkedIn, search for Oxfordshire
Local Enterprise Partnership.

00:50:31.489 --> 00:50:35.969
Perhaps you run a company or organisation
that's looking for some specific help

00:50:36.019 --> 00:50:39.509
or simply need a steer to the most
appropriate business advice available.

00:50:40.074 --> 00:50:42.634
Why not try the OxLEP
Business Support Tool?

00:50:42.634 --> 00:50:47.419
OxLEP Business Support Tool:
OxLEP's Business Support Tool

00:50:47.459 --> 00:50:48.939
is here to help your company.

00:50:49.509 --> 00:50:52.449
Whether you're just starting
out, growing, or ready to take

00:50:52.449 --> 00:50:53.869
on a new business challenge.

00:50:54.169 --> 00:50:58.159
If you're looking for the latest advice
and support, complete our business support

00:50:58.159 --> 00:51:02.589
tool today and get set to receive a
bespoke action plan for your organisation.

00:51:02.999 --> 00:51:05.109
Head to OxLEPbusiness.

00:51:05.369 --> 00:51:05.389
co.

00:51:05.499 --> 00:51:06.679
uk to find out more.

00:51:10.144 --> 00:51:12.364
Howard Bentham: Let's speak more
with Grae Laws, Managing Director

00:51:12.374 --> 00:51:16.684
of Beyond Touch, and Guy Gadney
from Charisma AI based in Oxford.

00:51:16.894 --> 00:51:21.354
Guy, I want to delve into the
ethics of AI and the safeguards

00:51:21.354 --> 00:51:24.004
that exist, or perhaps should exist.

00:51:24.444 --> 00:51:27.574
Are there ethical issues
around AI in your thoughts?

00:51:27.999 --> 00:51:31.049
Guy Gadney: Yes, I think there are ethical
issues and I think if you go back to,

00:51:31.059 --> 00:51:33.569
if you remember that scene in Jurassic
Park, in the first one, where they're

00:51:33.569 --> 00:51:38.279
all sitting around, having dinner and
Jeff Goldblum, I think it is, he says,

00:51:38.289 --> 00:51:42.099
you know, you scientists, just because
you could, didn't mean that you should

00:51:42.549 --> 00:51:46.999
and I think there's a real parallel and I
sort of hold that line in my head a lot.

00:51:47.039 --> 00:51:52.539
There is so much that can be done with
AI and we do need to have a little bit

00:51:52.589 --> 00:51:55.186
of restraint to say, should we do this?

00:51:55.586 --> 00:52:00.156
Now, that's all nice and easy to
say, but then from a practical

00:52:00.156 --> 00:52:01.526
perspective, what does that mean?

00:52:01.536 --> 00:52:04.376
Well, I think there are
two examples from my side.

00:52:04.386 --> 00:52:09.856
One is when we started building
Charisma AI back in 2016, 2017.

00:52:09.936 --> 00:52:12.476
Actually, one of the first things
we did was we wrote a white

00:52:12.476 --> 00:52:14.536
paper about bias in AI data sets.

00:52:14.836 --> 00:52:17.526
Now, at that point, it was like,
why are you thinking about that?

00:52:17.751 --> 00:52:22.211
But we were thinking about it because
we knew that as we were bringing the

00:52:22.211 --> 00:52:26.151
character, these virtual characters to
life that you were going to talk to, they

00:52:26.151 --> 00:52:30.351
may have biases and we wanted to know
what those biases were, so we wrote the

00:52:30.351 --> 00:52:34.761
white paper for that and that, I think,
was the start of a bit of a journey

00:52:34.811 --> 00:52:39.671
around ethics, around, you know, where
you source your data from, you know, is

00:52:39.671 --> 00:52:45.441
it someone's copyright, is it angry data,
like on social media, on some social media

00:52:45.441 --> 00:52:48.541
platforms, is it, academic data, whatever.

00:52:48.771 --> 00:52:52.541
So I think we've always kept that and then
in fact, we're working on a big Innovate

00:52:52.541 --> 00:52:58.551
UK funded project at the moment, which
we're halfway through and in the original

00:52:58.651 --> 00:53:03.821
bid application, we costed in a full time
digital ethicist and what I love about

00:53:03.821 --> 00:53:07.861
what she's doing, she calls it consequence
scanning, which is that she's involved

00:53:07.861 --> 00:53:11.461
in all of our meetings, you know, exec
meetings, production meetings, just got

00:53:11.601 --> 00:53:17.156
full access to everything that we do and
she just scans for intended and unintended

00:53:17.156 --> 00:53:18.676
consequences of what we're doing.

00:53:19.176 --> 00:53:23.006
She's always said, that doesn't mean
that I'm going to block you, but it

00:53:23.006 --> 00:53:26.556
means I'm going to raise it and then
you can then do the risk analysis to

00:53:26.556 --> 00:53:30.126
think about whether or not you're doing
something which is going to impact

00:53:30.126 --> 00:53:32.826
someone's job or whatever it may be.

00:53:33.146 --> 00:53:39.616
So the ethics are, I think, something
that people should think about, and

00:53:39.616 --> 00:53:43.756
it doesn't mean you have to move
away from your strategy, it just

00:53:43.756 --> 00:53:45.006
means you need to think about it.

00:53:45.236 --> 00:53:48.206
And our view is that if you're thinking
about it, you've got that, you know,

00:53:48.616 --> 00:53:53.466
should we question running through
your mind, you're probably about 60

00:53:53.466 --> 00:53:55.536
percent of the way to doing good.

00:53:55.686 --> 00:53:57.556
Howard Bentham: Because it feels,
from what we see in the media,

00:53:57.556 --> 00:54:01.606
that society is, shall we say, a
growing number of steps behind the

00:54:01.606 --> 00:54:04.616
technology as the technology advances.

00:54:05.006 --> 00:54:09.236
That's going to mean that our
laws become outdated or obsolete

00:54:09.296 --> 00:54:12.746
and that we, as you've used the
words, unintended consequences.

00:54:12.746 --> 00:54:17.256
We're sleepwalking into a world
of those unintended consequences.

00:54:17.376 --> 00:54:18.956
How can we stop that happening?

00:54:19.076 --> 00:54:21.716
Guy Gadney: Well, I think we're going,
regulation wise, we're going faster

00:54:21.766 --> 00:54:25.849
than a lot of governments have ever gone
before, you know, the AI acts that first

00:54:25.849 --> 00:54:29.299
came out of the EU, there's a draft in
the House of Lords, that's going out

00:54:29.319 --> 00:54:35.029
into the US as well, are faster than
how Europe and the rest of the world

00:54:35.029 --> 00:54:40.789
responded to the internet and you know,
the GDPR legislation around personal

00:54:40.789 --> 00:54:46.044
information has had a massive, and I
think very positive, impact on society

00:54:46.044 --> 00:54:48.874
as people have started to realise
the value of their own information.

00:54:49.454 --> 00:54:54.564
So I think that they have started to
realise that this needs significant

00:54:55.304 --> 00:54:59.914
regulation around it, not least
because of society and indeed, you

00:54:59.914 --> 00:55:04.384
know, military as a defense mechanism
and attack mechanism as well.

00:55:04.854 --> 00:55:09.587
So, I think they're going faster than
they have done still, it's slower than

00:55:09.587 --> 00:55:13.847
the technology industries and there
is a tension between innovation and

00:55:13.847 --> 00:55:17.497
regulation that I hear the whole time and
the innovators don't want to be slowed

00:55:17.507 --> 00:55:20.277
down because they feel that they're
going to lose a competitive commercial

00:55:20.277 --> 00:55:22.097
advantage because of regulation.

00:55:22.447 --> 00:55:26.857
At the same time, we've also seen some
pretty tragic examples where innovation

00:55:26.857 --> 00:55:30.697
has been, you know, the number one thing,
that people have gone down that, that's

00:55:30.697 --> 00:55:34.947
the pathway and conforming to regulation
has not happened, that tragic case of

00:55:34.947 --> 00:55:37.037
the submarine going down to the Titanic.

00:55:37.317 --> 00:55:40.627
Well, that was a conflict between
innovation and safety and regulation.

00:55:40.847 --> 00:55:44.377
So there is an absolute, very clear
reason why these things need to

00:55:44.387 --> 00:55:48.282
be in balance and my sense is that
people are very focused on this.

00:55:48.522 --> 00:55:51.532
You know, the first all party
parliamentary group in the UK on

00:55:51.542 --> 00:55:54.402
AI, I think, started in 2016, 2017.

00:55:54.442 --> 00:55:59.362
I was a part of it and it was sidelined
by Brexit because the parliamentarians

00:55:59.792 --> 00:56:01.472
were focusing on having to deal with that.

00:56:01.473 --> 00:56:03.992
There was a lot of work they needed
to do and they acknowledged that

00:56:03.992 --> 00:56:05.292
was going to set us back on AI.

00:56:05.692 --> 00:56:07.097
But it's coming back into play again.

00:56:07.107 --> 00:56:12.287
So I think it's naturally slower because
the regulators aren't sci fi writers,

00:56:12.287 --> 00:56:15.437
they don't look too much into the future,
they look into what's happening right now.

00:56:15.457 --> 00:56:19.307
But my sense is that it's
being given responsibility

00:56:19.367 --> 00:56:20.977
and an airtime that it needs.

00:56:21.303 --> 00:56:24.363
Howard Bentham: Great, Guy raised a
really interesting point there about bias.

00:56:24.513 --> 00:56:27.553
AI will have been input and
trained because we've talked about

00:56:27.553 --> 00:56:29.033
the training of it by someone.

00:56:29.613 --> 00:56:34.313
Does it leave an inherent
bias in the system somewhere?

00:56:34.556 --> 00:56:35.506
How do you eliminate it?

00:56:35.801 --> 00:56:38.761
Grae Laws: With great difficulty,
I would say, it's almost impossible

00:56:38.761 --> 00:56:39.871
to remove it once it's there.

00:56:39.881 --> 00:56:44.628
You ask Sam Altman or any of the
leaders of those organisations who

00:56:44.648 --> 00:56:49.167
are developing large language models,
exactly how does it do the sorting and

00:56:49.167 --> 00:56:52.367
processing and get to a point where
that foundation model can be used?

00:56:52.367 --> 00:56:56.588
They can't really answer it and they can't
easily go in and reverse engineer it.

00:56:56.588 --> 00:57:02.118
Now what they are trying to do is do
synthetic data, which actually has already

00:57:02.118 --> 00:57:09.178
been pre sanitised to remove bias and
think of it that it's one large language

00:57:09.198 --> 00:57:13.898
model using algorithms to train another
large language model and in doing so iron

00:57:13.898 --> 00:57:18.607
out some of those creases and you've got
like three different classes in a way.

00:57:19.077 --> 00:57:23.858
You have the closed language models
that the likes of OpenAI have

00:57:23.858 --> 00:57:26.838
and Grok, Elon Musk's X model.

00:57:27.118 --> 00:57:34.637
Then you have the Open models, such as
Meta's Lama 3, which is available for

00:57:34.637 --> 00:57:36.597
anybody to use as a language model.

00:57:37.097 --> 00:57:39.097
But of course, nobody really
knows what's inside those.

00:57:39.097 --> 00:57:42.717
They won't admit to what's in there
other than just to have to accept

00:57:42.727 --> 00:57:44.097
that, yes we scraped the internet.

00:57:44.377 --> 00:57:47.887
Yes, there is copyright issues, probably,
you know, yeah, we have to accept that.

00:57:48.387 --> 00:57:52.587
But actually that fine tuning is actually
quite often done by humans who have in

00:57:52.597 --> 00:57:58.112
their own innate biases, which creates
this third requirement in a way, which

00:57:58.112 --> 00:58:02.462
is, and this is a national infrastructure
level, and that is for governments to

00:58:02.462 --> 00:58:07.052
own their own language models and that's
where we're going to is actually safe,

00:58:07.322 --> 00:58:14.292
reliable, sanitised language models that
can be used freely or at minimal cost by

00:58:14.332 --> 00:58:19.372
businesses to build applications that will
be used in the real world by individuals,

00:58:19.842 --> 00:58:24.852
by businesses and most importantly, by
people who work in government themselves.

00:58:24.867 --> 00:58:28.557
Howard Bentham: There's the morality side
of this, cheating in exams, passing off AI

00:58:28.747 --> 00:58:30.617
generated work as your own, stealing IP.

00:58:31.557 --> 00:58:37.167
No one voted for AI,
it's just imposed on us.

00:58:37.667 --> 00:58:40.527
Is there a real ethical issue around that?

00:58:40.812 --> 00:58:43.792
Grae Laws: Impose, interesting
word, I guess it's like gas.

00:58:43.792 --> 00:58:46.542
It's all around us and you can't
move it out of the way really.

00:58:46.552 --> 00:58:50.542
Being it's been here more than a
decade and the more we use it, it's

00:58:50.872 --> 00:58:54.602
positive affirmation for any developer
to carry on doing much of the same.

00:58:54.992 --> 00:58:59.862
So we ourselves are feeding the machine
that creates the thing that we, some of us

00:58:59.862 --> 00:59:03.852
or some people may not want from an ethics
point of view, I mean, just maybe you

00:59:03.852 --> 00:59:08.232
can scale it out in a slightly different
way and ask people who have environmental

00:59:08.232 --> 00:59:12.472
concerns about industrialisation and
say, so what do you think about the

00:59:12.472 --> 00:59:14.572
energy use of large language models?

00:59:15.017 --> 00:59:18.747
You know, does it bother you that
every single day, ChatGPT alone

00:59:18.757 --> 00:59:23.587
requires 17, 000 homes worth of
energy just to keep it ticking over?

00:59:24.087 --> 00:59:28.817
Every time you put a prompt into a
ChatGPT or Perplexity, it's using

00:59:28.817 --> 00:59:34.142
one whole bottle of drinkable
water per prompt, think about that.

00:59:34.342 --> 00:59:38.972
Think about the sheer scale of
the technology, the compute that's

00:59:38.972 --> 00:59:43.652
required, all of those processors, all
of the electronic waste, all of the

00:59:43.732 --> 00:59:48.582
thermal pollution that's been created,
all of these things are byproducts

00:59:48.582 --> 00:59:50.332
and they're still ethical concerns.

00:59:50.612 --> 00:59:54.492
Now, are you willing to put them all
to one side, because you believe that

00:59:54.562 --> 01:00:00.462
overall the net benefit to humanity
through developing even greater language

01:00:00.492 --> 01:00:05.082
models than ones we already have can
put us at ease in thinking actually,

01:00:05.082 --> 01:00:08.239
you know, we are using a whole lot of
energy, you know, the US estimate that

01:00:08.249 --> 01:00:12.669
next year, 6 percent of their total
energy will be used by artificial

01:00:12.669 --> 01:00:14.529
intelligence and it will keep growing.

01:00:15.079 --> 01:00:19.869
Just this morning, it was announced by
OpenAI that ChatGPT 5 is coming out.

01:00:20.389 --> 01:00:26.539
Okay, so this has 100x the compute
of the current live chat GPT,

01:00:26.569 --> 01:00:29.669
100 times the processing power.

01:00:30.039 --> 01:00:33.209
Now, in terms of an order of magnitude,
our experience means it will feel

01:00:33.209 --> 01:00:36.489
like it's twice as good because it's
not a direct correlation between

01:00:36.489 --> 01:00:40.989
compute and the order of magnitude,
but we experienced, but this isn't

01:00:40.989 --> 01:00:42.499
slowing down, none of this is.

01:00:42.499 --> 01:00:46.729
We will keep consuming energy, we'll
keep consuming water, there will be

01:00:46.729 --> 01:00:50.374
a greater demand on us to be able
to do it and is that what you want?

01:00:50.524 --> 01:00:51.504
Because that's what's going to happen.

01:00:51.534 --> 01:00:54.884
Howard Bentham: Yeah, your idea of
a tsunami sounds absolutely right.

01:00:55.484 --> 01:00:59.014
Guy, just bringing it back to our
audience here, how can business leaders

01:00:59.014 --> 01:01:04.394
support their employees to learn about
AI and try and get behind this, encourage

01:01:04.394 --> 01:01:06.064
them to use it to support their work?

01:01:06.654 --> 01:01:09.634
Guy Gadney: Yeah, look, and I think
there's a connection here with your

01:01:09.634 --> 01:01:12.804
last question about no one voted
for this and the reason no one

01:01:12.804 --> 01:01:16.164
voted for it is because it's not
a government initiated technology.

01:01:16.174 --> 01:01:19.774
It was initiated by market forces,
you know, it was initiated by

01:01:19.774 --> 01:01:22.714
industry and it's had a huge
amount of tech up by people.

01:01:22.984 --> 01:01:27.064
So in the sense of how someone
might vote for whether they want

01:01:27.064 --> 01:01:30.194
something or not, buying it is
the market's way of doing that.

01:01:30.524 --> 01:01:35.474
So I think if you're looking at a business
and you're looking at your own staff.

01:01:35.514 --> 01:01:39.334
The first question is to ask
them, are they already using it?

01:01:39.364 --> 01:01:40.584
Are they using it at home?

01:01:40.934 --> 01:01:42.854
Are they using it in their day to day?

01:01:43.504 --> 01:01:45.564
Have it is in a safe space question.

01:01:45.564 --> 01:01:50.424
So if it's a hypothetical communications
manager called Rob, who happens to be

01:01:50.424 --> 01:01:54.934
working for an enterprise partnership
and his boss says, Rob, are you using

01:01:54.934 --> 01:01:56.784
ChatGPT to write press releases?

01:01:56.869 --> 01:02:01.399
It's okay for Rob to say yes, because
sort of everyone else is doing it.

01:02:01.429 --> 01:02:05.766
The education was fascinating, so from
the work we've been doing with Oxford

01:02:05.766 --> 01:02:08.846
University and various schools, we
know that within about two weeks of

01:02:08.846 --> 01:02:12.456
ChatGPT launching, pretty much every
teacher and every student in the UK

01:02:12.456 --> 01:02:14.516
was using it, but no one would say it.

01:02:14.856 --> 01:02:18.916
Now it's starting to evolve into
curriculum in really lovely ways,

01:02:18.936 --> 01:02:23.159
which is here is a piece of prose text
that was published and here's another

01:02:23.159 --> 01:02:25.096
one which was written by ChatGPT.

01:02:25.276 --> 01:02:28.246
Compare and contrast why these
two pieces of text are different.

01:02:28.786 --> 01:02:32.666
So that's getting the student
to think about it, acknowledging

01:02:32.666 --> 01:02:33.656
the existence of it.

01:02:33.936 --> 01:02:37.036
So I think the first step is acknowledge
that a lot of people are using AI,

01:02:37.176 --> 01:02:41.436
ask your staff, think about your own
strategy with OxLEP's support in that

01:02:41.816 --> 01:02:43.566
space of course, and do a quick audit.

01:02:44.206 --> 01:02:45.806
You know, where are you already using it?

01:02:45.836 --> 01:02:46.926
Where have people got ideas?

01:02:46.926 --> 01:02:50.176
Where have the employees got ideas of,
gosh, wouldn't it be cool if we had a

01:02:50.176 --> 01:02:52.636
corporate, you know, perplexity account?

01:02:53.026 --> 01:02:56.796
That would make my life easier and
discuss it, it's an open conversation.

01:02:56.861 --> 01:02:59.001
Howard Bentham: As OxTalks has
become the confessional now, Rob?

01:02:59.936 --> 01:03:03.486
Rob Panting: Well, yes,
I mean, I think, yes.

01:03:03.706 --> 01:03:06.016
Howard Bentham: Is that a yes?

01:03:06.296 --> 01:03:08.426
Rob Panting: I would be lying if
I said that I didn't embrace AI.

01:03:08.756 --> 01:03:11.696
I mean, yeah, picking up on Guy's
point there just briefly, I, I think

01:03:12.076 --> 01:03:17.976
from marketing communications, public
relations, journalism, I think it would

01:03:17.986 --> 01:03:22.026
be fair to say that probably we are the
group of professionals that are maybe

01:03:22.026 --> 01:03:24.046
most skeptical in terms of the use of AI,

01:03:26.641 --> 01:03:29.081
the potential for it
to take over our jobs.

01:03:29.081 --> 01:03:34.671
I think we see ourselves as
the creative guy that works

01:03:34.671 --> 01:03:36.161
within the creative industries.

01:03:36.161 --> 01:03:43.811
I think taking Guy's business approach in
terms of embracing AI is the way to go and

01:03:43.811 --> 01:03:48.087
I think similar with us you know, there's
a number of occasions where I've gone into

01:03:48.087 --> 01:03:51.987
work and perhaps I've been struggling a
little bit to come up with the right words

01:03:51.987 --> 01:03:56.847
to try and portray a particular programme
that we're launching or a particular fund

01:03:56.847 --> 01:04:00.917
that we're putting out and actually using
the AI to help me to get off the page,

01:04:00.917 --> 01:04:04.732
you know, get my ideas off the page and
to push me in the right direction as to

01:04:04.732 --> 01:04:09.112
what's going to work best to allow me to
communicate with our target markets has

01:04:09.112 --> 01:04:13.552
been massively helpful and like Grae says,
the majority of this stuff is there for

01:04:13.552 --> 01:04:15.672
you to use, it's free to a certain point.

01:04:15.992 --> 01:04:19.602
You would be silly not to embrace
it into your everyday operations,

01:04:19.642 --> 01:04:22.482
particularly I think from a
comms and marketing perspective.

01:04:22.522 --> 01:04:27.312
Certainly if we're an example of that,
you know, we're using it to a relatively

01:04:27.312 --> 01:04:31.206
small level, I would say, but undoubtedly
it's helped us to reach new audiences

01:04:31.206 --> 01:04:35.936
and probably in turn helped people like
Grae to connect with businesses who need

01:04:35.936 --> 01:04:38.126
the support, it really starts from there.

01:04:38.446 --> 01:04:42.366
Guy Gadney: The one caveat, Rob, I'd
say on that is that the way OpenAI

01:04:42.406 --> 01:04:47.456
works is that it uses what you type
into it to train its own model.

01:04:47.786 --> 01:04:52.696
So in essence, anything that you're
typing into ChatGPT is going to be public

01:04:53.086 --> 01:04:56.906
and there was an example recently of a
chief financial officer who was looking

01:04:56.916 --> 01:05:01.116
to summarise what the financial state of
their business was and so they typed in

01:05:01.116 --> 01:05:04.746
all of the financials for the company and
said, please summarise this, which then

01:05:04.746 --> 01:05:10.306
meant that all of the financial states
of that company were then searchable in

01:05:10.306 --> 01:05:16.112
OpenAI and our view on this is that it
is important for businesses ultimately

01:05:16.152 --> 01:05:20.672
to have their own sovereign model,
we'll get to that in the market at some

01:05:20.672 --> 01:05:22.262
point in the coming months and years.

01:05:22.722 --> 01:05:29.077
But just be aware that any personal
or confidential information that goes

01:05:29.077 --> 01:05:33.667
into ChatGPT will then form part of
its model and will likely be public.

01:05:33.842 --> 01:05:37.022
Grae Laws: I might add to that and just
say perplexity doesn't do that, Claude

01:05:37.252 --> 01:05:43.572
by Anthropic, doesn't do that and if
you have embedded Microsoft 365 Copilot

01:05:43.602 --> 01:05:49.202
into your enterprise, then it has a very
clear distinction between I can search

01:05:49.202 --> 01:05:53.632
the world wide web, flick the switch
across, I can search and use your internal

01:05:53.632 --> 01:05:55.082
data and it works exactly the same way.

01:05:55.082 --> 01:05:59.192
So if you were screenshotting your
annual report or your cash flow

01:05:59.192 --> 01:06:01.522
forecast and say, give me some
insight, where am I going wrong?

01:06:01.962 --> 01:06:06.002
That isn't going to inform the language
models that those organisations use.

01:06:06.042 --> 01:06:10.292
But it's a point well made, the free
versions are almost like you're the

01:06:10.292 --> 01:06:16.142
product and your data is helping
that tool understand the relationship

01:06:16.152 --> 01:06:21.062
between numbers and the written word
and your prompt and what that means.

01:06:21.072 --> 01:06:22.402
Why did that human ask that?

01:06:22.403 --> 01:06:24.762
That, you know, kind of
natural language processing.

01:06:25.257 --> 01:06:27.047
Howard Bentham: Let's finish
with a bit of a checklist on what

01:06:27.047 --> 01:06:30.937
business leaders should do if
they're seriously thinking about

01:06:30.967 --> 01:06:32.947
adopting AI into their organisations.

01:06:33.387 --> 01:06:35.657
Short and sweet, Grae,
give us your thoughts.

01:06:35.707 --> 01:06:40.147
Grae Laws: Contact OxLEP, number
one, do the diagnostic tool,

01:06:40.197 --> 01:06:41.987
and then you need to experiment.

01:06:41.997 --> 01:06:46.847
Start with safe tools, I can't
reiterate enough the importance of go

01:06:46.847 --> 01:06:51.517
to YouTube, it's a single best place
where you can search for content.

01:06:51.517 --> 01:06:54.952
It's the second biggest search engine
in the world and it actually has

01:06:54.952 --> 01:06:58.672
less bias than the number one search
engine tool in the world, Google.

01:06:59.112 --> 01:07:03.762
So go to YouTube, find some of those big
organisations and individuals who actually

01:07:03.762 --> 01:07:10.319
do regular, kind of weekly or bi weekly
shows introducing some new technologies.

01:07:10.639 --> 01:07:14.809
Talk to your peers, as a business
leader about what tools they use

01:07:15.139 --> 01:07:18.609
and something that Guy said earlier,
which is talk to your employees,

01:07:19.229 --> 01:07:20.729
find out what they're using.

01:07:20.729 --> 01:07:24.769
It's that person who is sitting in
the cafeteria on their mobile phone

01:07:24.769 --> 01:07:28.449
is using the app that your, you know,
head of operations should be doing.

01:07:29.104 --> 01:07:33.224
Understand from best practice, join
peer groups in your area, listen to

01:07:33.224 --> 01:07:37.324
what other people are using and just
try these things, try before you buy,

01:07:37.494 --> 01:07:39.164
you could even go to something like poe.

01:07:40.094 --> 01:07:43.634
com, try all the language models
for free, see which one you like.

01:07:44.074 --> 01:07:45.944
Go to something like huggingface.

01:07:45.964 --> 01:07:49.764
com, if you're really into this,
you can find the language models.

01:07:49.764 --> 01:07:53.494
You can find, you know, lots of user
examples of these types of things.

01:07:53.494 --> 01:07:56.964
Find out what works best for
you and then give it a go.

01:07:57.644 --> 01:08:02.084
Guy Gadney: I would start by saying
that probably businesses are already

01:08:02.084 --> 01:08:08.494
using AI and so the recommendation is to
understand how you're already doing it.

01:08:08.664 --> 01:08:12.424
Even though you may not know, you
probably are and build it from there.

01:08:12.764 --> 01:08:16.864
I think as we head into the next
few years, you know, there are

01:08:16.864 --> 01:08:21.814
challenges for businesses around
inflation, macroeconomic challenges.

01:08:22.024 --> 01:08:25.904
So there are going to be ways in
which businesses need to respond

01:08:26.134 --> 01:08:30.364
and AI can be a tool to help
you respond to those situations.

01:08:30.644 --> 01:08:35.249
So you're already doing it, do a quick
audit, as you say, talk to OxLEP, talk

01:08:35.249 --> 01:08:40.609
to employees, that person who's in the
cafe on the app could actually be using

01:08:40.619 --> 01:08:41.999
the app that could save your business.

01:08:42.449 --> 01:08:46.399
So, you know, find out what it is,
you know, not that your business may

01:08:46.399 --> 01:08:49.729
be a problem, but it could be the one
that elevates it suddenly as well.

01:08:50.029 --> 01:08:54.089
So I think people know more about
AI than they think they do already,

01:08:54.089 --> 01:08:57.449
actually and a quick, I'd say quick
audit, work out where you want to go.

01:08:57.769 --> 01:09:01.249
And then as with saying the barrier
to entry is low, so give it a go.

01:09:01.359 --> 01:09:03.049
Howard Bentham: Both of you,
again, briefly, if you would

01:09:03.059 --> 01:09:05.239
gaze into the AI crystal ball.

01:09:05.269 --> 01:09:08.289
What's this space going to look
like in five years, ten years time?

01:09:08.499 --> 01:09:10.659
Guy Gadney: Gosh, when you said
five years and then went 10 years, I

01:09:10.659 --> 01:09:11.739
thought you were gonna bring it down.

01:09:11.739 --> 01:09:12.969
Let's go, five years, two years.

01:09:13.299 --> 01:09:16.949
There are a couple of macro trends that
tend that I see, and I see these coming.

01:09:16.959 --> 01:09:21.869
The first one is that in any
new technology advanced, it goes

01:09:21.989 --> 01:09:22.929
low bandwidth to high bandwidth.

01:09:22.989 --> 01:09:27.409
So text, static image, audio,
video, and I think we're seeing the

01:09:27.409 --> 01:09:29.059
same thing play out now with AI.

01:09:29.109 --> 01:09:33.449
We saw ChatGPT creating text, and
then it was audio, and then that.

01:09:33.519 --> 01:09:37.249
We will move into everyone having
access to text to video, you

01:09:37.249 --> 01:09:41.969
know, very soon and I think that
is then a qualitative standpoint.

01:09:41.989 --> 01:09:44.769
So we see the innovation and
then the quality increase.

01:09:45.049 --> 01:09:49.989
That is going to change an enormous
amount of people's work, people's lives.

01:09:50.479 --> 01:09:53.519
There will be issues around deepfakes
that we already know, that we'll

01:09:53.519 --> 01:09:54.559
have to solve along the way.

01:09:54.569 --> 01:10:00.589
So I think the, to my mind, the one I'm
watching right now is generative AI video,

01:10:01.179 --> 01:10:02.579
and I think that's going to be critical.

01:10:02.609 --> 01:10:05.059
I think the other thing that
we'll see is going to be the

01:10:05.059 --> 01:10:06.974
fragmentation of language models.

01:10:06.984 --> 01:10:10.624
So we've got the ones that we mentioned
around OpenAI, the big two or three

01:10:10.624 --> 01:10:15.204
around the world and these will then
fragment into each business has its

01:10:15.204 --> 01:10:20.064
own, it'll be each country has its
own sovereign LLM, then each business

01:10:20.064 --> 01:10:24.184
has its own LLM and then all of us as
individuals will have our own ones that

01:10:24.554 --> 01:10:30.129
are the AI version of your answering
machine message, you know, that responds

01:10:30.199 --> 01:10:33.699
on an individual basis and I think we
will see that, in the next little bit.

01:10:33.699 --> 01:10:34.779
So I think those are the dead certs.

01:10:34.799 --> 01:10:38.139
The other ones are, you know, are
the new and sort of advancements

01:10:38.139 --> 01:10:39.349
and innovations that we'll see.

01:10:39.624 --> 01:10:40.324
Howard Bentham: What about you, Grae?

01:10:40.984 --> 01:10:41.654
Grae Laws: What Guy said.

01:10:43.464 --> 01:10:44.774
I think it's going to be exciting times.

01:10:44.774 --> 01:10:46.724
There's lots more investment needed.

01:10:47.194 --> 01:10:48.814
Guardrails have got to be put in place.

01:10:48.924 --> 01:10:50.194
We've got to embrace this.

01:10:50.224 --> 01:10:54.094
Every single living human being that
would wish to do so ought to embrace

01:10:54.094 --> 01:10:57.734
the technology if they can, even in a
light touch sense, because if they don't

01:10:57.734 --> 01:11:01.839
embrace it, it will embrace them and
then maybe they're not in control and

01:11:01.839 --> 01:11:06.879
I think that what we'll see within 10
years is, you know, artificial general

01:11:06.879 --> 01:11:09.819
intelligence where we suddenly realise
that there are technologies out there

01:11:09.819 --> 01:11:15.999
that most definitely exceed our capability
at every level, in every discipline.

01:11:16.009 --> 01:11:20.139
That's a given and I just hope
that we don't move beyond that to

01:11:20.139 --> 01:11:21.539
anything else, I'm sure we won't.

01:11:21.804 --> 01:11:25.904
Howard Bentham: And the one thing
you want to see AI change that it's

01:11:25.984 --> 01:11:32.144
beyond its capability at the moment,
the one sort of magic bullet or one

01:11:32.484 --> 01:11:37.284
thing that it solves globally or just
for humanity, what would you say?

01:11:37.719 --> 01:11:37.989
Grae Laws: Poverty.

01:11:38.719 --> 01:11:39.789
Complete that one, mate.

01:11:40.679 --> 01:11:42.309
what I'd ask GPT.

01:11:42.759 --> 01:11:43.209
Guy Gadney: Yeah.

01:11:43.209 --> 01:11:44.519
I'd go climate change.

01:11:44.917 --> 01:11:48.787
I think a lot of the challenges that AI
researchers we've seen outta DeepMind

01:11:48.787 --> 01:11:53.407
and so forth have been around Space
Invaders or Go, or Chess or, and I

01:11:53.407 --> 01:11:57.277
don't like those 'cause there's zero
sum games win or lose against a human.

01:11:57.307 --> 01:12:00.792
So you are, I, you know, I
sort of stand away from that.

01:12:01.062 --> 01:12:05.412
Give it something challenging, which
is an open-ended very complex pattern,

01:12:06.252 --> 01:12:10.842
like climate change, poverty, things
that are for the betterment of human

01:12:10.842 --> 01:12:13.992
evolution, and you know, and give us
the answer and it'll probably come

01:12:13.992 --> 01:12:18.662
out saying it's 42, and that's not
going to help too many people, Douglas

01:12:18.662 --> 01:12:20.512
Adams fans, but let's give it a go.

01:12:20.637 --> 01:12:24.167
Howard Bentham: Huge thanks to Grae
Laws, Managing Director of Beyond

01:12:24.187 --> 01:12:29.367
Touch and Guy Gadney, CEO of Charisma
AI based in Oxford and a big thank

01:12:29.367 --> 01:12:31.507
you also to OxLEPs Rob Panting too.

01:12:32.197 --> 01:12:34.197
Thank you for listening to OxTalks.

01:12:34.387 --> 01:12:37.377
There are a growing number of
editions of OxTalks available from

01:12:37.377 --> 01:12:38.777
where you normally get your podcasts.

01:12:39.647 --> 01:12:42.827
Check out some of the previous
editions, featuring Andy Edwards

01:12:42.827 --> 01:12:47.227
from MakeSpace Oxford on creating
opportunities for all communities.

01:12:47.567 --> 01:12:51.547
Chas Bountra, the Pro Vice Chancellor
for Innovation at the University

01:12:51.547 --> 01:12:54.917
of Oxford on how we can retain
the best talent in Oxfordshire.

01:12:55.407 --> 01:13:00.167
An apposite to our discussion today, the
changing face of the workplace with the

01:13:00.207 --> 01:13:05.337
MD of Legacy, an award winning sustainable
events agency, Abena Fairweather.

01:13:05.917 --> 01:13:08.247
Every episode, is well worth a listen.

01:13:08.367 --> 01:13:11.627
Please spread the word, tell your
friends or colleagues about us, and if

01:13:11.627 --> 01:13:13.727
you feel so inclined, leave us a review.

01:13:14.167 --> 01:13:17.337
You can share your thoughts and
suggestions on our social channels,

01:13:17.587 --> 01:13:21.117
and you can email your questions for
inclusion in future editions too.

01:13:21.427 --> 01:13:23.747
The address is on the podcast description.

01:13:24.587 --> 01:13:28.717
Business support in Oxfordshire is
just an email or a phone call away.

01:13:28.977 --> 01:13:30.907
OxLEP is at your service.

01:13:31.237 --> 01:13:34.697
The OxLEP business support
tool can signpost you to expert

01:13:34.697 --> 01:13:36.017
help in a matter of minutes.

01:13:36.517 --> 01:13:37.417
Why not take a look?

01:13:37.667 --> 01:13:40.507
Find it on our website, oxfordshirelep.

01:13:40.557 --> 01:13:40.897
com.

01:13:41.357 --> 01:13:46.067
But for now, from the whole OxLEP team,
and from me, Howard Bentham, it's goodbye.