Who's Responsible When AI Decides? Navigating Ethics Without Paralysis

1
00:00:00,120 --> 00:00:02,700
Mark Richards: G'day and welcome
to another episode of SBCs
2
00:00:02,700 --> 00:00:07,620
unleashed, almost in the process
of renaming it. We kind of
3
00:00:07,620 --> 00:00:10,080
agreed last week that it was
going to become agents
4
00:00:10,080 --> 00:00:13,980
unleashed. But, well, agents
unleashed, change unleashed. We
5
00:00:13,980 --> 00:00:16,200
had a few, but it looks like
it's going to be agents
6
00:00:16,200 --> 00:00:19,860
unleashed. So coming soon, but
we've got to go and do all the
7
00:00:19,860 --> 00:00:24,260
admin that sits behind the
decision. In the meantime, the
8
00:00:24,260 --> 00:00:28,340
boys were very excited to talk
about ethics, and I don't know
9
00:00:28,340 --> 00:00:31,940
whether Niko, Niko was just
predicting the conversations he
10
00:00:31,940 --> 00:00:35,240
was going to have with his kids
this week, or whether it was a
11
00:00:35,240 --> 00:00:38,000
bit of chicken and egg, but over
to you, Niko. Take us away.
12
00:00:39,320 --> 00:00:41,800
Nikolaos Kaintantzis: Thank you.
Then that then that agents
13
00:00:41,800 --> 00:00:46,540
unleashed. Yeah, talking about
ethics. Great. I prepared
14
00:00:46,540 --> 00:00:49,540
something, and I thought it
wouldn't be cool if some AI
15
00:00:49,840 --> 00:00:54,100
persona would read this. And
then I ran out of time, so I'm
16
00:00:54,100 --> 00:00:56,860
now AI, and I'm reading what
I've prepared. If it's okay with
17
00:00:56,860 --> 00:01:00,040
you, with a sexy voice, it was
the idea. So let me prepare my
18
00:01:00,040 --> 00:01:05,520
sexy voice. So and then this
show, three voices, one
19
00:01:05,520 --> 00:01:10,680
question, AI and ethics. Stephan
speaks of responsibility that
20
00:01:10,680 --> 00:01:15,360
should never hide in a black
box. Mark. Mark wants awareness
21
00:01:15,480 --> 00:01:19,860
without reflection, just another
algorithm running blind. And
22
00:01:19,860 --> 00:01:23,480
finally, Niko, that's me,
reminds us ethics is not
23
00:01:23,480 --> 00:01:27,680
compliance. Is a socio political
responsibility, one stretches
24
00:01:27,740 --> 00:01:31,100
far beyond the goal itself. The
riddle remains, is
25
00:01:31,100 --> 00:01:35,300
accountability? Accountability a
burden we carry, or the compass
26
00:01:35,300 --> 00:01:40,160
that carries us forward? This
summary was provided by AI, so I
27
00:01:40,160 --> 00:01:44,200
just feed in all the things I
thought you will say, or I've
28
00:01:44,200 --> 00:01:47,320
seen the preparation, and there
was a summary, and they asked
29
00:01:47,320 --> 00:01:50,980
repair something, yeah, yeah,
still not catchy. I think I
30
00:01:50,980 --> 00:01:52,300
tried myself again. I know
31
00:01:53,140 --> 00:01:55,900
Mark Richards: you needed some
background music.
32
00:01:56,860 --> 00:01:59,860
Nikolaos Kaintantzis: Oh, for
sure. Let me. Let me call the
33
00:01:59,860 --> 00:02:08,880
kids. No, exactly so we really
agreed on AI and ethics, and I
34
00:02:08,880 --> 00:02:12,660
think it's a broad topic, and
the first thing I would like to
35
00:02:12,660 --> 00:02:17,220
know for you, and not not what,
what is your passion, or what
36
00:02:17,220 --> 00:02:20,360
scares you, or whatever, but I
would like to know what comes
37
00:02:20,360 --> 00:02:24,140
first in your mind when you hear
AI and ethics.
38
00:02:24,380 --> 00:02:28,640
Mark Richards: So when I read
that, I went without any control
39
00:02:28,640 --> 00:02:31,340
whatsoever, my mind went
straight back to like six or
40
00:02:31,340 --> 00:02:36,800
seven years ago, and it was
actually a moment where my son
41
00:02:36,800 --> 00:02:39,500
had decided that he was not a
Math Science Kid, he was a
42
00:02:39,500 --> 00:02:42,280
humanities kid, and he was
starting to think about what
43
00:02:42,280 --> 00:02:45,280
possible futures might look
like. And we wound up in this
44
00:02:45,400 --> 00:02:49,840
one hour long discussion about
ethics and AI and driverless
45
00:02:49,840 --> 00:02:53,680
cars. And he at the time, and,
you know, his early teens at
46
00:02:53,680 --> 00:02:56,800
this point, and life was very
black and white to him, and I
47
00:02:56,800 --> 00:02:59,380
was taking him down this, you
know, very long discussion to
48
00:02:59,380 --> 00:03:02,520
go. Well, let's imagine how
there could be ethical dilemmas
49
00:03:02,520 --> 00:03:06,840
for a driverless car. And he's
like, Well, surely not, surely
50
00:03:06,900 --> 00:03:09,900
it's just going to make better
decisions than humans. And I
51
00:03:09,900 --> 00:03:13,080
said, Well, what if it is
choosing between two accidents,
52
00:03:13,380 --> 00:03:16,740
and the accidents involve
different types of people? And
53
00:03:16,740 --> 00:03:19,380
he went, Well, you know, you
pick the ones that have got the
54
00:03:19,380 --> 00:03:25,040
least remaining lifespan. So had
this huge tunnel we dug down to
55
00:03:25,040 --> 00:03:28,940
go, you know, how do you get to
a point where the AI is making a
56
00:03:28,940 --> 00:03:33,140
good decision about life and
death choices, controlling a
57
00:03:33,140 --> 00:03:36,380
car? And, of course, just then,
there were starting to be people
58
00:03:36,380 --> 00:03:40,720
actually doing PhDs and schools
of study emerging around, you
59
00:03:40,720 --> 00:03:44,260
know, the whole notion of the
psychology of AI and ethics. But
60
00:03:44,260 --> 00:03:47,320
I think the thing that struck me
when I finished reminiscing on
61
00:03:47,320 --> 00:03:52,000
that hour of torturing my son in
the car was back then, when I
62
00:03:52,000 --> 00:03:55,900
thought about AI and ethics, it
was big, grand choices like, you
63
00:03:55,900 --> 00:04:02,220
know, who does the car choose to
hurt in a crisis, as opposed to
64
00:04:02,460 --> 00:04:07,080
actually today, for most of us,
there are ethical AI decisions
65
00:04:07,080 --> 00:04:10,680
happening every few minutes in
our lives. It's moved from a
66
00:04:10,680 --> 00:04:15,360
kind of almost science fiction
imagination thing into a day to
67
00:04:15,360 --> 00:04:19,020
day reality for us, and it's
like that's kind of snuck up.
68
00:04:19,259 --> 00:04:22,039
Nikolaos Kaintantzis: Yeah, I
had something similar in my mind
69
00:04:22,219 --> 00:04:27,139
when I had to answer this
question myself. Was exactly the
70
00:04:27,139 --> 00:04:30,199
question not, can AI be
responsible, but who is
71
00:04:30,199 --> 00:04:34,039
responsible? So if AI chooses
that somebody trained that is
72
00:04:34,039 --> 00:04:39,019
responsibility of the ones who
train the AI, or Yeah, who is
73
00:04:39,019 --> 00:04:42,519
responsible for that? So if he
chooses the one who was older,
74
00:04:42,819 --> 00:04:46,479
maybe he was one day away from a
breakthrough curing diseases,
75
00:04:46,599 --> 00:04:50,139
and the other one was just a
mean kid that will never do
76
00:04:50,139 --> 00:04:53,079
anything society, if we can
judge it, I don't know. Maybe
77
00:04:53,319 --> 00:04:56,559
this kid will change in 10
years. So I really thought if
78
00:04:56,559 --> 00:05:00,659
you drive yourself, you do this
decision instantly. And you have
79
00:05:00,659 --> 00:05:03,479
to live with the choice if
somebody else is doing this
80
00:05:03,479 --> 00:05:06,659
decision, or if AI is doing
decision who is really
81
00:05:06,659 --> 00:05:11,279
responsible. So the first thing
that came in my mind now was not
82
00:05:11,339 --> 00:05:16,439
if a car has to decide, but, uh,
when he decides, or it decides
83
00:05:16,739 --> 00:05:21,979
who is responsible and what is
wrong. So deciding wrong is also
84
00:05:22,039 --> 00:05:26,179
a nice set. But what is really
wrong? Killing the old, killing
85
00:05:26,179 --> 00:05:30,559
the young. What is what is old?
So you see, I hadn't read an
86
00:05:30,559 --> 00:05:34,699
answer, so I already started
talking with my family, since I
87
00:05:34,699 --> 00:05:37,639
prepared this episode, the
questions on Saturday after this
88
00:05:37,639 --> 00:05:42,219
show, and we had the whole week
discussions at home, about about
89
00:05:42,219 --> 00:05:46,359
ethics, about the future, and
was really cool discussions.
90
00:05:47,139 --> 00:05:49,839
Stephan, your what comes first
in mind when you hear AI and
91
00:05:49,839 --> 00:05:51,579
ethics? Yeah, I kind
92
00:05:51,580 --> 00:05:54,280
Stephan Neck: of have mixed
feelings when I read your
93
00:05:54,280 --> 00:05:58,780
question or your trigger. As an
Information Systems Engineer,
94
00:05:59,140 --> 00:06:04,620
I'm aware that systems need
data, and my mixed feelings are
95
00:06:04,620 --> 00:06:11,280
about privacy of data. Someone
will collect data. Someone it
96
00:06:11,280 --> 00:06:15,180
has to be stored somewhere, the
way it's stored, and then the
97
00:06:15,180 --> 00:06:20,960
way it is used. So this popped
up when you came up with this
98
00:06:20,960 --> 00:06:26,120
trigger. And looking back a few
of our past episodes, we were
99
00:06:26,120 --> 00:06:31,820
dealing with data, we need that
data. And yep, that's why two
100
00:06:31,820 --> 00:06:33,380
cents to this trigger.
101
00:06:33,620 --> 00:06:36,980
Nikolaos Kaintantzis: Wow, wow.
Data. Talking about data, do we
102
00:06:36,980 --> 00:06:41,680
really have the choice what we
share? So can we say, just know,
103
00:06:41,680 --> 00:06:43,060
and then what happens?
104
00:06:45,460 --> 00:06:46,600
Unknown: Very good question.
105
00:06:46,780 --> 00:06:48,940
Mark Richards: Why don't you
keep the ball rolling, Stephan,
106
00:06:48,940 --> 00:06:51,520
since you opened it, because I
think I'm going to be a little
107
00:06:51,520 --> 00:06:53,800
bit more flippant, and you're
probably going to be more
108
00:06:53,800 --> 00:06:54,400
systematic.
109
00:06:56,260 --> 00:07:00,400
Stephan Neck: Yeah, you know,
I'm, I'm actually, I'm actually
110
00:07:00,400 --> 00:07:07,260
in Scotland for a short holiday
trip, and we just experienced
111
00:07:07,440 --> 00:07:12,480
storm Amy. And you can be
structured as much as looking
112
00:07:12,480 --> 00:07:14,640
over regarding our flights,
113
00:07:14,880 --> 00:07:16,860
Mark Richards: it's Oh, you are
hacking
114
00:07:16,860 --> 00:07:21,180
Stephan Neck: over how much
debris you have in front on of
115
00:07:21,180 --> 00:07:24,920
the house on your lawn all the
day branches from the trees,
116
00:07:24,920 --> 00:07:25,220
right?
117
00:07:25,219 --> 00:07:28,879
Mark Richards: So looks like
you're coming and going with Wi
118
00:07:28,879 --> 00:07:33,319
Fi. By the way, Stephan, you
kind of were there and here, but
119
00:07:33,619 --> 00:07:34,219
I'll jump in.
120
00:07:35,000 --> 00:07:36,800
Nikolaos Kaintantzis: I think
your wife should stop watching
121
00:07:36,800 --> 00:07:40,600
Netflix and Disney plus and give
you the one bandwidth back. Just
122
00:07:40,600 --> 00:07:44,860
Just kidding. We just missed the
last, the last sentence of
123
00:07:44,860 --> 00:07:45,340
years.
124
00:07:46,180 --> 00:07:49,600
Stephan Neck: Yeah, what I said
is, what I said is, can you
125
00:07:49,600 --> 00:07:53,920
control everything now? Storm
Amy has taken over in Scotland,
126
00:07:53,980 --> 00:07:59,440
and this morning is kind of
under restructuring, how we
127
00:07:59,800 --> 00:08:03,180
behave, what we do and how it
affects our lives,
128
00:08:03,540 --> 00:08:06,480
Nikolaos Kaintantzis: yeah, but
it's not the same. So until now,
129
00:08:06,600 --> 00:08:11,520
I was in charge of my data, my
decisions. So storm is not
130
00:08:11,520 --> 00:08:14,760
really something I decide. And
then I realized, oh, I have no
131
00:08:14,760 --> 00:08:19,500
choice. It's something that just
happens. What's about data I
132
00:08:19,500 --> 00:08:23,780
have? So there comes a window
and says, log in with your that
133
00:08:23,780 --> 00:08:26,480
that account, and with that, you
give your soul and your privacy
134
00:08:26,480 --> 00:08:31,340
away. You can say yes or no.
Will it be okay to say no, and
135
00:08:31,340 --> 00:08:33,680
what happens if no? So what
happens with that decision?
136
00:08:34,459 --> 00:08:37,459
Stephan Neck: Yeah, let me be
picky on that one. Are we really
137
00:08:37,459 --> 00:08:44,139
in control, or are we just
thinking we are in control. I
138
00:08:44,139 --> 00:08:48,819
think we have the same basic
questions to answer with AI as
139
00:08:48,819 --> 00:08:54,939
well, but it's speeding up. You
get more opportunities to deal
140
00:08:54,939 --> 00:09:00,239
with data. The question about is
it, is it proper data? Is a
141
00:09:00,239 --> 00:09:05,759
clean data we are using stuff
like that is adding to the
142
00:09:05,759 --> 00:09:11,039
overall challenges we still
have. I don't think if I look
143
00:09:11,039 --> 00:09:16,379
back my last 3035, years in the
trade, yeah, we thought we could
144
00:09:16,499 --> 00:09:22,519
decide we could master
everything, but yeah, it opens
145
00:09:22,519 --> 00:09:23,899
up a new whole field for me.
146
00:09:24,860 --> 00:09:27,920
Mark Richards: I think for me, a
lot of people think they're in
147
00:09:27,920 --> 00:09:31,460
control today, but they don't
necessarily have the
148
00:09:31,460 --> 00:09:35,780
mindfulness. Like, what control
do I have? What kind of data am
149
00:09:35,780 --> 00:09:39,200
I sharing? Do I really have
awareness? And it's one of those
150
00:09:39,200 --> 00:09:42,760
things where I'm going to do two
kind of generational looks.
151
00:09:42,880 --> 00:09:45,520
There's a generational look for
me, which is, I've got a 20 year
152
00:09:45,520 --> 00:09:49,060
old son who grew up in the
digital age, and if I look at
153
00:09:49,060 --> 00:09:53,020
the way he treats data about
himself, he is so much more
154
00:09:53,020 --> 00:09:56,980
mindful about his personal data
than I am. He's mindful
155
00:09:56,980 --> 00:09:59,920
everybody. Somebody takes a
photo with him in it. He's mind.
156
00:10:00,060 --> 00:10:03,660
Full of his ability to control
people's access, right? He
157
00:10:03,660 --> 00:10:06,000
turned around to me a little
while ago and said, Dad, nobody
158
00:10:06,000 --> 00:10:09,000
shares phone numbers with each
other anymore. You share instas
159
00:10:09,300 --> 00:10:14,640
Because you can block somebody
from your Insta, right? And you
160
00:10:14,640 --> 00:10:19,260
get control. So these kids of
today, they grow up with layered
161
00:10:19,560 --> 00:10:22,640
views of their personal data and
how they're exposing it to the
162
00:10:22,640 --> 00:10:25,580
world. It's just built into
them, almost just a part of
163
00:10:25,580 --> 00:10:28,820
being society. But us older
people didn't grow up like that.
164
00:10:29,120 --> 00:10:35,240
You know, of course, I share my
phone number and and, you know,
165
00:10:35,240 --> 00:10:38,060
do I have different emails that
I use for different purposes?
166
00:10:38,060 --> 00:10:41,440
Well, I've often wished that I
did, but I don't. But any kid of
167
00:10:41,440 --> 00:10:45,400
today is going to have that, and
I think that that same level of
168
00:10:45,640 --> 00:10:49,900
am I aware, has to start to
creep in to questions I'm
169
00:10:49,900 --> 00:10:53,920
running with AI, right? Do I
know whether I have decisions
170
00:10:53,920 --> 00:10:58,120
about whether or not to share
information with chat GPT or
171
00:10:58,120 --> 00:11:01,080
with Claude? Do I know the
repercussions of different
172
00:11:01,080 --> 00:11:04,560
choices I make. You know, am I
even checking? There was
173
00:11:04,560 --> 00:11:07,320
recently a big furor because
Claude put out an update that
174
00:11:07,320 --> 00:11:11,160
ship switched the toggle from
automatically not sharing to
175
00:11:11,160 --> 00:11:14,520
automatically sharing, and there
was such a controversy blew up
176
00:11:14,520 --> 00:11:18,000
about it that a day later, they
reversed the decision. And we've
177
00:11:18,000 --> 00:11:21,620
got this world where people are
becoming more and more conscious
178
00:11:21,620 --> 00:11:24,500
of the commercial value of data,
right? We've been talking about
179
00:11:24,500 --> 00:11:30,080
it for years, and companies like
Google and Amazon and Facebook
180
00:11:30,200 --> 00:11:33,860
have built multi billion,
potentially multi trillion
181
00:11:33,860 --> 00:11:37,280
dollar empires on the basis of
the commercial value of data.
182
00:11:37,280 --> 00:11:40,360
Most of us don't, but you look
around and what you're seeing at
183
00:11:40,360 --> 00:11:43,360
the moment, and all of the
companies, let's say LinkedIn
184
00:11:43,360 --> 00:11:46,180
works very hard to make sure the
only way you get at LinkedIn
185
00:11:46,660 --> 00:11:50,260
data is logged in manually and
tracking. Instagram, it's pretty
186
00:11:50,260 --> 00:11:53,140
easy to get at their data, but
they're busy locking it down, so
187
00:11:53,140 --> 00:11:56,140
it's harder to get at their
data. People are getting a lot
188
00:11:56,140 --> 00:11:59,980
more focused about putting
messages on their websites.
189
00:11:59,980 --> 00:12:02,880
Right there's a file, I think
it's called llms, dot text,
190
00:12:03,180 --> 00:12:06,420
which actually communicates what
you are and aren't willing to
191
00:12:06,420 --> 00:12:11,220
share from your website to llms.
Is everybody using that? Do you
192
00:12:11,220 --> 00:12:14,100
have the awareness to know what
you're sharing? Because if you
193
00:12:14,100 --> 00:12:16,860
don't know what you're sharing,
you've never got any choices
194
00:12:16,860 --> 00:12:19,860
going on, because you're not
making choices. You're just
195
00:12:19,860 --> 00:12:20,480
letting it
196
00:12:20,480 --> 00:12:24,200
Stephan Neck: fly by, but that
that leads to a situation of
197
00:12:24,200 --> 00:12:29,240
take it or leave it right, and
you're deciding in that context
198
00:12:29,240 --> 00:12:33,380
with your data, because your
data is like has a price tag as
199
00:12:33,380 --> 00:12:39,320
well. Am I sharing it? Am I? Am
I blocking it? And if I block
200
00:12:39,320 --> 00:12:44,140
too much, I'm probably out of
the inner circle. I'm not within
201
00:12:44,140 --> 00:12:50,560
the fence. I can't use it. So
that makes me think, and for me,
202
00:12:50,560 --> 00:12:55,240
the producers, the creators,
have a huge responsibility
203
00:12:55,240 --> 00:13:01,560
regarding the usage of data,
collection of data, bringing it
204
00:13:01,560 --> 00:13:07,080
in, into the systems, and where
do you position this Take it or
205
00:13:07,080 --> 00:13:11,880
leave it right? Is it a binary
thing, or is it kind of an area
206
00:13:11,880 --> 00:13:17,460
where you really are in control,
where you start deciding, am I
207
00:13:17,460 --> 00:13:24,560
in and to what kind of range Am
I in, in a system, or in an in
208
00:13:24,560 --> 00:13:26,300
an even in an ecosystem,
209
00:13:27,559 --> 00:13:29,839
Nikolaos Kaintantzis: the more I
think, the less I think I have
210
00:13:29,839 --> 00:13:33,379
it under control. So let's go
into WhatsApp. When you have
211
00:13:33,379 --> 00:13:36,199
WhatsApp and somebody has your
number, you're already part of
212
00:13:36,199 --> 00:13:40,579
the system, even if you never
agreed, and they are saying bye,
213
00:13:40,579 --> 00:13:44,619
bye, if you are part of
WhatsApp, you took care that
214
00:13:44,619 --> 00:13:46,899
everybody on your phone book
agree that they are on the phone
215
00:13:46,899 --> 00:13:50,919
book. So what the hell question
is that? So I'm sure it's not
216
00:13:50,919 --> 00:13:53,859
really legal to say your
responsibility that everybody in
217
00:13:53,859 --> 00:13:58,539
your phone book agrees that your
phone book now is public. Yet, I
218
00:13:58,839 --> 00:14:01,139
think Stephan, you're right. You
convinced somehow we are not
219
00:14:01,139 --> 00:14:06,659
really under control of the
data. And Mark, I think it's
220
00:14:06,659 --> 00:14:13,499
also the way you or we handle
data, because I see also kids.
221
00:14:13,799 --> 00:14:16,979
Let's go to my youngest
daughter. They have no idea
222
00:14:16,979 --> 00:14:20,419
about the stuff, so the older
ones, I think they're much more
223
00:14:20,419 --> 00:14:26,539
careful. I remember last time
when my middle kid watched me
224
00:14:26,539 --> 00:14:30,259
how I just accept all cookies
without reading it, and she was
225
00:14:30,319 --> 00:14:34,039
really, how can you? And
thought, oh, that's just a dummy
226
00:14:34,039 --> 00:14:36,679
browser which just does
everything. And then I close it,
227
00:14:36,679 --> 00:14:40,659
and then I'm a completely fresh
new person. Ah, okay, because
228
00:14:40,659 --> 00:14:45,279
she really tries to handle what
she wants to give away or not.
229
00:14:45,639 --> 00:14:50,679
But Stephan, you're right if
you're anyway part of it. And
230
00:14:50,679 --> 00:14:53,199
the other question is how much
you want to interact with that.
231
00:14:53,199 --> 00:14:56,679
So for instance, our youngest
doesn't want to be part of
232
00:14:56,679 --> 00:15:02,519
what's up. Just she's missing a
lot of. Loss gossip, which, by
233
00:15:02,519 --> 00:15:05,519
the way, I think it's good that
she misses, but that's another
234
00:15:05,519 --> 00:15:08,939
point she maybe misses, also,
also important, important stuff.
235
00:15:09,000 --> 00:15:11,220
Mark Richards: So what's such a
great example to me of a
236
00:15:11,220 --> 00:15:15,480
generational divide? Thing we
all know about the fact that,
237
00:15:15,480 --> 00:15:18,660
basically, by continuing to use
it, we're giving Facebook a
238
00:15:18,660 --> 00:15:24,560
stupid amount of our data. What
my children know is that their
239
00:15:24,560 --> 00:15:27,920
parents are old and they want to
use WhatsApp. Therefore, because
240
00:15:27,920 --> 00:15:30,740
they don't view family
interaction as particularly data
241
00:15:30,740 --> 00:15:33,380
sensitive, they're willing to
use it for family interaction,
242
00:15:35,000 --> 00:15:38,180
but they would never use
WhatsApp, right? They literally,
243
00:15:38,180 --> 00:15:40,840
and I watch, I watch them talk
about these things, make these
244
00:15:40,840 --> 00:15:44,680
calls, they literally go, what
are the privacy concerns I've
245
00:15:44,680 --> 00:15:47,140
got with different ways of
communicating, and which way am
246
00:15:47,140 --> 00:15:50,380
I going to communicate based on
the privacy and I care about my
247
00:15:50,380 --> 00:15:55,540
ownership of that data
awareness, so
248
00:15:55,540 --> 00:15:59,260
Nikolaos Kaintantzis: you guys
read the whole stuff. So before
249
00:15:59,260 --> 00:16:01,320
you sign up, sign in, sign up
for something.
250
00:16:02,040 --> 00:16:04,380
Mark Richards: I had a friend
who did a PhD on whether people
251
00:16:04,380 --> 00:16:04,920
read those.
252
00:16:08,100 --> 00:16:10,620
Nikolaos Kaintantzis: You do the
meta reading, okay? Stephan,
253
00:16:12,240 --> 00:16:18,000
Stephan Neck: No, obviously not.
But I the past few years, maybe
254
00:16:18,000 --> 00:16:22,040
once or twice, I really took the
time to read it. First of all,
255
00:16:22,100 --> 00:16:25,100
do I really understand
everything? No, you should be a
256
00:16:25,100 --> 00:16:31,040
legal person to understand
every, every part of such a such
257
00:16:31,040 --> 00:16:37,400
a text, in the end, and I must
confess, I don't care, because I
258
00:16:37,400 --> 00:16:40,420
want to take part in that
service. I want to have this
259
00:16:40,420 --> 00:16:44,500
action. So I have to agree, it's
again this, take it or leave it
260
00:16:44,500 --> 00:16:49,480
right, and within seconds, or
pretty fast, you have to make a
261
00:16:49,480 --> 00:16:51,400
decision. But I
262
00:16:51,760 --> 00:16:53,680
Mark Richards: think, like
there's a little bit deeper
263
00:16:53,680 --> 00:16:56,080
there for us, right? So if you
talk about what you just said
264
00:16:56,080 --> 00:16:59,680
there, Stephan was, I'm going to
make a mindful choice. Yeah, I
265
00:16:59,680 --> 00:17:02,940
know what I'm giving up, and I'm
willing to pay it. But if you
266
00:17:02,940 --> 00:17:05,580
think about being in a world
where you're trying to be an AI
267
00:17:05,580 --> 00:17:07,920
change agent, you're trying to
help a company, think about
268
00:17:07,920 --> 00:17:11,700
their AI strategy, what data
they're potentially giving away
269
00:17:12,000 --> 00:17:14,640
when they use chat GPT to help
them write an email, what
270
00:17:14,640 --> 00:17:17,160
personally identifiable
information are they potentially
271
00:17:17,160 --> 00:17:22,220
sharing with the world in doing
that? And do you have an
272
00:17:22,220 --> 00:17:25,640
awareness of the potential
exposure of the types of
273
00:17:25,640 --> 00:17:28,700
information that we tend to
share in an AI setting? If you
274
00:17:28,700 --> 00:17:31,400
don't have that, then you're not
going to help your conversation
275
00:17:31,400 --> 00:17:33,800
client think about making
mindful choices. So I think
276
00:17:33,800 --> 00:17:37,640
there's a real duty of care to
us if we're coaching clients in
277
00:17:37,640 --> 00:17:41,200
the AI space to be actually
highly aware of some of the
278
00:17:41,200 --> 00:17:45,580
implications around data and the
various ways that data can get
279
00:17:45,580 --> 00:17:48,640
out of your world from an
ethical perspective, Yeah,
280
00:17:48,700 --> 00:17:49,060
agreed.
281
00:17:49,059 --> 00:17:53,559
Stephan Neck: Mark as a
consultant, I get all of a
282
00:17:53,559 --> 00:18:00,279
sudden different NDAs to sign
right extended ones, or now
283
00:18:00,339 --> 00:18:03,659
companies really being aware, as
you mentioned, they're giving
284
00:18:03,659 --> 00:18:07,619
away more than they think.
Right? Talking to a customer
285
00:18:07,619 --> 00:18:11,399
about strategy, talking to a
customer about a lean portfolio
286
00:18:11,399 --> 00:18:17,819
management, it opens up a whole
new range of possibilities, but
287
00:18:17,819 --> 00:18:23,119
also access to data, which then,
with the help of AI, gets a
288
00:18:23,119 --> 00:18:28,339
whole new a whole new sphere of
options and possibilities in my
289
00:18:28,339 --> 00:18:28,759
trade.
290
00:18:29,360 --> 00:18:31,880
Mark Richards: But that's what
makes it really fascinating to
291
00:18:31,880 --> 00:18:35,300
me, right? Is every potential
internal application of AI is
292
00:18:35,300 --> 00:18:39,560
gonna be fueled by data, right?
So you want more and more data
293
00:18:39,560 --> 00:18:42,400
available to AI because it opens
up more and more possibilities
294
00:18:42,400 --> 00:18:45,220
to you, but at the same time, it
opens up more and more risk
295
00:18:45,220 --> 00:18:48,520
surface. And I think, you know,
the world is caught between
296
00:18:48,520 --> 00:18:52,120
companies that are just blindly
taking an awful lot of risks and
297
00:18:52,120 --> 00:18:54,640
companies that are so concerned
about the risk they're not using
298
00:18:54,640 --> 00:18:56,920
AI at all. And I think there's
gonna be a really interesting
299
00:18:56,920 --> 00:19:00,160
time for us in the next couple
of years as we try and find the
300
00:19:00,160 --> 00:19:01,620
space to hold between those two.
301
00:19:01,620 --> 00:19:06,240
Nikolaos Kaintantzis: I would
like to add on that the leaders
302
00:19:06,240 --> 00:19:09,840
think the people don't use AI
because they just restricted
303
00:19:09,840 --> 00:19:13,020
everything. But the other thing
is, we have aI on our pocket, on
304
00:19:13,020 --> 00:19:16,800
our on our mobile phone, so they
use just different ways, which
305
00:19:16,800 --> 00:19:19,620
is much more dangerous because
now I use it as a private
306
00:19:19,620 --> 00:19:22,640
person. And we just heard how
Stephan act as a private person,
307
00:19:22,640 --> 00:19:26,840
different as a consultant. So so
now we're doing it on our mobile
308
00:19:26,840 --> 00:19:32,000
phones, taking pictures of our
screens and feeding to a free
309
00:19:32,000 --> 00:19:36,980
AI, which has no limitations at
all. So instead of having a
310
00:19:36,980 --> 00:19:42,344
proper governance, we took
decision, Oh, it's nothing for
311
00:19:42,344 --> 00:19:46,364
us, but people still using it so
and that's one of the huge
312
00:19:46,364 --> 00:19:50,504
problems, or one of the we
talked about blind spots, I
313
00:19:50,504 --> 00:19:54,284
think two episodes ago, it's
something people don't or
314
00:19:54,284 --> 00:19:57,764
management, manager, manager,
don't realize, if you just
315
00:19:58,784 --> 00:20:04,020
banish. AI, it's still there,
just on the legal way or in the
316
00:20:04,020 --> 00:20:07,680
private way. Yeah, so in my
case, what I'm doing with with
317
00:20:07,680 --> 00:20:11,640
the whole privacy part. When I
was younger, I read everything,
318
00:20:11,700 --> 00:20:17,940
and I remember, I remember days
when it was on paper and they
319
00:20:17,940 --> 00:20:22,040
just take a pen and remove parts
I didn't like before I signed a
320
00:20:22,040 --> 00:20:25,340
contract, which in Switzerland
is completely legal if the other
321
00:20:25,340 --> 00:20:29,180
person says nothing, that's the
new contract. That's the new new
322
00:20:29,180 --> 00:20:33,080
part of the contract. And it
changed because now too much
323
00:20:33,080 --> 00:20:37,820
reading, it's too many tools,
too many things. And what I'm
324
00:20:37,820 --> 00:20:42,040
doing now is really asking AI to
summarize what they want from
325
00:20:42,040 --> 00:20:45,700
me, and ask also, how is it
compared to something else? You
326
00:20:45,700 --> 00:20:49,780
see, things I shouldn't sign, at
least I'm aware of, but I'm with
327
00:20:49,780 --> 00:20:53,320
Stephan, have no choice. I have
to be in or out, and at least I
328
00:20:53,320 --> 00:20:55,120
try to find out what it means
when I'm in.
329
00:20:55,120 --> 00:20:58,180
Mark Richards: And then, just to
top it off, what you're doing at
330
00:20:58,180 --> 00:21:01,020
that moment is you're giving AI
data about your internal
331
00:21:01,020 --> 00:21:04,020
contractual arrangements with
your clients that may well be
332
00:21:04,020 --> 00:21:04,800
confidential.
333
00:21:05,940 --> 00:21:10,500
Stephan Neck: Yeah, exactly. It
sounds like the start of a
334
00:21:10,500 --> 00:21:11,580
vicious circle, right?
335
00:21:13,200 --> 00:21:15,720
Nikolaos Kaintantzis: Yeah, I
didn't say it was NDAs, but
336
00:21:15,720 --> 00:21:19,020
you're right, yeah, if you
extend it on this way, you
337
00:21:19,020 --> 00:21:23,600
realize, okay, you give it a lot
of information, yes, yes. But
338
00:21:24,500 --> 00:21:27,080
even if you just send it per
email, you give the information
339
00:21:27,080 --> 00:21:30,140
also the whole world, because
it's just a postcard from one
340
00:21:30,140 --> 00:21:35,060
person or the other. But wow,
let's, let's go to a little
341
00:21:35,060 --> 00:21:38,120
different topic. But on a
section about future
342
00:21:38,120 --> 00:21:40,780
generations, we talked about
kids, about they are different.
343
00:21:40,780 --> 00:21:45,820
Their handle data, how aware
they are. I personally think
344
00:21:45,880 --> 00:21:50,080
they are aware because parents
were aware too. Because I'm also
345
00:21:50,080 --> 00:21:53,020
sure some parents don't care,
and the kids will also not care.
346
00:21:53,440 --> 00:21:56,800
So by the way, if I imagine how
many people, how many parents
347
00:21:56,860 --> 00:22:02,040
just post pictures of naked
babies and naked young kids. I'm
348
00:22:02,820 --> 00:22:05,760
not sure if these kids are more
aware or later, if they have a
349
00:22:08,760 --> 00:22:13,440
if it comes back again when they
are older, but this brought me
350
00:22:13,440 --> 00:22:17,340
to the question about future
generations. What more
351
00:22:17,340 --> 00:22:20,960
obligations do we have when
we're deploying AI today? What
352
00:22:20,960 --> 00:22:24,440
can we destroy? What can we
what's Robert G obligate,
353
00:22:24,440 --> 00:22:27,500
obligate. Ah, come on, English
obligations.
354
00:22:29,000 --> 00:22:32,480
Mark Richards: Who wants to
start? I drew a blank. So it's
355
00:22:32,480 --> 00:22:36,080
over to you guys. So first time
ever I sat and thought five
356
00:22:36,080 --> 00:22:38,420
minutes I went, I really don't
know what I'm going to answer to
357
00:22:38,420 --> 00:22:39,020
this. So you're
358
00:22:39,020 --> 00:22:43,000
Nikolaos Kaintantzis: too old
for that shit or what to know.
359
00:22:43,060 --> 00:22:47,740
Stephan Neck: I like your
question. Niko and I scribbled
360
00:22:47,740 --> 00:22:51,580
down a few things, and then you
mentioned your family, and
361
00:22:51,820 --> 00:22:57,100
thinking about my family, my
kids, my grandkids, and when I
362
00:22:57,100 --> 00:23:04,080
summarized my sketches and my
notes again using AI. It came up
363
00:23:04,080 --> 00:23:09,600
with the term of stewardship,
right? I'm not telling my kids
364
00:23:09,960 --> 00:23:14,880
how to live their life, but what
is my stewardship, towards the
365
00:23:14,880 --> 00:23:20,240
family, towards my kids, the
next few generations, and I'm
366
00:23:20,240 --> 00:23:24,560
not using the word obligations,
although I feel like that. I do
367
00:23:24,560 --> 00:23:28,700
have my obligations for the
future, although I'll disappear
368
00:23:28,700 --> 00:23:32,840
one day, and that's it. And I
think it's about, we already
369
00:23:32,840 --> 00:23:36,260
talked about that. How do you
control environments? How do you
370
00:23:36,320 --> 00:23:39,680
control certain things, right?
And some are control freaks.
371
00:23:40,040 --> 00:23:43,780
Some don't even care. There's
the whole range and and behind
372
00:23:43,900 --> 00:23:49,420
this, this whole range for me is
I want to prevent harm, for my
373
00:23:49,420 --> 00:23:53,620
family, for other people, for my
friends. So this is, this is one
374
00:23:53,620 --> 00:23:57,400
of the obligations I probably
have as a producer, as a creator
375
00:23:57,400 --> 00:24:03,660
of AI stuff. The other one is,
every culture has values, and we
376
00:24:03,660 --> 00:24:11,400
as human beings, what I cherish
is dignity, right to live a life
377
00:24:11,760 --> 00:24:16,020
in dignity. And if you look at
this world, first world, second
378
00:24:16,020 --> 00:24:19,560
world, third world, there's not
the same dignity for everyone.
379
00:24:19,560 --> 00:24:23,840
And now a new technology comes
in, new stuff comes in, which
380
00:24:23,840 --> 00:24:31,160
probably shakes up those values
and how we we keep, yeah,
381
00:24:31,160 --> 00:24:36,200
dignity, and with that comes
what I like. If I look back in
382
00:24:36,200 --> 00:24:40,120
my life, it's what is a fair
share. A fair share I know
383
00:24:40,120 --> 00:24:44,080
something. How do I share my
knowledge I do have something.
384
00:24:44,080 --> 00:24:48,400
How do I share with my friends?
Am I a good host at home? Is
385
00:24:48,400 --> 00:24:54,160
there an open door policy? Do we
have guests over for a coffee or
386
00:24:54,400 --> 00:25:00,660
for an event? Or is it No, this
is my this is my castle. Uh,
387
00:25:00,660 --> 00:25:06,480
everyone else is out. And the
funny thing, AI came back with
388
00:25:06,900 --> 00:25:12,600
an idea, calling it a user
manual for the future. So when
389
00:25:12,600 --> 00:25:18,780
it comes to the stewardship, are
we not only leaving traces and
390
00:25:18,780 --> 00:25:25,040
systems? Are we using it with a
manual so people understand why
391
00:25:25,040 --> 00:25:29,420
we built stuff the way we built
it. How should it be used?
392
00:25:29,480 --> 00:25:33,380
Because that's that's, for me,
what the user manual is, right?
393
00:25:34,460 --> 00:25:38,060
My brother eats user manuals. I
don't. I just fiddle around with
394
00:25:38,060 --> 00:25:42,580
the new stuff, and then I
probably hit the wall, and I
395
00:25:42,580 --> 00:25:46,060
have to go back to the user
manual. Why? While my brother
396
00:25:46,180 --> 00:25:51,220
reading the user manual is much
cleverer than I am, and he gets
397
00:25:51,220 --> 00:25:57,040
faster in no time understanding
what the system is, how it works
398
00:25:57,160 --> 00:25:59,860
and how it should be used. And
this leading back to the
399
00:25:59,860 --> 00:26:02,880
stewardship we have as as
people,
400
00:26:02,999 --> 00:26:04,319
Nikolaos Kaintantzis: my
thoughts is going now in a
401
00:26:04,319 --> 00:26:07,679
direction. If it becomes too
political, please stop me and we
402
00:26:07,679 --> 00:26:10,439
do something else, because I
still want to visit some
403
00:26:10,439 --> 00:26:15,299
countries in the next years.
What really touched me, and I
404
00:26:15,299 --> 00:26:20,999
agree with that, is the idea of
fair share, and we have now a
405
00:26:21,059 --> 00:26:25,999
technology that takes away,
maybe jobs, but that can makes
406
00:26:25,999 --> 00:26:29,659
us much more wealthier, and now
we can share. But when I look
407
00:26:29,659 --> 00:26:34,279
what happens in the world in
politics, I'm just wondering,
408
00:26:34,879 --> 00:26:37,639
will it be a nice share, or we
would just be the richer
409
00:26:37,639 --> 00:26:42,459
Richards, so what I'm with you
is I really want to avoid some
410
00:26:42,639 --> 00:26:48,699
in reverse, in reversible, in
reversible damage. But I don't
411
00:26:48,699 --> 00:26:52,659
know if those people who are in
charge really think that way, or
412
00:26:52,659 --> 00:26:55,119
they're just playing now
whatever they have to play the
413
00:26:55,119 --> 00:26:57,999
role. But there's a wife, and
that's what makes me a little
414
00:26:57,999 --> 00:27:01,019
bit sad. But I think now we are
going we derailed somewhere in a
415
00:27:01,019 --> 00:27:05,459
discussion which is not part of
agent unleashed, which is just a
416
00:27:05,699 --> 00:27:09,359
criticism on society or
criticism of the riches I don't
417
00:27:09,359 --> 00:27:10,619
know. Mark help me.
418
00:27:12,480 --> 00:27:17,100
Mark Richards: I was just about
Ali on the side track because so
419
00:27:17,100 --> 00:27:22,340
my son is very much an
environmental activist, and he
420
00:27:22,340 --> 00:27:25,400
is working on making me more
environmentally aware every
421
00:27:25,400 --> 00:27:30,140
week. So we have a lot of
discussions about, you know, the
422
00:27:30,140 --> 00:27:35,000
challenge of environmentalism is
that some countries may be doing
423
00:27:35,000 --> 00:27:39,980
something about it, while a lot
of the world is not. And how do
424
00:27:39,980 --> 00:27:42,040
you deal with that, because it's
the same kind of thing, right?
425
00:27:42,340 --> 00:27:45,820
If you're ethical
environmentally, but your
426
00:27:45,820 --> 00:27:49,300
neighbors aren't, and we've got
a great example with Australia,
427
00:27:49,300 --> 00:27:51,640
we've been trying to be more
ethical about managing our fish
428
00:27:51,640 --> 00:27:56,680
stocks and putting a lot of
licensing controls over it to
429
00:27:56,680 --> 00:28:00,040
allow fish stocks to replenish.
But we have neighboring
430
00:28:00,040 --> 00:28:03,300
countries who are poor, and they
come and they fish in our waters
431
00:28:04,620 --> 00:28:08,460
because they are not living the
same things. And I think you get
432
00:28:08,460 --> 00:28:13,260
that same question with ethics,
becomes really quite
433
00:28:13,260 --> 00:28:16,260
fascinating, but I reckon we
should have a crack at Harm's
434
00:28:16,260 --> 00:28:20,960
answer before we get distracted
by jiggles. So harm came in and
435
00:28:20,960 --> 00:28:24,380
said, What about a scenario with
using AI to enforce ethics? And
436
00:28:24,380 --> 00:28:26,900
actually, the thing that came
straight to mind was, I was
437
00:28:26,900 --> 00:28:32,180
watching a video a couple of
days ago, somebody was going
438
00:28:32,180 --> 00:28:34,820
through, of course, they leaked
the news Claude 4.5 system
439
00:28:34,820 --> 00:28:38,300
prompt. And somebody was going
through doing a bit of a
440
00:28:38,300 --> 00:28:40,900
dissection about interesting
things they found in the system
441
00:28:40,900 --> 00:28:45,040
prompt, and one of the things
they found was an awful lot of
442
00:28:45,100 --> 00:28:48,520
anti plagiarism prompting. So
there are a whole bunch of
443
00:28:48,520 --> 00:28:52,480
sections in the new 4.5 system
prompt to actually stop it from
444
00:28:52,480 --> 00:28:58,600
doing direct cited quotes that
might breach copyright. And you
445
00:28:58,600 --> 00:29:01,560
can see they're starting to try
and get to, well, people are
446
00:29:01,560 --> 00:29:05,520
using it, and it's doing a whole
bunch of copyright breaches. You
447
00:29:05,520 --> 00:29:08,220
know, the suspicion, of course,
is they're protecting themselves
448
00:29:08,220 --> 00:29:11,280
from lawsuits. People are
actually starting to get into,
449
00:29:12,000 --> 00:29:14,820
how do I prompt, whether it's in
a system prompt, or whether it's
450
00:29:14,820 --> 00:29:19,200
in an Asian definition, Agent
definition, how do I prompt for
451
00:29:19,200 --> 00:29:23,120
ethical behavior and or, How do
I prompt to detect unethical
452
00:29:23,120 --> 00:29:27,020
behavior? So people are starting
to do it. I think the really
453
00:29:27,020 --> 00:29:30,320
interesting question becomes,
how do you codify your ethics in
454
00:29:30,320 --> 00:29:32,480
a way that a prompt is capable
of interpreting?
455
00:29:32,900 --> 00:29:36,800
Stephan Neck: Let me pick up
this one mark, going back to
456
00:29:37,520 --> 00:29:41,320
Stephan beers, viable system
model. There's this system
457
00:29:41,320 --> 00:29:45,220
number five, which is the
normative system, ring fencing,
458
00:29:45,220 --> 00:29:51,460
the other systems in a viable
system. And I high, I'm a high
459
00:29:51,580 --> 00:29:56,140
advocate on this question that
comes from harm. It's about the
460
00:29:56,140 --> 00:30:03,900
scenario that allows the. A
company to from onboarding
461
00:30:03,900 --> 00:30:07,500
people during their work until
they leave a company to
462
00:30:07,500 --> 00:30:11,400
accompany them in a very
supportive way, right? Even
463
00:30:11,400 --> 00:30:15,600
saying no to certain actions, as
you mentioned. Am I allowed to
464
00:30:15,600 --> 00:30:20,840
use such a prompt and before the
prompt gives back a result, you
465
00:30:20,840 --> 00:30:25,280
probably get back a message,
this is not how we think as a
466
00:30:25,280 --> 00:30:29,840
company. And what you just, what
you just mentioned, mark for me
467
00:30:30,140 --> 00:30:34,460
is, I would say, and I'm being
very blunt about this, 190, 9%
468
00:30:34,880 --> 00:30:39,200
of all companies don't have
ethics. They can't express it.
469
00:30:39,500 --> 00:30:43,540
How can you bring it into a
scenario, and bring it into the
470
00:30:43,540 --> 00:30:47,800
AI world if you don't know what
you are talking about, if you
471
00:30:47,800 --> 00:30:53,140
don't know how to how to
describe it and to even script
472
00:30:53,140 --> 00:30:57,520
or code it in a system. But the
scenario, yes, we have to deal
473
00:30:57,520 --> 00:31:02,100
with that one. And it's again,
going back to leadership, this
474
00:31:02,100 --> 00:31:07,200
is one of the big obligations
you have as a C level to make it
475
00:31:07,200 --> 00:31:08,940
happen in your organization.
476
00:31:09,240 --> 00:31:12,180
Nikolaos Kaintantzis: I'm now
sad and sorry, mad and happy in
477
00:31:12,180 --> 00:31:15,420
the same time, because I just
realized it's something we have
478
00:31:15,420 --> 00:31:18,240
to do as companies, as
individuals. I just looked the
479
00:31:18,240 --> 00:31:21,080
whole time for an answer. How
can the whole society enforce
480
00:31:21,080 --> 00:31:24,140
this. And that's much more
difficult, because I still think
481
00:31:24,140 --> 00:31:28,640
some companies, some crazy
people, will not do that. So,
482
00:31:29,480 --> 00:31:33,260
but yeah, being there as a
company, as a person, as a
483
00:31:33,260 --> 00:31:36,500
human, and saying this, this is
our ethic, and I want to enforce
484
00:31:36,500 --> 00:31:40,180
this in our AI, that makes
absolutely sense. I haven't seen
485
00:31:40,180 --> 00:31:44,140
a solution, how we can enforce
it worldwide. So one discussion
486
00:31:44,140 --> 00:31:48,340
we had with the family was, what
happens to masks AI, which was
487
00:31:48,340 --> 00:31:53,200
somehow half an Nazi, half
whatever. So where was there the
488
00:31:53,260 --> 00:31:57,340
ethics, or is it normal? What is
what it's doing? So I think some
489
00:31:57,340 --> 00:32:01,020
companies will will have, as I
said, have ethics and should and
490
00:32:01,020 --> 00:32:04,320
will enforce it, and others will
just take whatever they have. I
491
00:32:04,740 --> 00:32:07,320
also thought about it, what's,
what's with open source and AI?
492
00:32:07,560 --> 00:32:11,340
If an AI is open source, I can
remove all the control parts.
493
00:32:11,340 --> 00:32:13,980
That was where I'm going with
the questions I'm really happy
494
00:32:13,980 --> 00:32:16,380
you answered in a different way.
There was the happy part and the
495
00:32:16,380 --> 00:32:18,780
sad part was what I just said,
sorry if I make the day.
496
00:32:18,780 --> 00:32:21,680
Everybody said it wasn't,
wasn't, the intention.
497
00:32:21,740 --> 00:32:22,820
Mark Richards: That's not your
job. Niko,
498
00:32:26,360 --> 00:32:29,180
Nikolaos Kaintantzis: clever
guy, thank you so much. Let's
499
00:32:29,180 --> 00:32:32,480
use also, let's mess up with our
brains. So let's, let's go to
500
00:32:32,480 --> 00:32:38,180
jiggle and where, each time I
have the chance to give you
501
00:32:39,080 --> 00:32:42,760
Yeah, to activate other parts of
the brain, and this time, it's a
502
00:32:42,760 --> 00:32:46,840
hard one. Well, also hard for
me, if AI and ethics were an
503
00:32:46,840 --> 00:32:51,640
exhibit in a museum, again, if
AI and ethics was an exhibit in
504
00:32:51,640 --> 00:32:56,260
a museum, what does it look
like, and which museum exhibits
505
00:32:56,260 --> 00:32:56,620
it
506
00:32:56,860 --> 00:32:59,080
Mark Richards: so I'm going to
go first. And when I first
507
00:32:59,080 --> 00:33:01,800
looked at this one, I went, I've
got absolutely no idea. I left
508
00:33:01,800 --> 00:33:04,560
it alone, and then I picked it
up again, and it was like I got
509
00:33:04,560 --> 00:33:08,040
an instant answer. So here's my
instant answer. It's a very
510
00:33:08,040 --> 00:33:11,400
lightly re flavored Mongols
visit. So if you think about the
511
00:33:11,400 --> 00:33:14,940
story of the Mongols in a
museum, they came, they saw they
512
00:33:14,940 --> 00:33:18,300
conquered. They became the thing
that they conquered. And the
513
00:33:18,300 --> 00:33:21,260
question that I leave with you
is, in this scenario, are the
514
00:33:21,260 --> 00:33:23,300
humans, the Mongols, or the
Chinese?
515
00:33:24,020 --> 00:33:24,860
Unknown: That's a good one.
516
00:33:26,060 --> 00:33:28,040
Nikolaos Kaintantzis: Oh yeah.
But by the way, before, before I
517
00:33:28,040 --> 00:33:34,400
hand over to Stephan, each time
we we use Mark's custom chat GPT
518
00:33:34,400 --> 00:33:37,520
to generate an image. And this
time, I really love the image,
519
00:33:37,520 --> 00:33:40,720
because it showed exactly what
you're seeing. So who is which?
520
00:33:40,780 --> 00:33:46,000
So who is? Who is the origin,
who is the master, who is giving
521
00:33:46,000 --> 00:33:50,860
whom the torch? What the heck is
the kids doing? So I think,
522
00:33:50,860 --> 00:33:53,920
unfortunately, it just
disappears from LinkedIn as soon
523
00:33:53,920 --> 00:33:57,040
as we, as we, as we did the
show. So I think maybe we should
524
00:33:57,040 --> 00:34:00,780
have, like a gallery walk around
all the images this time I
525
00:34:00,780 --> 00:34:04,980
really liked what the co
creation with AI was. And at the
526
00:34:04,980 --> 00:34:09,480
end, I think this time AI was,
was, was the master of the
527
00:34:09,480 --> 00:34:12,660
ideas, and I just tweaked
around. So I really like this,
528
00:34:12,660 --> 00:34:16,620
this picture and this one just
symbolize your story. Mark at
529
00:34:16,680 --> 00:34:19,380
the end, I like this very much.
Stephan,
530
00:34:19,740 --> 00:34:23,420
Stephan Neck: yeah, well, I You
said, Where would you see it? In
531
00:34:23,420 --> 00:34:27,980
a museum. And if you, if you go
to museums, you look at things.
532
00:34:28,040 --> 00:34:34,460
And the idea behind magical was,
what if you look at something
533
00:34:34,460 --> 00:34:37,400
and it looks back and you
recognize, oh, it's a mirror.
534
00:34:37,460 --> 00:34:41,080
It's an algorithmic mirror,
because it's about math, it's
535
00:34:41,080 --> 00:34:45,160
about statistics, it's about
data and how we deal with that
536
00:34:45,160 --> 00:34:49,360
so reflecting our digital
selves. And where would I go to
537
00:34:49,420 --> 00:34:55,900
the Deutsches Museum in Munich,
which probably has written marks
538
00:34:56,560 --> 00:35:01,740
exhibit, they show the part.
Past, but they also show what
539
00:35:01,740 --> 00:35:06,420
could the future look like? So
they really span ages, and
540
00:35:06,420 --> 00:35:11,520
according to your interest, you
go and you have a look. And so
541
00:35:11,760 --> 00:35:16,980
the algorithmic mirror. It's our
biases we already talked about
542
00:35:16,980 --> 00:35:20,900
that, our blind spots, our
hopes, our flaws, but also our
543
00:35:20,900 --> 00:35:26,600
values, because we said it's
about ethics. And if I could
544
00:35:26,600 --> 00:35:30,860
create such an exhibit, it
should be highly interactive. No
545
00:35:30,860 --> 00:35:34,100
matter who is in front of the
mirror, it acts differently,
546
00:35:34,100 --> 00:35:38,060
right? Oh, and instead of the
signature of Yeah, and instead
547
00:35:38,060 --> 00:35:42,340
of the signature of an artist,
there will be a quote that says,
548
00:35:42,340 --> 00:35:46,180
The future is not yet written.
You are the architect. How do
549
00:35:46,180 --> 00:35:46,960
you deal with it?
550
00:35:47,260 --> 00:35:53,500
Nikolaos Kaintantzis: Sounds
like matrix, no kind of so I was
551
00:35:53,500 --> 00:35:57,700
somehow just thinking and then
starting writing levels
552
00:35:57,700 --> 00:36:01,140
thinking. So I thought it has to
be something transparent,
553
00:36:01,440 --> 00:36:04,620
because we talk about ethics, it
should be transparent. And then
554
00:36:04,620 --> 00:36:07,860
I realized transparent has
different meanings in German and
555
00:36:07,860 --> 00:36:11,400
in English. So if in English,
something is transparent, you
556
00:36:11,400 --> 00:36:15,000
can look through it like a
glass. And that's the picture I
557
00:36:15,000 --> 00:36:17,400
had first in my mind. And then I
realized in German, it's
558
00:36:17,400 --> 00:36:22,160
something you look how it works,
so you see all the gears and
559
00:36:22,160 --> 00:36:24,980
everything there, because it's
transparent. You can see how the
560
00:36:24,980 --> 00:36:28,400
machine is working. It's not a
black box, it's you see
561
00:36:28,400 --> 00:36:30,860
everything, and that's
transparent. And then English is
562
00:36:30,860 --> 00:36:34,040
something I don't care. I just
look through and then I realize
563
00:36:34,400 --> 00:36:37,040
I would like to see the gears
somehow. But still, it's now
564
00:36:37,040 --> 00:36:42,280
some gears in the glass that's
now the art the exhibit I'm
565
00:36:42,280 --> 00:36:45,700
having there, and they're
working there, two little girls
566
00:36:45,700 --> 00:36:49,960
just working, working in a huge
glass. And the museum is
567
00:36:50,200 --> 00:36:54,460
everywhere. There are just
drones carrying this glass
568
00:36:54,820 --> 00:36:59,800
around our heads and flies
around. And hopefully it has
569
00:36:59,800 --> 00:37:04,200
enough ethics not to let it fall
down on my head. That was what I
570
00:37:04,200 --> 00:37:09,120
came with. So was a cool journey
in my head. At least, sounds
571
00:37:09,120 --> 00:37:13,920
good about to continue is what
to continue. Let's go to the
572
00:37:13,920 --> 00:37:19,980
story. Mark you started with the
story about the car and deciding
573
00:37:20,340 --> 00:37:24,200
who to kill or who to hurt more
if you have to do an accident
574
00:37:24,200 --> 00:37:28,280
you cannot break on time. I
would like to know over what you
575
00:37:28,280 --> 00:37:32,060
think AI should be allowed to
decide, and if so, who's
576
00:37:32,060 --> 00:37:33,860
accountable Mark, what do you
577
00:37:33,860 --> 00:37:37,460
Mark Richards: think so? Look,
it's all for me about the risk
578
00:37:37,460 --> 00:37:41,620
and the consequence of the wrong
decision in any decision. But
579
00:37:41,620 --> 00:37:44,860
then I think if I put an ethics
over on it, because that's like,
580
00:37:44,860 --> 00:37:49,000
that's the default answer if you
are using AI in some fashion for
581
00:37:49,300 --> 00:37:53,200
automation of a capability, or
automation of a decision or a
582
00:37:53,200 --> 00:37:56,920
workflow, or anything else, you
got to look at. Well, how much
583
00:37:56,920 --> 00:37:59,740
human control do I need? Is a
combination of the
584
00:37:59,740 --> 00:38:03,420
sophistication of my ability to
implement the AI and the risk
585
00:38:03,420 --> 00:38:06,720
level of something being made as
a wrong decision. Ethics just
586
00:38:06,720 --> 00:38:09,180
happens to be a branch, and I
think that's where it gets
587
00:38:09,180 --> 00:38:12,060
really interesting, because
again, it goes back to that
588
00:38:12,060 --> 00:38:16,980
awareness. And if you again, if
you look a little bit more
589
00:38:16,980 --> 00:38:19,920
seriously at the way people are
approaching engineering with AI
590
00:38:19,920 --> 00:38:23,540
these days, you'll hear a term
that you know the hardware guys
591
00:38:23,540 --> 00:38:26,540
love, and occasionally you hear
it in software, which is failure
592
00:38:26,540 --> 00:38:29,780
modes. And I still remember back
when I was an architect, I would
593
00:38:29,780 --> 00:38:32,660
occasionally have to write a
failure modes effect analysis.
594
00:38:33,200 --> 00:38:36,800
How might this thing fail? What
would be the impact? How might I
595
00:38:36,800 --> 00:38:41,800
get mitigate for it? Etc, etc,
etc. And everybody looking at
596
00:38:41,800 --> 00:38:46,780
engineering AI solutions
stresses failure modes analysis
597
00:38:47,560 --> 00:38:53,320
through a behavioral evaluation
perspective. And the challenge
598
00:38:53,320 --> 00:38:55,660
to you is think of the different
ways it might fail. It might
599
00:38:55,660 --> 00:38:58,180
fail because it gets the wrong
answer. It might fail because it
600
00:38:58,180 --> 00:39:00,960
uses too many tokens. It might
fail because it takes too long.
601
00:39:01,500 --> 00:39:04,980
Goodness knows. But I think the
question is, you ask yourself,
602
00:39:04,980 --> 00:39:08,700
is when I'm applying AI in a
situation, how might it fail
603
00:39:08,760 --> 00:39:11,940
ethically, and what am I going
to do to defend against that
604
00:39:12,000 --> 00:39:15,900
potential ethical failure? Now,
is it potentially that it's
605
00:39:15,900 --> 00:39:19,260
going to have bias? But let's
say I'm using I'm an HR team,
606
00:39:19,260 --> 00:39:22,520
and I'm using it as part of
screening resumes. When I
607
00:39:22,700 --> 00:39:27,140
advertise job listings, really
logical productivity application
608
00:39:27,140 --> 00:39:29,300
of it, well, what's the chance
that it's going to have bias
609
00:39:29,300 --> 00:39:32,180
about how it picks job listings?
It doesn't mean you can't use it
610
00:39:32,180 --> 00:39:35,600
to screen resumes, but it does
mean that when you think about
611
00:39:35,600 --> 00:39:39,620
using it to screen resumes,
you've got to consider the way
612
00:39:39,620 --> 00:39:43,240
in which bias or unethical
decisions might enter that
613
00:39:43,240 --> 00:39:46,300
decisioning process and be
explicit how you're going to
614
00:39:46,300 --> 00:39:50,020
actually validate that it hasn't
done so. And that's where I
615
00:39:50,020 --> 00:39:55,240
think the combo, for me is about
going, on the one hand, what's
616
00:39:55,240 --> 00:39:59,140
the risk associated with a wrong
decision, and recognizing that
617
00:39:59,140 --> 00:40:02,220
ethical risk is. Very much a
part of that potential risk. But
618
00:40:02,220 --> 00:40:06,120
the other is, what is my current
ability to safeguard against
619
00:40:06,120 --> 00:40:10,140
that risk? What sophistication
do I have in the way that I
620
00:40:10,140 --> 00:40:14,700
currently employ and deploy AI
to enable me to effectively
621
00:40:14,700 --> 00:40:17,880
manage it? And if my
sophistication is low and my
622
00:40:17,880 --> 00:40:21,780
risk is high, I probably
shouldn't be doing it at all.
623
00:40:22,560 --> 00:40:26,300
But if my sophistication grows,
then realistically, my risk
624
00:40:26,300 --> 00:40:29,720
becomes more effectively
controlled. So you know, who's
625
00:40:29,720 --> 00:40:31,700
accountable? Well, at the end of
the day, the company's
626
00:40:31,700 --> 00:40:34,700
accountable. They're taking a
stance, and it's a risk posture,
627
00:40:36,140 --> 00:40:39,680
but it's not black and white.
It's about awareness, asking
628
00:40:39,680 --> 00:40:42,700
yourself questions, and
evaluating the trade offs, and
629
00:40:42,700 --> 00:40:46,000
every discussion I have with
anybody in business about
630
00:40:46,000 --> 00:40:51,220
responsible AI full stop, right?
Is you are choosing a risk
631
00:40:51,220 --> 00:40:54,700
posture. Make sure you're making
an informed choice.
632
00:40:56,200 --> 00:40:59,740
Stephan Neck: And this mark will
be related for me to what are
633
00:40:59,740 --> 00:41:04,440
the stakes? Right? Is it? Do I
have low stakes? I would say,
634
00:41:04,440 --> 00:41:09,240
Okay, go for it. Help me, right?
You have high autonomy. My
635
00:41:09,240 --> 00:41:13,260
stakes are low, so I probably
have low risks in it. The
636
00:41:13,260 --> 00:41:18,000
question is, when it comes to a
medium stake, like we talked
637
00:41:18,000 --> 00:41:23,180
about PILOTs and CO pilots. The
last episode, right? You're a
638
00:41:23,180 --> 00:41:29,900
pilot in a plane. Your plane is
full of AI helping you. There's
639
00:41:30,560 --> 00:41:36,500
an automatic navigation system.
There's GPS whatever. There is
640
00:41:36,500 --> 00:41:42,280
kind of a co working space
where, based on my stakes, I
641
00:41:42,280 --> 00:41:48,280
have to decide, where do I
interact? Where do I set my
642
00:41:48,280 --> 00:41:53,500
stakes? And say, now I decide.
And the other one where I have a
643
00:41:53,500 --> 00:41:56,860
high stake, let's think about
I'm in a hospital in palliative
644
00:41:56,920 --> 00:42:01,620
care. I don't want the machine
to decide if I live or not, it's
645
00:42:01,620 --> 00:42:06,360
still what I've written down on
a sheet of paper as a patient
646
00:42:06,480 --> 00:42:10,980
that someone, for me, will
decide if I'm out of mind, if
647
00:42:10,980 --> 00:42:15,060
I'm not able to do that. So for
me, it's about stakes, and as
648
00:42:15,060 --> 00:42:19,080
you said, Mark, yeah, in the
end, those who create those
649
00:42:19,080 --> 00:42:24,980
systems and are using AI, they
have the primary responsibility
650
00:42:24,980 --> 00:42:27,920
also for the future, as you
mentioned, Niko, for the future
651
00:42:27,920 --> 00:42:32,000
generations. And again, this
leads me back to, as you said,
652
00:42:32,000 --> 00:42:36,140
Mark fitness for use, right? You
have an architecture that is fit
653
00:42:36,140 --> 00:42:42,100
for use, and on top comes
regulatory aspects, government,
654
00:42:42,280 --> 00:42:46,900
aspects that have to be baked in
as well, which is part of
655
00:42:46,900 --> 00:42:50,860
ethics, where there are certain
values behind it. Regulator
656
00:42:50,860 --> 00:42:56,500
tries to, tries to protect the
majority of the people against
657
00:42:56,500 --> 00:43:01,860
harm, and how you deal with
certain risks. So, yeah, there's
658
00:43:01,860 --> 00:43:05,400
a there's a huge range of
stakes, and based on the
659
00:43:05,400 --> 00:43:09,360
context, again, based on the
stakes, I would answer the
660
00:43:09,360 --> 00:43:13,500
question, who's responsible,
who's dealing with it, and how
661
00:43:13,500 --> 00:43:14,100
you deal with
662
00:43:15,000 --> 00:43:16,620
Nikolaos Kaintantzis: it? Let's
make the practical. I like the
663
00:43:16,620 --> 00:43:21,860
discussion. We call the episode
unleashed agents. Unleashed
664
00:43:22,520 --> 00:43:26,120
change, unleashed whatever, but
unleashed. What should we
665
00:43:26,300 --> 00:43:31,280
unleash to make this practical,
or let's ask differently. What
666
00:43:31,340 --> 00:43:34,700
cultural shifts do we need to be
prepared for this
667
00:43:34,700 --> 00:43:39,320
accountability, for ethics to be
part of our daily lives? So what
668
00:43:39,320 --> 00:43:45,100
can we do? What can our
spectaculars do? Our listeners
669
00:43:45,760 --> 00:43:47,260
do? Destroy the
670
00:43:53,980 --> 00:43:57,880
Stephan Neck: whole flow. Niko
as a moderator, is fully
671
00:43:57,880 --> 00:44:02,520
unleashed, and we're following,
okay, so let me throw in two
672
00:44:02,520 --> 00:44:06,060
things that are still the same.
It shouldn't change the culture
673
00:44:06,060 --> 00:44:09,300
that we have. And if we say
culture is the sum of our
674
00:44:09,300 --> 00:44:14,040
behavior, we still should build
carefully. And if things are
675
00:44:14,040 --> 00:44:16,800
broken, you have to fix it.
That's how I was brought up,
676
00:44:16,800 --> 00:44:22,280
right? You fix it. If you break
it or you clean up. Second one
677
00:44:22,280 --> 00:44:26,180
we talked about it is, data is
still a liability, right? You
678
00:44:26,180 --> 00:44:32,120
leave traces, you give your
data, and now what changes
679
00:44:32,120 --> 00:44:38,300
drastically, and I as an old
dog, compared to mark son, who's
680
00:44:38,360 --> 00:44:42,820
a digital native, right? What is
a the question for me is, what
681
00:44:42,820 --> 00:44:46,300
is the digital digital
citizenship? We are all
682
00:44:46,300 --> 00:44:51,160
citizens. We all belong to a
certain group, to a client, to a
683
00:44:51,160 --> 00:44:56,620
city, to people. We have
different languages. So how does
684
00:44:56,620 --> 00:44:59,500
that evolve? That would be one
question that's for sure, a
685
00:44:59,500 --> 00:45:07,500
digital. Little shift. And the
other one is, yeah, still, I
686
00:45:07,500 --> 00:45:12,540
think that should be the same,
like in the old days. Nowadays.
687
00:45:12,540 --> 00:45:16,980
It's healthy skepticism. There's
always an option. There's new
688
00:45:16,980 --> 00:45:23,660
stuff to use, but, yeah, staying
skeptical, skeptic a little bit.
689
00:45:24,860 --> 00:45:29,060
Don't just jump into the pool.
It might not be deep enough, and
690
00:45:29,060 --> 00:45:29,720
you will hit your
691
00:45:30,620 --> 00:45:32,360
Nikolaos Kaintantzis: head. I
lacked your positivity. How you
692
00:45:32,360 --> 00:45:35,060
started and say, we are doing
this anyway. You cannot change
693
00:45:35,060 --> 00:45:37,940
the culture. We are in an
environment where we'll care for
694
00:45:37,940 --> 00:45:41,860
it. You just have to continue
caring. I love, I love this part
695
00:45:41,860 --> 00:45:45,700
of the answer from for me, I
think what we have to change or
696
00:45:45,700 --> 00:45:50,860
do differently is I still think
leadership management thinks the
697
00:45:50,860 --> 00:45:54,280
whole AI thing is an IT thing,
and ethics should be a
698
00:45:54,280 --> 00:45:59,560
compliance issue. So let's move
it to a compliance department
699
00:45:59,620 --> 00:46:03,060
and or HR department, if there
are some other rules there and
700
00:46:03,060 --> 00:46:06,720
that will solve the problem. And
as you heard before, I'm out,
701
00:46:06,720 --> 00:46:10,920
and I'm now the sad guy this
episode, the warning guy, I
702
00:46:10,920 --> 00:46:14,940
think, no, not. I think, I know
it's a responsibility of the
703
00:46:14,940 --> 00:46:20,280
whole society. So we need, we
need models who are traceable.
704
00:46:20,280 --> 00:46:24,680
We need to know where data come
from. We need to know what is
705
00:46:24,680 --> 00:46:28,940
fake or not. Is it generated by
AI or not? So I'm deeply
706
00:46:28,940 --> 00:46:33,020
concerned seeing pictures that
look like original pictures from
707
00:46:33,020 --> 00:46:36,860
the past, and now we are faking
and changing history, and I
708
00:46:36,860 --> 00:46:42,220
think this is not continue as we
are doing now. It's really being
709
00:46:42,220 --> 00:46:45,340
aware that the society issue and
the society have to do
710
00:46:45,340 --> 00:46:48,640
something, and it's not just the
IT department or department or
711
00:46:48,640 --> 00:46:52,660
compliance department. So I
think we need not a culture
712
00:46:52,660 --> 00:46:55,540
shift, but we need a shift in
thinking it's our all
713
00:46:55,540 --> 00:46:59,920
responsibility, and not the tech
moguls that will take care for
714
00:46:59,920 --> 00:47:02,160
us. They will not at the moment,
what I see,
715
00:47:02,520 --> 00:47:05,460
Mark Richards: I'm gonna go
slightly different way, just for
716
00:47:05,460 --> 00:47:10,260
change. Have either of you seen
Mad Men in the streets
717
00:47:10,260 --> 00:47:15,840
everywhere? But there's a movie.
It was a TV series, and it was
718
00:47:15,840 --> 00:47:20,720
set in advertising in the world
of the 60s, and fundamentally,
719
00:47:20,720 --> 00:47:23,240
it told stories about the
craziness of the world of the
720
00:47:23,240 --> 00:47:27,020
60s through the lens of somebody
in advertising. So of course,
721
00:47:27,020 --> 00:47:31,280
this is still the days when
cigarettes paid for an awful lot
722
00:47:31,280 --> 00:47:35,840
of advertising. So big tobacco
was was a part of the game. It
723
00:47:35,840 --> 00:47:39,140
was a world where women's place
in the office was very
724
00:47:39,140 --> 00:47:44,140
different. And I watched it with
my son two or three years ago
725
00:47:44,140 --> 00:47:49,000
now, and we sat there every
episode, and I would sit going,
726
00:47:49,420 --> 00:47:52,060
like, I know the workplace has
changed a lot since I was young,
727
00:47:52,120 --> 00:47:56,380
but it's a startling to see how
it was then. And and he would
728
00:47:56,380 --> 00:47:59,440
look at it and go, Dad, is that
actually possible? Like, is this
729
00:47:59,440 --> 00:48:02,760
real? Surely they've got to be
inventing this because the way
730
00:48:02,760 --> 00:48:07,020
they're treating different types
of people and and you look at
731
00:48:07,500 --> 00:48:11,340
the way our workplace has
changed in our lifetimes, I know
732
00:48:11,340 --> 00:48:14,100
when I started working, nobody
got training about sexual
733
00:48:14,100 --> 00:48:18,480
harassment or bullying, talk
about privacy and security and
734
00:48:18,480 --> 00:48:22,040
personal information. All the
things that we have learned in
735
00:48:22,040 --> 00:48:25,400
our businesses are important to
establish expectations for with
736
00:48:25,400 --> 00:48:29,480
people and to teach them about
because we needed to change our
737
00:48:29,480 --> 00:48:33,500
workplace, right? And a lot of
those changes were really about
738
00:48:33,500 --> 00:48:36,320
changing our workplace to be
more inclusive for everybody who
739
00:48:36,320 --> 00:48:40,040
was a part of it. And you know,
they took years, because I would
740
00:48:40,040 --> 00:48:43,240
periodically go, hey, the life
of women in the corporate world
741
00:48:43,240 --> 00:48:46,240
is much better now than it used
to be. And then I would coach
742
00:48:46,240 --> 00:48:50,140
women who would go, No, it's not
despite years and years and
743
00:48:50,140 --> 00:48:53,260
years of mindfulness and
training and awareness campaigns
744
00:48:53,260 --> 00:48:57,280
and all the rest of it. And I'm
not going to suggest that AI
745
00:48:57,280 --> 00:48:59,500
agents being part of the
workplace is anything like the
746
00:48:59,500 --> 00:49:02,880
issues we had with equality in
the workplace, but I'm going to
747
00:49:02,880 --> 00:49:05,100
go it's really disruptive,
because now we're going to have
748
00:49:05,100 --> 00:49:09,000
a workplace that's got people,
humans and agents, and it's
749
00:49:09,240 --> 00:49:12,180
going to have a whole bunch of
expectations raised about
750
00:49:12,180 --> 00:49:16,200
awareness, whether it's
awareness of agents for ethics
751
00:49:16,680 --> 00:49:19,920
or awareness of people for the
way in which ethical concerns
752
00:49:19,920 --> 00:49:24,500
arise In the world of AI, it's
going to take years of real
753
00:49:24,740 --> 00:49:31,580
education, mindfulness and clear
action, because I know, I don't
754
00:49:31,580 --> 00:49:34,280
know how many times I've been
through sexual harassment
755
00:49:34,280 --> 00:49:36,860
training to make sure that I'm
aware of it, but every now and
756
00:49:36,860 --> 00:49:39,620
again, I'll see something in it
while I'm doing it for the 20th
757
00:49:39,620 --> 00:49:43,420
Time at a different company, and
I go, Oh, that's a really
758
00:49:43,420 --> 00:49:45,760
interesting scenario. I had
never thought of that as
759
00:49:45,760 --> 00:49:49,480
potential bullying or
harassment, and it's actually
760
00:49:49,480 --> 00:49:52,240
awakening me to the possibility.
And I think right now for a lot
761
00:49:52,240 --> 00:49:54,940
of people, it's about just being
awakened to the possibility. But
762
00:49:54,940 --> 00:49:56,860
Nikolaos Kaintantzis: you said
we needed years, and still, the
763
00:49:56,860 --> 00:50:01,320
situation isn't better. Now we
are much more faster. So what
764
00:50:01,320 --> 00:50:02,580
can we do different this time?
765
00:50:02,580 --> 00:50:04,140
Mark Richards: Well, that's,
that's the really interesting
766
00:50:04,140 --> 00:50:07,500
question about it, right? I
mean, there are wells in which
767
00:50:07,800 --> 00:50:11,040
you know the fundamental deal
with AI is, the more that you
768
00:50:11,040 --> 00:50:13,740
try and make your company AI
powered, the more you organize
769
00:50:13,740 --> 00:50:16,860
your information. And as part of
organizing your information, you
770
00:50:16,860 --> 00:50:19,140
find things that need to be
codified, and whether that's
771
00:50:19,140 --> 00:50:23,780
coding standards or values or
ethical principles. You know,
772
00:50:23,780 --> 00:50:26,300
there's all kinds of things that
need to be codified. Once you
773
00:50:26,300 --> 00:50:30,680
get them codified, you can get
agents to scan for them. I
774
00:50:30,680 --> 00:50:34,220
could, for example, have some
kind of ethical behaviors,
775
00:50:34,520 --> 00:50:38,120
ethical interactions. Agent that
scans the transcript of every
776
00:50:38,120 --> 00:50:42,400
meeting that's held on teams
looking for potential ethical
777
00:50:42,400 --> 00:50:46,480
violations and raising red flags
to go it seems like we're seeing
778
00:50:46,480 --> 00:50:49,420
patterns of a particular type of
ethical violation. So I can get
779
00:50:49,480 --> 00:50:52,840
AI to help me to monitor it and
be aware of it, and design
780
00:50:52,840 --> 00:50:55,900
action and response if that
happens to be something I
781
00:50:55,900 --> 00:50:58,900
choose. But of course, to choose
that, I have to also look at the
782
00:50:58,900 --> 00:51:01,620
ethical dilemma of allowing an
agent to scan every interaction
783
00:51:01,620 --> 00:51:06,720
they've had on teams between
people for analysis. So you can
784
00:51:06,720 --> 00:51:10,440
develop a system that starts to
develop a certain amount of self
785
00:51:10,440 --> 00:51:15,060
monitoring and self policing
there, but then it even just
786
00:51:15,060 --> 00:51:18,060
raises the potential complexity
even more, because what if your
787
00:51:18,300 --> 00:51:21,200
ethical self monitoring agents
start to monitor for the wrong
788
00:51:21,200 --> 00:51:23,960
type of behavior, because, like,
if you're using Croc, there's a
789
00:51:23,960 --> 00:51:24,800
fair chance of it,
790
00:51:27,920 --> 00:51:31,100
Stephan Neck: but doesn't it
start with how you are brought
791
00:51:31,100 --> 00:51:34,820
up. Because if I look at my
kids, if I look at my
792
00:51:34,820 --> 00:51:40,220
granddaughter, how do I
exemplify what good behavior is?
793
00:51:40,760 --> 00:51:45,880
So they can follow, they can
learn and and, yes, the future
794
00:51:45,880 --> 00:51:49,480
will be totally different after
I'm gone, but it's about
795
00:51:49,480 --> 00:51:56,020
behavior of human beings. And
does AI help? Yes, of course,
796
00:51:56,260 --> 00:52:02,220
does it maybe enforce? Yes,
let's, let's build it in. But,
797
00:52:02,220 --> 00:52:06,840
but the question will be, how do
we create an environment,
798
00:52:06,960 --> 00:52:10,920
starting in a family, in
partnerships, going through our
799
00:52:10,920 --> 00:52:14,880
workspaces, as you mentioned,
Mark, where we where we are kind
800
00:52:14,880 --> 00:52:19,140
to each other, where we treat
each other like equals, like
801
00:52:19,140 --> 00:52:22,940
human beings. Because this is,
this is the behavior I would
802
00:52:22,940 --> 00:52:27,740
like to see, the degree of
support by AI. It depends right
803
00:52:28,220 --> 00:52:31,160
again, it depends on the
context, it depends on the
804
00:52:31,160 --> 00:52:35,240
stakes we have. But just be kind
and do the good job.
805
00:52:35,480 --> 00:52:38,240
Nikolaos Kaintantzis: Great. I
would like to close with much
806
00:52:38,240 --> 00:52:41,020
more open questions than I
prepared. What can we do
807
00:52:41,020 --> 00:52:44,980
differently? How can this look
concrete in practice? We still
808
00:52:44,980 --> 00:52:49,900
called unleashed agents
unleashed. How can successful AI
809
00:52:49,900 --> 00:52:52,960
accountability look like in
practice? You said already some
810
00:52:52,960 --> 00:52:56,020
parts of it, I know, yeah,
811
00:52:56,080 --> 00:52:58,780
Mark Richards: and look, I'd
really stay with if I boil it
812
00:52:58,780 --> 00:53:01,860
all back down. And you know,
it's like a broken record
813
00:53:01,860 --> 00:53:06,720
mindfulness. You've got to start
having the conversations. And I
814
00:53:06,720 --> 00:53:09,540
think it's really important you
start having the conversations
815
00:53:09,540 --> 00:53:13,860
in a way that does not scare
people into not engaging. You
816
00:53:13,860 --> 00:53:16,680
know, there's a there's a world
which is there are people there.
817
00:53:16,680 --> 00:53:19,200
Some people are excited about
AI, some people are scared about
818
00:53:19,200 --> 00:53:22,340
AI, some people who are worried
that it's going to take their
819
00:53:22,340 --> 00:53:24,920
job. Some people are worried
it's going to take over the
820
00:53:24,920 --> 00:53:27,920
world. Like there's all kinds of
attitudes of people towards AI.
821
00:53:27,920 --> 00:53:30,740
Then there's people
understanding how AI might help
822
00:53:30,740 --> 00:53:34,160
them in different types of jobs.
But then there's also, what are
823
00:53:34,160 --> 00:53:37,520
the risks of AI? And I think
there's a risk, and this was
824
00:53:37,520 --> 00:53:42,640
kind of, I'll go there right for
at least a year, the surface for
825
00:53:42,640 --> 00:53:48,280
AI in safe was responsible AI,
so the whole surface that safe
826
00:53:48,280 --> 00:53:50,680
happened, obviously, they've
changed a lot over the last six
827
00:53:50,680 --> 00:53:54,700
to eight months, but for a good
while, if you went AI in safe,
828
00:53:54,760 --> 00:53:57,640
all you saw was responsible AI,
which is fundamentally, how do
829
00:53:57,640 --> 00:54:02,640
we govern AI and ethical AI,
right? And it was missing the
830
00:54:02,640 --> 00:54:07,260
what's exciting about AI, you
just read this long laundry list
831
00:54:07,260 --> 00:54:09,540
of, oh my goodness, these are
all the corporate governance
832
00:54:09,540 --> 00:54:12,240
measures you can have to put in
place before you can responsibly
833
00:54:12,240 --> 00:54:16,860
use this new thing. And that
overwhelms and it drowns
834
00:54:17,040 --> 00:54:20,660
evolution where, for me, I
think, just as you're for most
835
00:54:20,660 --> 00:54:24,080
organizations, you're entering a
mindfulness journey to say, how
836
00:54:24,080 --> 00:54:27,920
can we apply AI in our
organization in a meaningful way
837
00:54:27,920 --> 00:54:30,800
where we get return for our
investment? I think we've got to
838
00:54:30,800 --> 00:54:33,200
have a parallel conversation
that also says, Well, how can we
839
00:54:33,200 --> 00:54:36,560
do that ethically? What are the
implications, and how are we
840
00:54:36,560 --> 00:54:41,800
starting to evolve and take on
an organizational stance in
841
00:54:41,800 --> 00:54:45,580
terms of our belief about AI and
ethics and let it, let it
842
00:54:45,580 --> 00:54:45,940
unfold.
843
00:54:46,300 --> 00:54:47,980
Nikolaos Kaintantzis: Sounds
like you're advocating becoming
844
00:54:47,980 --> 00:54:50,620
teenagers again, being
responsive, responsible,
845
00:54:50,620 --> 00:54:54,100
responsible, responsible, but
sometimes little bit stretch of
846
00:54:54,100 --> 00:54:56,620
boundaries, and then say Sorry,
sorry, sorry, and do it the next
847
00:54:56,620 --> 00:55:00,840
time better and learn and try.
Sounds great. I. I had many,
848
00:55:00,840 --> 00:55:03,480
many bingo, so I listened to
you. For me, it's really get
849
00:55:03,480 --> 00:55:06,720
people involved who are
affected, like mindfulness, you
850
00:55:06,720 --> 00:55:10,500
said before, and build in these
correlations in the code that we
851
00:55:10,500 --> 00:55:14,760
can be sure, yeah, that we have
ethics, yeah, those who are
852
00:55:14,760 --> 00:55:18,840
affected, and some kind of what
harm suggested in beginning to
853
00:55:18,840 --> 00:55:22,460
use AI to enforce this ethics.
Stephan, you have to add
854
00:55:22,460 --> 00:55:24,500
something. You said some things
about about this,
855
00:55:25,160 --> 00:55:29,900
Stephan Neck: yeah, just suing
what we just discussed. People
856
00:55:29,900 --> 00:55:36,020
will redress with AI, how do we
empower them? Organizations,
857
00:55:36,080 --> 00:55:40,340
corporations are responsible for
what they built, how they deploy
858
00:55:40,340 --> 00:55:45,400
it, how it's used. So this is
this accountability. The third
859
00:55:45,400 --> 00:55:51,640
level, where I'm a bit hesitant,
is government. Governments, for
860
00:55:51,640 --> 00:55:55,240
me, at the moment, worldwide,
are not doing a good job when it
861
00:55:55,240 --> 00:55:58,960
comes to this item. We just had
an election in Switzerland about
862
00:55:58,960 --> 00:56:04,440
an electronic ID, almost 50% of
the people are not yet
863
00:56:04,440 --> 00:56:09,360
convinced. But hey, now it's
imposed. It's we have to do it.
864
00:56:09,360 --> 00:56:14,160
So how do we deal with that? And
how do we make it really
865
00:56:14,160 --> 00:56:18,540
successful for human beings on
all different levels?
866
00:56:19,260 --> 00:56:22,280
Nikolaos Kaintantzis: Just one
takeaways, my friends, just one
867
00:56:22,280 --> 00:56:25,640
takeaway. Which one would it be?
Wrapping up, wrapping up the
868
00:56:25,640 --> 00:56:26,240
episodes
869
00:56:26,600 --> 00:56:29,660
Mark Richards: I'll jump in. I'm
going to steal from Mark Twain.
870
00:56:30,020 --> 00:56:32,420
It ain't what you know don't
know that gets you into trouble.
871
00:56:32,420 --> 00:56:35,600
It's what you know for sure that
just don't soak so my question
872
00:56:35,600 --> 00:56:38,480
to you is, what do you know
about AI and ethics?
873
00:56:39,140 --> 00:56:43,840
Stephan Neck: Wow, yeah, that's
that's a strong one mark. So I'm
874
00:56:43,840 --> 00:56:47,140
still looking into the mirror.
AI is reflecting the best and
875
00:56:47,140 --> 00:56:51,460
the worst of our own humanity,
forcing us to decide which
876
00:56:51,460 --> 00:56:56,200
version of ourselves we want to
become we want to encode into
877
00:56:56,260 --> 00:56:56,800
the future.
878
00:56:57,160 --> 00:56:59,800
Nikolaos Kaintantzis: Oh, that's
so great. So some mind now loses
879
00:56:59,800 --> 00:57:04,200
up. My mind sounds like it's
responsibility of our society,
880
00:57:04,260 --> 00:57:07,980
of us as parents, us as
neighbors, us as humans. So
881
00:57:07,980 --> 00:57:12,180
ethics is a socio political
responsibility. And with that, I
882
00:57:12,240 --> 00:57:16,140
give you to the hands of Mark,
and he can wrap up and tell you
883
00:57:16,140 --> 00:57:20,480
what's next week you want to
mute somehow mark on my side, at
884
00:57:20,480 --> 00:57:20,960
least,
885
00:57:21,080 --> 00:57:25,160
Mark Richards: I had to mute
myself to now I'm back and some
886
00:57:25,160 --> 00:57:30,140
post production editing. Next
week, I think will be the first
887
00:57:30,140 --> 00:57:32,960
of a few episodes where we're
going to look at different roles
888
00:57:33,980 --> 00:57:36,560
and talk about and I think for
me, this is kind of inspiring,
889
00:57:36,560 --> 00:57:38,540
because it's what I'm finding a
lot of my coaching
890
00:57:38,540 --> 00:57:42,820
conversations. Is that beginning
of saying to someone, how might
891
00:57:42,820 --> 00:57:45,880
AI be relevant to you and your
job, right? What are the
892
00:57:45,880 --> 00:57:49,480
possible places you could use
it? And, you know, we kicked off
893
00:57:49,480 --> 00:57:52,780
season three with a look at AI
and user stories, and how I
894
00:57:52,780 --> 00:57:55,960
might be disrupting user
stories. We got a bit broader
895
00:57:55,960 --> 00:57:58,300
next week, and we're going to
talk about AI and the product
896
00:57:58,300 --> 00:58:01,860
management role. So looking
forward to that one, and we will
897
00:58:01,860 --> 00:58:05,640
be in stephan's capable hands.
In the meantime, thank you for
898
00:58:05,640 --> 00:58:09,780
joining us, and hopefully you
don't have too many ethical
899
00:58:09,780 --> 00:58:11,640
dilemmas between now and next
week.