Will AI chatbots replace physicists?

1
00:00:04,790 --> 00:00:05,570
Physics.
2
00:00:05,570 --> 00:00:08,280
World. Hello and welcome to the
Physics World Stories Podcast.
3
00:00:08,780 --> 00:00:11,520
I'm Andrew G. Lester, and I'm
suffering a little bit with hay fever.
4
00:00:11,980 --> 00:00:16,840
So I'll try and keep my voice to a minimum
as we explore the influence of chat
5
00:00:16,960 --> 00:00:20,840
G p T and other AI bots on
the lives of physicists.
6
00:00:21,490 --> 00:00:26,440
We'll focus on how physics education may
change as a response to these chatbots,
7
00:00:26,940 --> 00:00:30,320
but we'll also look at how physics
as a discipline may change.
8
00:00:30,510 --> 00:00:32,840
What are the skills that
future physicists will need,
9
00:00:33,180 --> 00:00:36,680
and what are the tasks that may be
lost to artificial intelligence?
10
00:00:37,250 --> 00:00:41,520
Later we'll hear from a student whose
life as a physicist is really just
11
00:00:41,520 --> 00:00:46,240
beginning as the world around us changes
as a result of artificial intelligence
12
00:00:46,270 --> 00:00:49,240
breakthroughs. But first,
we'll hear from Philip Moti.
13
00:00:49,740 --> 00:00:51,320
But for this conversation to make sense,
14
00:00:51,720 --> 00:00:55,360
I do need to mention that I'm also a
lecturer in sciences communication.
15
00:00:55,740 --> 00:00:57,560
Here's Phil Marriott.
16
00:00:57,620 --> 00:01:00,920
I'm a physicist at University
of Nottingham. I've been
here a very long time.
17
00:01:01,280 --> 00:01:04,280
Actually next year I'll be
in Nottingham for 30 years.
18
00:01:05,220 --> 00:01:08,760
So as well as the lecturing side of
things, which I really do enjoy there,
19
00:01:08,850 --> 00:01:12,960
there are research focuses
on single atom manipulation,
20
00:01:12,960 --> 00:01:15,920
single molecule manipulation
with scanning, probe microscopes,
21
00:01:16,230 --> 00:01:19,760
looking at the forces
between atoms. Ultimately,
22
00:01:20,310 --> 00:01:25,040
what I sometimes describe it as though
this is a little bit more applied than
23
00:01:25,040 --> 00:01:27,800
really what we're going for, it's
really quite fundamental. Researches.
24
00:01:28,020 --> 00:01:32,160
Can we do 3D printing with atoms? So
can we build structures up atom by atom?
25
00:01:32,580 --> 00:01:35,480
So there's a lot of interest as well
within the group and machine learning
26
00:01:35,630 --> 00:01:38,000
methods. Um, so chat,
27
00:01:38,120 --> 00:01:43,080
G p t and g p t four and things
like that are certainly of
28
00:01:43,280 --> 00:01:48,120
interest. The the problem is
with G P T four, it's, um,
29
00:01:48,430 --> 00:01:53,160
like the success of the chat G P T,
but they've been a great deal more, um,
30
00:01:54,420 --> 00:01:56,880
uh, how can I put it?
Closed door about it.
31
00:01:57,020 --> 00:02:01,880
So in terms of how it's been
trained, in terms of just what, uh,
32
00:02:01,980 --> 00:02:06,000
you know, parameter space, there
is just what types of, of network.
33
00:02:07,140 --> 00:02:10,560
Um, it's obviously one of these
pre generative transformer models,
34
00:02:10,770 --> 00:02:15,720
which is where the G P T
thing comes from. Um, uh,
35
00:02:16,380 --> 00:02:20,560
but in terms of beyond that, they're
getting a little bit more, um,
36
00:02:20,670 --> 00:02:21,840
proprietary about things.
37
00:02:21,840 --> 00:02:24,920
So it's actually quite difficult to
tell what's going on. It's interesting,
38
00:02:24,930 --> 00:02:29,760
we've evolved to the point where our
artificial neural networks are just
39
00:02:29,780 --> 00:02:33,640
as complicated and impossible
to understand as the
organic neural networks in
40
00:02:33,640 --> 00:02:35,400
our heads. So in terms of all the layers.
41
00:02:35,640 --> 00:02:38,280
I wanted to know why AI
bots such as chat G P T,
42
00:02:38,280 --> 00:02:40,080
are significant from
a physics perspective.
43
00:02:40,580 --> 00:02:42,320
How can they help
physicists in their work?
44
00:02:42,700 --> 00:02:45,960
But I also wanted to get to the heart
of where some of the fears come from.
45
00:02:46,710 --> 00:02:49,600
They, the interesting thing
from physics perspective, it,
46
00:02:49,870 --> 00:02:52,480
it's now capable of interpreting images.
47
00:02:53,180 --> 00:02:58,120
Its spatial reasoning seems to be a heck
of a lot better than shot G P T. Um,
48
00:02:58,380 --> 00:03:03,280
and it's one paper I've just
pulled off the archive, um,
49
00:03:03,430 --> 00:03:05,880
said that has has put uh,
50
00:03:06,000 --> 00:03:10,080
G P T four through its paces on
something called the force concept
51
00:03:10,430 --> 00:03:11,263
inventory,
52
00:03:11,330 --> 00:03:16,200
which is generally if you want to
test the efficacy of different, um,
53
00:03:16,200 --> 00:03:19,680
undergraduate courses, particularly
first year courses, Newton's laws,
54
00:03:19,890 --> 00:03:20,880
kinematics, et cetera,
55
00:03:21,380 --> 00:03:24,720
you give them this F c I force
concept inventory at the start,
56
00:03:25,140 --> 00:03:28,600
you give them at the end and you hope
that you've actually had a positive
57
00:03:28,600 --> 00:03:33,000
influence on them. Chat G P T got
28, sorry, not chat. G P T chat.
58
00:03:33,120 --> 00:03:37,320
G P T actually got about
a 50% score g PT four.
59
00:03:37,650 --> 00:03:40,200
First time it saw that, uh, inventory,
60
00:03:40,510 --> 00:03:45,200
that set of questions got 28 outta
30, which is rather scary .
61
00:03:46,760 --> 00:03:48,260
So it can increasingly do physics.
62
00:03:48,660 --> 00:03:49,620
. But why is that scary?
63
00:03:49,880 --> 00:03:52,580
Uh, it's interesting, um, in terms of
64
00:03:54,250 --> 00:03:56,420
it's both scary and very impressive,
65
00:03:56,500 --> 00:04:00,260
I guess scary almost in an impressive
sense. I think we've got a,
66
00:04:00,260 --> 00:04:02,580
there's an awful lot of hand
ringing and there's an awful lot of,
67
00:04:02,830 --> 00:04:05,940
let's let's try and, um,
shut this down mixture.
68
00:04:05,940 --> 00:04:09,820
Students don't get access to it, let's
ban it. That's obviously not gonna work.
69
00:04:10,050 --> 00:04:12,300
It's a really powerful tool,
70
00:04:13,320 --> 00:04:17,220
but we have to learn
how to exploit it. Um,
71
00:04:17,220 --> 00:04:21,900
there's a broader question as well in
that, you know, even if we do ban it,
72
00:04:22,240 --> 00:04:26,860
what's the point in sending students
out into the big bad world with skills
73
00:04:27,130 --> 00:04:30,980
that chat that that G P
T for and its successors,
74
00:04:31,760 --> 00:04:32,980
you know, can just,
75
00:04:33,090 --> 00:04:37,140
they can do it in a fraction of the
time a student could do it or a graduate
76
00:04:37,140 --> 00:04:42,100
could do it. We have to be
giving them skillsets that
77
00:04:42,750 --> 00:04:46,820
compliment what an AI can do
otherwise, what's the point of us?
78
00:04:47,530 --> 00:04:50,580
What is the point of us,
you know, more tangibly,
79
00:04:50,650 --> 00:04:54,940
what do physicists now who are,
you know, going through education,
80
00:04:54,940 --> 00:04:55,980
what should they be?
81
00:04:56,090 --> 00:05:00,700
What skills should they be learning so
that they're still needed in 2040 or
82
00:05:00,920 --> 00:05:01,660
beyond?
83
00:05:01,660 --> 00:05:02,160
I think the,
84
00:05:02,160 --> 00:05:06,860
the fun most fundamental
skill we can give them is not
85
00:05:07,120 --> 00:05:12,060
how to, you know, uh, think about
and write down a Hamiltonian.
86
00:05:12,060 --> 00:05:15,500
It's not to think about how they're gonna
do some problem in quantum mechanics
87
00:05:15,500 --> 00:05:18,540
or classical fields or how to
do one bit of Dave gra or curl.
88
00:05:18,650 --> 00:05:21,450
It's how to educate themselves. That's,
89
00:05:21,510 --> 00:05:23,850
that's the skill we really
should be teaching them.
90
00:05:23,850 --> 00:05:28,250
That's to have that
independence, self-reliance,
91
00:05:28,450 --> 00:05:32,010
resilience to be able
to teach themselves. Um,
92
00:05:32,160 --> 00:05:34,970
obviously we have to do that
within the context of physics.
93
00:05:35,020 --> 00:05:39,410
We're teaching physics,
but the, the thing is,
94
00:05:39,470 --> 00:05:43,250
if we're playing catch up really with
the technology and we're gonna lose,
95
00:05:43,630 --> 00:05:46,330
we really are gonna lose.
Cuz it's, it's developing.
96
00:05:47,010 --> 00:05:49,810
I know exponentially is a word that's
thrown around a return that's thrown
97
00:05:49,810 --> 00:05:50,450
around a great deal,
98
00:05:50,450 --> 00:05:54,130
but I think it's pretty clear
it's developing exponentially and
99
00:05:55,700 --> 00:06:00,080
yeah, I think we radically need to
change what we're doing. You know,
100
00:06:00,080 --> 00:06:04,680
the old style exam question of
please regurgitate pages 42 to
101
00:06:04,680 --> 00:06:08,800
47 of your notes, this derivation,
that doesn't work anymore.
102
00:06:08,980 --> 00:06:13,960
And it didn't work when I was a
undergraduate or any of us were an
103
00:06:13,960 --> 00:06:14,480
undergraduate.
104
00:06:14,480 --> 00:06:19,320
I remember clearly in third
year being able to do all
105
00:06:19,340 --> 00:06:23,040
the equations in terms of
minority and ma majority, um,
106
00:06:23,230 --> 00:06:26,680
carrier occurrence in a
P junction. You asked me,
107
00:06:26,780 --> 00:06:30,360
sit me down and ask me to explain in
a few sentences what is a P junction,
108
00:06:30,500 --> 00:06:35,400
how does it work? I don't know, but
I can regurgitate those derivations.
109
00:06:35,540 --> 00:06:39,110
So we have to think, yeah,
110
00:06:39,290 --> 00:06:41,110
we radically need to change things. Yeah.
111
00:06:42,460 --> 00:06:46,110
It's an interesting one.
I I think back to sort of,
112
00:06:46,790 --> 00:06:51,390
I dunno what March, 2020 with
Covid on the, you know, suddenly,
113
00:06:51,850 --> 00:06:56,070
uh, arriving and I conversations happening
at universities that were sort of,
114
00:06:56,070 --> 00:07:00,310
you know, well maybe by January
we'll be back to normal.
115
00:07:01,250 --> 00:07:05,790
And the, you know, how that changed
the way that we had to teach and,
116
00:07:05,890 --> 00:07:09,790
and the way we had to do things.
It was kind of, you know,
117
00:07:10,570 --> 00:07:13,230
one thing, but this feels like we,
118
00:07:13,290 --> 00:07:17,110
we don't know where if church G B
T four, as you say, exponential,
119
00:07:17,110 --> 00:07:20,750
if church G B T four can do what it
is just being talked about doing,
120
00:07:20,780 --> 00:07:24,550
then what does 5, 6, 7, 8, 10 do?
121
00:07:24,570 --> 00:07:29,390
And so we can't plan now
based on what church G B T
122
00:07:29,490 --> 00:07:31,270
and others can do now. Yeah. Right.
123
00:07:32,090 --> 00:07:35,670
So I was dis I've gotta admit
with the, with the, uh, you know,
124
00:07:35,670 --> 00:07:38,150
COVID pandemic et cetera,
and, and all the changes,
125
00:07:38,350 --> 00:07:43,070
I really thought we had a
massive opportunity to grasp the
126
00:07:43,070 --> 00:07:46,030
nettle and shake things up.
That's not what happened.
127
00:07:46,410 --> 00:07:49,590
And some of that might
be her fault in general,
128
00:07:49,820 --> 00:07:52,590
sort of academic university inertia,
129
00:07:52,730 --> 00:07:54,550
but a lot of it's also
coming from the students.
130
00:07:54,920 --> 00:07:58,510
There is nothing easier for the students
and for us for them to sit in a lecture
131
00:07:58,510 --> 00:08:02,150
theater among 200 other students and
listen to us drone on for 50 minutes.
132
00:08:02,850 --> 00:08:03,420
And it's,
133
00:08:03,420 --> 00:08:07,830
it's really disappointing that that's
seen as the pinnacle of education
134
00:08:09,300 --> 00:08:12,360
and that's what they want or that's
what they claim to want, they want,
135
00:08:12,660 --> 00:08:14,160
you know, well, it's not what they claim.
136
00:08:14,230 --> 00:08:16,640
It's absolutely what
quite a few of them want.
137
00:08:17,780 --> 00:08:19,800
So I don't wanna generalize
across the board.
138
00:08:19,800 --> 00:08:23,280
There are others who are keen on the sort
of more interactive approach and sort
139
00:08:23,280 --> 00:08:27,090
of trying to shake things up a little
bit, but quite a few just wanted,
140
00:08:27,250 --> 00:08:29,890
I want to go to the lecture
theater, I wanna listen to you,
141
00:08:30,290 --> 00:08:33,370
I wanna note it down and then I wanna
regurgitate it in the exam. The.
142
00:08:33,430 --> 00:08:33,670
The,
143
00:08:33,670 --> 00:08:38,090
the simplest solution from an education
point of view seems to me to be that in
144
00:08:38,090 --> 00:08:38,830
terms of assessment,
145
00:08:38,830 --> 00:08:43,330
we go back to close book exams that happen
in a room at a particular time. Yeah.
146
00:08:43,430 --> 00:08:45,770
Is that, is that the only
solution or what, I mean.
147
00:08:46,010 --> 00:08:49,250
I think it will be disappointing
if that's the only solution,
148
00:08:49,470 --> 00:08:54,410
but particularly with chat G P T and
whatever G P T models in general,
149
00:08:54,990 --> 00:08:58,000
when they can, they're
now getting, as I said,
150
00:08:58,000 --> 00:09:01,960
28 outta 30 on on first
year, uh, physics problems.
151
00:09:02,660 --> 00:09:07,440
And I also noticed, um,
Scott Aronson, who's, uh,
152
00:09:07,440 --> 00:09:10,920
quantum computing expert is
bloggers brilliant by the way. Um,
153
00:09:11,260 --> 00:09:16,240
he set a G P T for his
quantum computing exam paper,
154
00:09:16,460 --> 00:09:21,120
so fairly advanced quantum
concepts and it got 69, 60 9%.
155
00:09:22,300 --> 00:09:25,480
So just, just slightly
off a first. Mm-hmm.
156
00:09:26,020 --> 00:09:28,040
So if the,
157
00:09:28,740 --> 00:09:32,880
if we continue to set traditional exam
papers, it has to be in, as you say,
158
00:09:32,880 --> 00:09:36,960
in under those closed conditions.
But that just seems archaic to me.
159
00:09:38,030 --> 00:09:39,160
What would you like to see then?
160
00:09:39,350 --> 00:09:43,520
Yeah, that's a very good question.
. It's a very, so the,
161
00:09:43,600 --> 00:09:48,440
I think we need to embed, um, G
P T, um, and these, these models,
162
00:09:49,050 --> 00:09:52,840
theis of whatever, um, stripe into our,
163
00:09:53,060 --> 00:09:56,440
our teaching and indeed into
our assessment. I don't know,
164
00:09:57,020 --> 00:10:01,800
we get to the point where
we tell G P T four, um, to,
165
00:10:01,860 --> 00:10:04,160
it can already, G P T
four can already do this,
166
00:10:04,160 --> 00:10:06,920
but it can simulate certain
physical systems, I don't know,
167
00:10:06,920 --> 00:10:09,320
like a simple pendulum
or a driven pendulum.
168
00:10:09,780 --> 00:10:11,600
We ask the students to protect the output,
169
00:10:11,600 --> 00:10:16,200
but we could even do things
like ask G P T to deliberately
170
00:10:16,510 --> 00:10:20,440
include errors in the physics
and ask for the students to,
171
00:10:20,660 --> 00:10:25,120
to pick those out. I think we have to be
really open with the students. I think,
172
00:10:25,220 --> 00:10:29,720
you know, we say use G P T, um, use ais,
173
00:10:30,140 --> 00:10:33,000
use whatever you can in
terms of writing your report,
174
00:10:33,980 --> 00:10:36,200
but tell us where you've used that.
175
00:10:38,480 --> 00:10:39,780
The issue then is, of course,
176
00:10:39,790 --> 00:10:43,380
let's rely on a great deal in the
students honesty and that might happen.
177
00:10:43,480 --> 00:10:48,420
So we have to think about
just how we build in use
178
00:10:49,560 --> 00:10:53,980
of, of these models,
these, these ais into our,
179
00:10:54,250 --> 00:10:57,500
into our assessment and indeed our
teaching processes where it's great,
180
00:10:57,560 --> 00:11:00,540
for example, on the teaching side
rather than the assessment side is,
181
00:11:00,560 --> 00:11:01,660
and I've already used to do,
182
00:11:01,720 --> 00:11:04,660
do this generates some multiple
choice questions for me on this topic,
183
00:11:05,240 --> 00:11:06,240
and that's great.
184
00:11:06,260 --> 00:11:10,800
And then you use those in the class
and with whatever online polling thing.
185
00:11:11,100 --> 00:11:15,120
So it's a real benefit. Then,
um, there's also things like, um,
186
00:11:15,500 --> 00:11:19,230
in principle, uh, you
know, close captioning,
187
00:11:19,420 --> 00:11:21,110
putting captions on videos,
188
00:11:21,110 --> 00:11:26,070
lecture videos or whatever we could ask
it to do that might still struggle with
189
00:11:26,070 --> 00:11:30,310
my accent . Um, but,
uh, yeah, so there are,
190
00:11:30,310 --> 00:11:32,670
there are a lot of,
there a lot of benefits.
191
00:11:32,930 --> 00:11:36,590
But the assessment side
is, is the tricky one,
192
00:11:37,370 --> 00:11:40,870
as you say. Well, how, where's, where's
your stance? How do you feel? Do you,
193
00:11:40,870 --> 00:11:43,830
are you happy with traditional
exams as they are? No.
194
00:11:44,290 --> 00:11:48,710
I'm, I mean, so I don't teach physics.
I teach wildlife filmmaking, podcasting,
195
00:11:49,050 --> 00:11:50,990
uh, science filmmaking. So it's,
196
00:11:50,990 --> 00:11:54,880
it's quite easy for me to move into
areas where, at least at the moment,
197
00:11:54,890 --> 00:11:59,280
artificial intelligence isn't particularly
good in terms of making podcasts and
198
00:11:59,290 --> 00:11:59,820
films.
199
00:11:59,820 --> 00:12:01,600
But there's the thing,
why, so that's exactly it.
200
00:12:01,780 --> 00:12:05,470
So that's the type of thing we should be
asking our students to do. Make videos,
201
00:12:05,470 --> 00:12:08,110
make short YouTube videos, make podcasts,
202
00:12:09,210 --> 00:12:12,510
add in a lot more sort of, of
the creative side of things.
203
00:12:12,570 --> 00:12:15,550
One thing that does drive me up the wall,
204
00:12:16,130 --> 00:12:18,110
and particularly as my kids
have gone through school,
205
00:12:18,130 --> 00:12:22,510
is that we have the STEM subjects over
here, and that's for all the drones.
206
00:12:22,730 --> 00:12:26,950
And over here we have the creative
subjects and there's so much creativity,
207
00:12:27,010 --> 00:12:30,590
of course, in science, but we need
to exploit that a little bit more in,
208
00:12:30,730 --> 00:12:34,710
in undergraduate degrees. Yeah.
And indeed postgrad, you know,
209
00:12:34,710 --> 00:12:35,510
postgraduate degrees.
210
00:12:35,510 --> 00:12:38,030
As well. No, I, I mean, you're not
gonna find me disagree with that,
211
00:12:38,030 --> 00:12:42,270
obviously . Cause it's my, it's
my bread and butter. But, um, yeah, no,
212
00:12:42,310 --> 00:12:43,550
I th I mean, it's,
213
00:12:43,610 --> 00:12:46,390
it seems to me though that there's
possibly a time in the future where,
214
00:12:46,970 --> 00:12:51,870
you know, podcasts are being generated
by artificial intelligence. I,
215
00:12:52,430 --> 00:12:54,710
I thought maybe they're, but
maybe they're not that good.
216
00:12:55,170 --> 00:12:58,350
Oh, I think there's a time
coming. Um, there's a,
217
00:12:58,730 --> 00:13:02,510
in the not too distant future where
if I want to generate a lecture video,
218
00:13:03,110 --> 00:13:04,550
I type, I give it some,
219
00:13:04,940 --> 00:13:09,390
some video of me and the key
themes and it generates a deep f
220
00:13:10,150 --> 00:13:12,270
I don't think that time is too far off.
221
00:13:12,930 --> 00:13:14,390
No. Would you use that.
222
00:13:15,470 --> 00:13:19,710
? No, I wouldn't. For the same
reason that I wouldn't, I like to write,
223
00:13:20,490 --> 00:13:23,550
um, and I wouldn't, I wouldn't use,
224
00:13:24,150 --> 00:13:28,790
I know this is increasingly common to
use chat G P T or g p t four to actually
225
00:13:28,790 --> 00:13:30,150
write the introductions to papers.
226
00:13:30,510 --> 00:13:32,470
Journals have said that
they're not gonna accept that.
227
00:13:32,490 --> 00:13:37,190
But I've read quite a few blog posts and
quite a few articles where people are
228
00:13:37,190 --> 00:13:41,510
freely admitted. They've used G P T
four in some sort of creative sense.
229
00:13:41,530 --> 00:13:45,750
And that doesn't sit well
with me cuz it's essentially
230
00:13:46,440 --> 00:13:47,273
plagiarism.
231
00:13:47,680 --> 00:13:50,190
We've had essay mills
for some time though,
232
00:13:50,190 --> 00:13:53,110
and maybe the people who make essay
mills should be the most worried. Cause.
233
00:13:53,110 --> 00:13:56,830
They don't need that. Oh, they are.
They really are. So it, it, but again,
234
00:13:57,220 --> 00:14:02,030
there's, there's, um, some on
the education side actually,
235
00:14:02,050 --> 00:14:06,550
the legal, well they would claim the
legal as well, allegedly . Um,
236
00:14:07,370 --> 00:14:09,630
um, but Eva Can Academy, for example,
237
00:14:09,850 --> 00:14:14,560
is going to use G P T four
as a virtual tutor. So it's,
238
00:14:14,790 --> 00:14:18,520
it's, um, and that, that's
also a good idea. In fact,
239
00:14:18,580 --> 00:14:23,160
we could use G P T four
as a virtual shooter in
240
00:14:23,470 --> 00:14:27,200
sort of real time in a, in a
classroom or indeed in a lab.
241
00:14:28,340 --> 00:14:32,790
It's, you know, the one is a big issue
in, in, if you've got a class, I, um,
242
00:14:32,790 --> 00:14:36,190
one of my first jobs here was to
revamp her second year, uh, lab.
243
00:14:36,530 --> 00:14:41,090
And I was second year, uh, lab
organizer for, oh God, seven years,
244
00:14:41,090 --> 00:14:42,130
eight years, I can't remember.
245
00:14:42,590 --> 00:14:45,530
And one of the issues is always
having enough demonstrators.
246
00:14:45,780 --> 00:14:49,170
Cause particularly in a Friday afternoon
when the experiments aren't working and
247
00:14:49,450 --> 00:14:53,480
14 hands are up, you're gonna, if we
had a sort of virtual lab assistant,
248
00:14:53,480 --> 00:14:56,200
particularly now because
it can interpret images,
249
00:14:56,340 --> 00:14:58,960
you could almost envision a a,
250
00:14:59,360 --> 00:15:03,760
a situation whereby the student takes a
a photo or takes a video of this piece
251
00:15:03,760 --> 00:15:06,280
of kid isn't working. What
are your suggestions? So.
252
00:15:06,280 --> 00:15:08,760
Uh, can we just step away from education
for a bit? Because there's, it's,
253
00:15:08,830 --> 00:15:11,440
it's not just gonna affect
physicists who are learning. Is it,
254
00:15:11,440 --> 00:15:14,600
it's gonna affect physicists who
are, uh, in practice, isn't it?
255
00:15:14,750 --> 00:15:18,680
It's interesting, this paper that I, um,
just from the York, a guy called, um,
256
00:15:18,850 --> 00:15:21,920
Colin G West who's at the
University of Colorado,
257
00:15:22,700 --> 00:15:27,080
and it's was uploaded actually just
April, April the 16th. There's a great,
258
00:15:27,220 --> 00:15:28,880
the final paragraph of his paper,
259
00:15:29,130 --> 00:15:31,480
which is well worth reading
and get in the archive.
260
00:15:31,550 --> 00:15:33,960
I'll just give everybody the
tightly they wanna chase it up.
261
00:15:33,960 --> 00:15:37,800
It's advances in apparent conceptual
physics reasoning in G P T four.
262
00:15:38,920 --> 00:15:40,820
And the final paragraph is ultimately,
263
00:15:40,820 --> 00:15:44,540
perhaps what is most intriguing
is what G P T four tells us.
264
00:15:44,560 --> 00:15:46,380
Not about the future of our classrooms,
265
00:15:46,680 --> 00:15:49,940
but about our future conceptions
of physics as a discipline.
266
00:15:50,780 --> 00:15:52,780
G p t four succeeds at physics tasks,
267
00:15:52,850 --> 00:15:57,100
even though it is only explicitly
programmed to emulate human speech and not
268
00:15:57,100 --> 00:16:00,340
provided directly with any of the formal
structure of mathematics or the laws of
269
00:16:00,340 --> 00:16:02,140
physics. So it's,
270
00:16:02,770 --> 00:16:07,340
it's changing not only how we teach,
but it will change how we do research.
271
00:16:07,400 --> 00:16:11,500
In fact, the, the computer scientists
are already talking about, you know,
272
00:16:11,600 --> 00:16:16,580
coding will evolve to the point where
it's prompt engineering, you, you,
273
00:16:16,580 --> 00:16:20,100
you engineer the right prompts to
feed to an a AI to give you the code.
274
00:16:20,800 --> 00:16:25,500
And we're again, probably
not too far off. In fact,
275
00:16:25,520 --> 00:16:29,780
in some cases it's already doing this
where you feed the data and it picks out
276
00:16:29,780 --> 00:16:33,860
correlations in that
data and it's okay that,
277
00:16:33,860 --> 00:16:34,940
that you can sort of accept.
278
00:16:35,120 --> 00:16:38,860
But then it goes the further step and
comes up with a physical model to explain
279
00:16:39,700 --> 00:16:43,520
tho those, those correlations.
That's very close, I would say.
280
00:16:44,260 --> 00:16:48,350
And then again, where does
that leave us ?
281
00:16:49,090 --> 00:16:53,430
So, so if we're gonna work alongside
these things, we need to find,
282
00:16:55,060 --> 00:16:56,260
I guess it's again that,
283
00:16:56,260 --> 00:16:59,780
that that collaboration between the
artificial neuronets and the organic
284
00:16:59,850 --> 00:17:03,900
neuronets, we will complement
each other and we'll hopefully,
285
00:17:04,200 --> 00:17:08,810
I'd like to think that
it's, it's creativity.
286
00:17:08,910 --> 00:17:13,290
The AI's creativity will be
sort of complimenting ours.
287
00:17:13,290 --> 00:17:18,010
It works in a slightly different
way, perhaps even then. It's,
288
00:17:18,080 --> 00:17:19,170
it's strange thing, you know,
289
00:17:19,170 --> 00:17:22,090
we've got billions upon
billions of neurons in there.
290
00:17:22,090 --> 00:17:26,010
You've got effectively
billions upon billions of
artificial neurons for the, um,
291
00:17:26,590 --> 00:17:30,600
for some of these very sophisticated
models. What's the difference?
292
00:17:31,500 --> 00:17:35,000
We could get into a bit of
an existential crisis here, .
293
00:17:36,150 --> 00:17:39,120
Yeah, yeah, absolutely.
I, so like if you are,
294
00:17:39,300 --> 00:17:43,760
if you're working at LIGO or you're
working at cern or is this affecting you?
295
00:17:44,360 --> 00:17:49,040
I, I think it could be in terms of, of
data analysis in terms of, it's already,
296
00:17:49,420 --> 00:17:54,200
so we're trying to build in different
neural nets into what we do,
297
00:17:54,200 --> 00:17:57,720
which is all about image
interpretation and, oh,
298
00:17:57,860 --> 00:18:01,360
at least partly about image
interpretation, but also
about spectral analysis.
299
00:18:01,580 --> 00:18:04,840
So looking for correlations
in different spectra.
300
00:18:04,840 --> 00:18:06,320
There's already been a couple of papers,
301
00:18:07,200 --> 00:18:12,040
a number of papers published over
the last few years where they fed
302
00:18:12,800 --> 00:18:17,470
a lot of data over, say the
course of a decade to, uh,
303
00:18:18,390 --> 00:18:23,110
unsupervised neural net and asked it to
find the correlations and the data that
304
00:18:23,110 --> 00:18:27,230
humans didn't spot. And there's actually
been a nature paper published on this,
305
00:18:27,230 --> 00:18:30,750
whether the AI spotted those
correlations that we didn't spot.
306
00:18:33,260 --> 00:18:38,160
So I think it's, it's like all these
things, it's not gonna be of, you know,
307
00:18:38,410 --> 00:18:39,880
black or white issue. It's gonna sort,
308
00:18:39,880 --> 00:18:43,280
everything's gonna sort of fall in that
gray area where we have some sort of
309
00:18:43,520 --> 00:18:45,440
convergence, I hope. Well,
310
00:18:45,440 --> 00:18:49,960
it's either that or really or AI
overloads are gonna take over
311
00:18:50,260 --> 00:18:50,960
doomed.
312
00:18:50,960 --> 00:18:53,640
. Yeah, yeah, yeah.
Are you a science fiction fan?
313
00:18:54,280 --> 00:18:57,640
I am, yeah. I'm a big Star Trek fan.
Oh, okay. For one thing. Yeah. Okay.
314
00:18:57,980 --> 00:19:02,680
But I mean, should we be watching,
315
00:19:03,180 --> 00:19:08,160
you know, 2001, um, ex Machina mm-hmm.
316
00:19:08,200 --> 00:19:09,160
and thinking
mm-hmm. .
317
00:19:09,700 --> 00:19:10,533
No, that's a classic.
318
00:19:10,570 --> 00:19:11,280
Isn't it? Both.
319
00:19:11,280 --> 00:19:13,200
Of those are classics, of
course. Yeah. Right. Yeah.
320
00:19:13,200 --> 00:19:15,760
Yeah. Her, you know, this, there's,
321
00:19:15,760 --> 00:19:18,120
there's a number of films out
there about this, this topic.
322
00:19:18,420 --> 00:19:22,440
How close are they to sort of
the reality that we're gonna see?
323
00:19:22,580 --> 00:19:25,720
If you'd asked me 10 years ago,
I'd said, oh, that's just so,
324
00:19:26,020 --> 00:19:29,040
so many centuries ahead now. Um,
325
00:19:29,690 --> 00:19:34,320
we're they're pretty close, I would
say. I would say they're pretty close.
326
00:19:34,900 --> 00:19:37,750
And, um, yeah, it's,
327
00:19:38,020 --> 00:19:41,950
well already we've had Musk at Ireland
about, I dunno how many it is in total,
328
00:19:42,110 --> 00:19:44,790
a thousand, um, thought leaders as I,
329
00:19:44,850 --> 00:19:49,150
I'm sure they'd like to
describe themselves, um, call
for sort of moratorium on,
330
00:19:49,170 --> 00:19:51,310
on AI research. So it's,
331
00:19:52,930 --> 00:19:56,970
I think that's misguided
as well. Cuz again,
332
00:19:56,970 --> 00:19:58,770
trying has shut it down for one thing,
333
00:19:59,030 --> 00:20:02,010
you'll just drive stuff underground.
334
00:20:02,510 --> 00:20:06,850
And plus Musk with his Neurolink
stuff, which is not the most open, um,
335
00:20:07,130 --> 00:20:09,170
research platform in the world, um,
336
00:20:09,310 --> 00:20:12,610
is it's just a little bit hypocritical
for him to call for a moratorium.
337
00:20:13,190 --> 00:20:17,890
So yeah, it's, I, in terms
of the science fiction,
338
00:20:18,240 --> 00:20:22,490
yeah, I think there's lots that
we can learn from the dystopian
339
00:20:23,650 --> 00:20:25,850
scenarios in some science
fiction novels. Yeah.
340
00:20:26,400 --> 00:20:29,170
Whether we would or not is another
question, isn't it? I mean, I've,
341
00:20:29,370 --> 00:20:30,010
I dunno about you.
342
00:20:30,010 --> 00:20:34,730
I've read Red Ready Player One and I
can see that happening right before our
343
00:20:34,730 --> 00:20:38,410
eyes at this particular moment, so I
don't see why. Yeah, I mean they, it's,
344
00:20:38,560 --> 00:20:41,450
it's all very well and good
academic sitting around saying,
345
00:20:41,670 --> 00:20:44,010
we need to work out how
to do this. But, you know,
346
00:20:44,790 --> 00:20:49,610
so are the other people who
are maybe not as motivated
347
00:20:49,610 --> 00:20:50,443
in the same way?
348
00:20:50,880 --> 00:20:54,170
Precisely Absolutely. And
Ready Player one. Of course,
349
00:20:54,610 --> 00:20:58,410
I was bitterly disappointed with the
movie because they stuck out the bit about
350
00:20:58,480 --> 00:20:59,210
Rush.
351
00:20:59,210 --> 00:21:01,290
. Yeah. Okay. Yeah.
352
00:21:03,050 --> 00:21:06,450
I was so happy of Brooke
that included Rush's 2112.
353
00:21:06,450 --> 00:21:10,490
There's a dystopian future for you as
well. Wow. Um, as a core part of the plot.
354
00:21:10,730 --> 00:21:11,280
. Yeah.
355
00:21:11,280 --> 00:21:14,450
Yeah, yeah, yeah. Oh, we're gonna go
off topic slightly, but let's do that.
356
00:21:14,450 --> 00:21:16,010
So have you seen them live Rush?
357
00:21:16,830 --> 00:21:19,730
Oh, lots of times, yeah. Oh, you
a fan as well? Yeah, yeah, yeah.
358
00:21:19,730 --> 00:21:19,930
Yeah.
359
00:21:19,930 --> 00:21:22,090
Yeah. Oh, oh, oh, brilliant. Yeah. Um,
360
00:21:22,530 --> 00:21:26,050
I saw them four times on the time machine
tour cause they did moving pictures in
361
00:21:26,070 --> 00:21:30,010
its entirety. Brilliant.
So, um, favorite Mike? So.
362
00:21:30,010 --> 00:21:34,650
The conversation did slightly veer
off at this moment as we went into
363
00:21:34,830 --> 00:21:38,450
our favorite concerts by
the Canadian Rock Band Rush.
364
00:21:38,710 --> 00:21:42,850
And as interesting as I'm sure you'd find
that we did return to the conversation
365
00:21:42,850 --> 00:21:47,830
of artificial intelligence. So I did
speak to somebody who, who signed that,
366
00:21:48,210 --> 00:21:51,150
um, that letter the moratorium
with, with Elon Musk.
367
00:21:51,150 --> 00:21:54,910
It wa it wasn't Elon Musk that I spoke
to , but it was someone else who,
368
00:21:54,950 --> 00:21:55,910
I won't, I won't name it.
369
00:21:55,910 --> 00:21:59,430
And they said that essentially
that was a line in the sand.
370
00:21:59,660 --> 00:22:03,390
They didn't think it was gonna change
anything. It was just to say, look,
371
00:22:03,390 --> 00:22:05,030
we know this is coming and we're alerting.
372
00:22:05,030 --> 00:22:08,630
You. Yeah. It's a pr exercise, really.
Yeah, absolutely. No, I, I, I agree.
373
00:22:08,630 --> 00:22:13,590
And it's, it, I think there
is some value in that,
374
00:22:14,260 --> 00:22:16,990
even if it is just a
sort of pr stunt to, to,
375
00:22:17,050 --> 00:22:21,710
to flag up that we need to
be concerned about this. Um,
376
00:22:22,370 --> 00:22:23,750
but it's, you know,
377
00:22:23,750 --> 00:22:27,790
it really is a question of bolting the
stable door after the horse run away.
378
00:22:29,230 --> 00:22:31,570
Uh, I don't think, you know,
the genie's out of the bottle.
379
00:22:31,770 --> 00:22:36,290
I don't think if I can
pile around , um, I,
380
00:22:36,970 --> 00:22:40,810
I don't think it's gonna be
easy to, to stop this as you,
381
00:22:40,870 --> 00:22:45,330
for us exactly the reason you say
you might well have the academic
382
00:22:46,010 --> 00:22:49,650
research in the academic sector shut
down, but not everybody's a good actor.
383
00:22:50,230 --> 00:22:51,063
So yeah.
384
00:22:51,590 --> 00:22:55,810
If you are, um, listening to this
podcast and you are also, you know, in,
385
00:22:55,830 --> 00:22:59,490
in school now, the world
looks very different. We.
386
00:22:59,720 --> 00:23:04,490
Need to ensure that we
are preparing kids for a,
387
00:23:05,410 --> 00:23:08,010
a very much a brave new world, um,
388
00:23:08,100 --> 00:23:12,370
where a lot of the skills that we've
taught in the past will be completely
389
00:23:12,530 --> 00:23:13,190
redundant.
390
00:23:13,190 --> 00:23:14,890
So there's creativity,
391
00:23:16,480 --> 00:23:20,450
there's being able to educate
yourself. What does that mean?
392
00:23:20,740 --> 00:23:25,050
Don't rely on, you know, us
to go, there's the textbook,
393
00:23:25,160 --> 00:23:29,410
there's what you should read. It's like
I should in principle be able to go,
394
00:23:29,410 --> 00:23:33,750
right. Here's a problem in
quantum mechanics. If you
go, that's where I'd like,
395
00:23:34,370 --> 00:23:36,910
you know, you may say that's just
making us completely redundant,
396
00:23:36,910 --> 00:23:40,470
but it's a question of
how we design the, um,
397
00:23:41,170 --> 00:23:44,750
the activities, how we
design the, the, um,
398
00:23:44,850 --> 00:23:48,120
the assessment as well that will
lead students to that point.
399
00:23:48,260 --> 00:23:51,600
So at least by the time they're in their
final year or the penultimate year,
400
00:23:51,750 --> 00:23:53,800
they're at the point where sort of,
401
00:23:54,020 --> 00:23:58,120
we are almost redundant in
terms of delivering material.
402
00:23:59,000 --> 00:24:03,490
It's more about how they seek
out that material process,
403
00:24:03,720 --> 00:24:05,770
that material, analyze that,
404
00:24:05,990 --> 00:24:09,880
be that with chat g p t and the like. Um,
405
00:24:10,580 --> 00:24:14,960
and then also present it.
And that's coming back to
the podcast thing, you know,
406
00:24:14,960 --> 00:24:19,360
being able to present it, being
able to be persuasive, um,
407
00:24:19,660 --> 00:24:23,000
in terms of either written,
oral communication, whatever,
408
00:24:23,650 --> 00:24:25,840
those are valuable skills,
409
00:24:26,660 --> 00:24:30,320
but they're far outside the traditional
skills we'd expect of a physicist.
410
00:24:30,550 --> 00:24:31,040
There's.
411
00:24:31,040 --> 00:24:32,400
Part of me that wants to say, well,
412
00:24:32,440 --> 00:24:34,560
I do teach on a master's
in science communication.
413
00:24:34,560 --> 00:24:36,880
It sounds like we're
gonna be needed, but this.
414
00:24:36,880 --> 00:24:38,920
Yeah, you absolutely will. Um, but oh,
415
00:24:38,920 --> 00:24:42,440
we should get you to Nottingham sometimes
soon to tell us all honestly I'm,
416
00:24:42,470 --> 00:24:46,760
yeah. Um, to tell us all
about that. Yeah. Um,
417
00:24:46,950 --> 00:24:49,320
because we've got quite a few
students who are interested in science
418
00:24:49,600 --> 00:24:53,520
communicators and I used to teach, um,
no, that's absolutely the wrong word.
419
00:24:54,500 --> 00:24:58,400
Uh, and I don't like the word deliver
at all. Uh, how about convene?
420
00:24:58,640 --> 00:25:02,040
I used to convene a module, a
fourth year module called, um,
421
00:25:02,040 --> 00:25:06,040
the Politics Perception and Philosophy
of Physics, uh, which is still running.
422
00:25:06,140 --> 00:25:09,680
I'm no longer doing, I, I paid my Jews,
I did it for seven or eight years,
423
00:25:10,060 --> 00:25:14,400
but it was very much developed along
the lines of much more like a humanities
424
00:25:14,400 --> 00:25:17,200
module whereby discussion, debate.
425
00:25:18,170 --> 00:25:23,030
And we cover things like, uh, the
so-called science wars where, you know,
426
00:25:23,030 --> 00:25:27,030
you've got one group of sociologists
saying that there's nothing truly
427
00:25:27,590 --> 00:25:30,790
discovered by science evidence,
culturally related, et cetera.
428
00:25:31,170 --> 00:25:33,270
And you've obviously got the
scientists who work at that,
429
00:25:33,570 --> 00:25:38,550
but also things like peer review, how
does peer review work, um, you know,
430
00:25:38,810 --> 00:25:41,030
gender balance, all those types of things.
431
00:25:41,730 --> 00:25:45,870
And the first time I ran it and tried
to sort of prompt this discussion,
432
00:25:46,030 --> 00:25:46,990
I was really,
433
00:25:47,090 --> 00:25:52,030
really worried because it's a group of
40 physics graduates Oh, undergraduates.
434
00:25:52,290 --> 00:25:54,950
And I'm thinking, okay, this is
gonna be tumble weight. So, okay.
435
00:25:55,370 --> 00:25:59,070
And I asked for the first, do
the little intro and then, um,
436
00:25:59,290 --> 00:26:03,990
ask for the first question and
a hand goes up and then the
437
00:26:03,990 --> 00:26:06,630
floodgates opened and I had to kick
them out after an hour and a half.
438
00:26:07,090 --> 00:26:08,790
So some of the, in that case,
439
00:26:08,820 --> 00:26:12,470
they knew the module they were going
into and they knew what it was like,
440
00:26:12,580 --> 00:26:15,550
what it was gonna be like. So they
were self-selecting in that sense.
441
00:26:15,650 --> 00:26:19,950
But I think this idea that absolutely
every physicist is completely
442
00:26:20,480 --> 00:26:25,110
inept and entirely incapable of,
443
00:26:25,530 --> 00:26:28,950
um, communicating. He says as he
struggles to get through that sentence,
444
00:26:29,390 --> 00:26:33,150
, um, is
is not always the case.
445
00:26:33,650 --> 00:26:37,390
So yes, the ability to communicate is,
446
00:26:38,490 --> 00:26:43,400
uh, just essential. And again, it's
something that's not traditionally in,
447
00:26:43,700 --> 00:26:47,040
in university physics courses,
although there's a lot of focus,
448
00:26:47,170 --> 00:26:49,360
increasing focus on, um,
449
00:26:49,760 --> 00:26:52,600
presentation in terms of
scientific presentation,
450
00:26:52,600 --> 00:26:55,640
technical presentation less so on.
451
00:26:55,990 --> 00:26:59,600
Well explain to, you know,
somebody with absolutely, you know,
452
00:26:59,600 --> 00:27:00,440
background in science,
453
00:27:01,260 --> 00:27:04,880
why we should care about your particular
research in three sentences, golf.
454
00:27:05,330 --> 00:27:06,140
Absolutely.
455
00:27:06,140 --> 00:27:10,120
You've recently been on the Physics World
Weekly podcast speaking about quantum
456
00:27:10,620 --> 00:27:15,530
woo. Yeah. And that's, I
think that's my concern.
457
00:27:15,670 --> 00:27:20,610
My main concern with life generally is
people following stuff that's not true
458
00:27:20,610 --> 00:27:25,290
and, and leading themselves down a or
being led down particular paths which are
459
00:27:25,560 --> 00:27:27,210
harmful to them and others.
460
00:27:27,430 --> 00:27:30,570
And with G P T and others
like it making up stuff,
461
00:27:30,570 --> 00:27:32,170
including making up references.
462
00:27:32,210 --> 00:27:37,090
W there are a number of cases
where students have used
G P T and the way they've
463
00:27:37,090 --> 00:27:41,490
been caught is not in terms of that,
that, you know, this you're nodding,
464
00:27:41,490 --> 00:27:45,090
you know, this has not, but for
those who perhaps don't, they, um,
465
00:27:45,310 --> 00:27:47,730
has not been in terms
of the writing style,
466
00:27:48,040 --> 00:27:52,970
it's been that G P T has made up
references and that's the only way it was
467
00:27:52,970 --> 00:27:53,200
caught.
468
00:27:53,200 --> 00:27:54,130
Yeah, no, absolutely.
469
00:27:54,760 --> 00:27:58,930
It's happened to me with a particular
student doing a dissertation
470
00:27:59,710 --> 00:28:01,570
and essentially the
whole dissertation was,
471
00:28:01,790 --> 00:28:04,850
was done via chat G P T.
472
00:28:05,750 --> 00:28:08,670
And well, there's no
policy really .
473
00:28:08,700 --> 00:28:12,310
Like there certainly wasn't at the
time for, um, what we do with that.
474
00:28:12,890 --> 00:28:17,870
And the, not only was it only really
recognizable by the fact that the,
475
00:28:18,410 --> 00:28:23,150
um, citations and references
were all made up, but um,
476
00:28:23,460 --> 00:28:26,230
that was the only way we could
sort of say anything about it.
477
00:28:26,500 --> 00:28:28,630
Yeah, exactly. Exactly.
478
00:28:28,970 --> 00:28:33,110
So this is what I mean about sort of
genie and the bot genie outta the bottle.
479
00:28:33,340 --> 00:28:35,390
It's, if we can't track this,
480
00:28:35,390 --> 00:28:38,030
and I think there have been
some claims with chat G P T,
481
00:28:38,030 --> 00:28:41,510
at least in terms of looking at the,
482
00:28:42,090 --> 00:28:46,150
the way it uses or constructs
the sentences. There may be,
483
00:28:46,760 --> 00:28:50,270
there may be possibilities of
detecting that it is AI generated.
484
00:28:50,830 --> 00:28:55,030
I don't think with G P T four, that's
not really the case. I could be wrong,
485
00:28:55,050 --> 00:28:56,110
but I it's if,
486
00:28:56,130 --> 00:29:00,190
so it's much less the case and as it
evolves it'll get to the point where it's
487
00:29:00,190 --> 00:29:01,023
indistinguishable.
488
00:29:01,270 --> 00:29:05,590
I i, if we're teaching students
to learn for themselves,
489
00:29:06,410 --> 00:29:08,710
the, the tool they're gonna
use for that is the internet.
490
00:29:08,740 --> 00:29:12,670
Part of that is chat g p t four
or whatever it is in the future.
491
00:29:13,610 --> 00:29:17,350
And is anybody in control
of that? Who's is that?
492
00:29:17,350 --> 00:29:22,110
Nobody's in control of the
internet. So it's just how,
493
00:29:22,690 --> 00:29:25,870
how do you discern what's true? Exactly.
494
00:29:26,380 --> 00:29:31,270
Exactly. And that's a skill, that's
a skill that many of us struggle to,
495
00:29:31,330 --> 00:29:36,030
to, to, to keep on top of. So getting
that across to the next generation,
496
00:29:36,030 --> 00:29:39,550
getting that across to students is,
is, is certainly gonna be problematic.
497
00:29:39,610 --> 00:29:44,600
But it's absolutely essential to be,
um, what's an authoritative source,
498
00:29:45,220 --> 00:29:46,053
you know,
499
00:29:46,100 --> 00:29:50,120
is I have my certain broadsheets that
I like and even by using the term
500
00:29:50,120 --> 00:29:52,840
broadsheet, I date myself
. Absolutely. You know,
501
00:29:52,860 --> 00:29:56,840
as the guardian and authoritative
source is the Times and authoritative
502
00:29:56,840 --> 00:30:01,360
source, you know, is Fox
News an authoritative source?
503
00:30:01,940 --> 00:30:06,160
Why should they use, not use Fox News
and use the Guard Unified? Those are,
504
00:30:07,020 --> 00:30:07,630
you know,
505
00:30:07,630 --> 00:30:12,040
very deep questions because of course the
guardian can also get it wrong and you
506
00:30:12,040 --> 00:30:14,320
know, journals can get it
wrong. They've been examples of,
507
00:30:14,420 --> 00:30:18,960
of many examples of where peer reviews
completely failed and there's junk out
508
00:30:18,960 --> 00:30:22,960
there. So it's, yeah, it's a difficult,
509
00:30:23,030 --> 00:30:25,520
difficult can of worms.
510
00:30:25,880 --> 00:30:29,320
, there's possibly a world, I
can't imagine this world of course,
511
00:30:29,500 --> 00:30:31,160
but there's possibly a
world where, you know,
512
00:30:31,230 --> 00:30:33,600
even political leaders
might not tell the truth.
513
00:30:33,840 --> 00:30:36,880
, it's like one of
those parallel universes. Yeah.
514
00:30:37,400 --> 00:30:39,360
. But no, but
joking apart, you know,
515
00:30:39,580 --> 00:30:44,320
not too long ago you could probably
speak to people who would say, you know,
516
00:30:44,320 --> 00:30:46,720
you just have to respect
the people in authority.
517
00:30:46,870 --> 00:30:49,880
They are the people who have the, the.
518
00:30:49,880 --> 00:30:52,560
Truth. Yeah, definitely not so
Sony in my lifetime. Absolutely.
519
00:30:52,650 --> 00:30:55,640
We've gone from that, you
know, authorities there,
520
00:30:55,700 --> 00:31:00,360
you don't dare speak truth to
power. What now everybody is,
521
00:31:00,820 --> 00:31:03,480
is questionable at every level. And.
522
00:31:03,740 --> 00:31:08,000
Is there a sense in which these
tools might be used to actually
523
00:31:08,940 --> 00:31:10,040
divine what is true,
524
00:31:10,100 --> 00:31:14,000
to actually be able to
sift through the internet,
525
00:31:14,420 --> 00:31:17,960
to sift through the, the sum of human
knowledge, the sum of understanding.
526
00:31:18,500 --> 00:31:23,320
Is there a some something that
this AI could do to find the
527
00:31:23,420 --> 00:31:26,240
ans the big, the answers to the
big questions and find the truth.
528
00:31:26,500 --> 00:31:30,880
And come, finally come up with the
answer of 42, uh, .
529
00:31:33,020 --> 00:31:37,160
But The question is, okay,
530
00:31:37,160 --> 00:31:38,960
so how would we quantify that? How would,
531
00:31:39,020 --> 00:31:43,240
how do we quantify the truth after all
these things are just processing numbers,
532
00:31:43,450 --> 00:31:45,360
looks like words, but
really it's all numbers.
533
00:31:45,820 --> 00:31:49,960
So how do we quantify the truth? You know,
534
00:31:50,500 --> 00:31:54,360
how do we quantify the truth in science
is, is an interesting question as well.
535
00:31:54,490 --> 00:31:57,530
Cause a big element of that is, you know,
536
00:31:58,270 --> 00:32:03,210
and one of the things I used to cover in
this fourth year module is the blessed
537
00:32:03,320 --> 00:32:05,690
fineman. I'm a big fan of fineman, but he,
538
00:32:05,710 --> 00:32:08,850
it is something of a cliche for a
physicist to wheel out the fineman in.
539
00:32:08,950 --> 00:32:10,810
But one of the,
540
00:32:11,030 --> 00:32:15,810
the more famous Fomen quotes
is he was asked to define
541
00:32:15,960 --> 00:32:19,450
what science is to a, the believe
of conference of school teachers.
542
00:32:19,670 --> 00:32:23,250
And his one sentence definition was
science is the belief in the ignorance of
543
00:32:23,250 --> 00:32:26,210
experts. Which is sort of gives
us all a warm fuzzy feeling,
544
00:32:26,310 --> 00:32:29,850
yet it's all about being objective and
it's a, but that's not how science works.
545
00:32:30,150 --> 00:32:30,510
You know,
546
00:32:30,510 --> 00:32:34,970
the first few paragraphs or indeed in
some cases the first few pages of a
547
00:32:34,970 --> 00:32:36,890
scientific paper are what's come before.
548
00:32:37,390 --> 00:32:39,640
And it's really about building
on what's come before.
549
00:32:39,640 --> 00:32:42,440
And a big element of
that truth is consensus.
550
00:32:43,700 --> 00:32:47,680
So you could argue that you train
these things to look for consensus,
551
00:32:47,860 --> 00:32:51,920
but of course then the issue is consensus
can certainly build about something
552
00:32:51,920 --> 00:32:55,040
that's complete nonsense . So it's,
553
00:32:55,140 --> 00:32:57,880
it has to come back to empirical evidence,
554
00:32:58,020 --> 00:33:02,520
it has to come back to empirical dead
and it has to come back to the weight of
555
00:33:02,520 --> 00:33:04,430
the evidence. And then,
556
00:33:04,530 --> 00:33:07,510
and that's certainly how science works
so people can have their opinions but it
557
00:33:07,510 --> 00:33:08,750
always comes back to the experiment.
558
00:33:08,810 --> 00:33:11,070
People can of course
screw up the experiment,
559
00:33:11,490 --> 00:33:14,910
but if a hundred people or a hundred
groups do the experiment and they get the
560
00:33:14,910 --> 00:33:19,710
same result, that's getting us
closer to the truth. We never, ever,
561
00:33:19,860 --> 00:33:24,230
ever get to what the truth is
capital T in both cases. Um,
562
00:33:24,690 --> 00:33:28,790
but it gets us closer.
Could that be built in,
563
00:33:29,510 --> 00:33:33,110
I think in the case of sort of
analyzing scientific data? Yeah,
564
00:33:33,610 --> 00:33:38,070
in terms of analyzing
political data? Yeah.
565
00:33:42,910 --> 00:33:46,630
I wanted to get the perspective
of somebody who's currently
a student with their
566
00:33:46,630 --> 00:33:48,590
career in front of them, his corral green,
567
00:33:48,880 --> 00:33:53,110
who's a regular contributor to the
physics world magazine and website.
568
00:33:53,770 --> 00:33:56,430
I'm uh, an astrophysics PhD student.
569
00:33:56,810 --> 00:33:59,070
I'm doing my PhD at the
University of Nottingham.
570
00:33:59,470 --> 00:34:04,350
I study like black holes and galaxy
evolution in the most simplest
571
00:34:04,400 --> 00:34:05,140
terms.
572
00:34:05,140 --> 00:34:09,390
There's like a specific type that I'm
looking at and I'm trying to like see what
573
00:34:09,390 --> 00:34:13,030
they do to the galaxies that they're in
and maybe how they're triggered in the
574
00:34:13,030 --> 00:34:16,430
first place. Cause we're not a hundred
percent sure. So they're called um,
575
00:34:16,610 --> 00:34:21,470
active galactic nuclei or a G N for
sure. That's what everyone calls them.
576
00:34:22,050 --> 00:34:26,350
And basically in the middle of every
galaxy you have a super massive back hole
577
00:34:26,490 --> 00:34:30,510
and that's what all like the stars and
the dust and the soda systems and things
578
00:34:30,510 --> 00:34:32,950
like that orbit around. Um,
579
00:34:33,170 --> 00:34:38,150
and if just for whatever reason
there's enough dust close enough to
580
00:34:38,210 --> 00:34:39,710
the center of a galaxy,
581
00:34:39,710 --> 00:34:44,670
it can start falling onto the black
hole and all of the gravity from like
582
00:34:44,690 --> 00:34:49,070
the densest object in the universe
basically causes it to spew out a load of
583
00:34:49,070 --> 00:34:49,903
energy.
584
00:34:50,090 --> 00:34:54,310
So some galaxies look like they have a
massive bright point in the middle of
585
00:34:54,310 --> 00:34:57,750
them where obviously most of
them, the black hole's invisible.
586
00:34:58,410 --> 00:35:00,590
So you're looking out in the
universe and you're just like,
587
00:35:00,590 --> 00:35:03,230
why is that one really bright?
What's sort of going on there?
588
00:35:03,770 --> 00:35:06,470
And essentially I'm looking at
those galaxies and being like,
589
00:35:06,530 --> 00:35:08,910
why are they bright? You know,
590
00:35:09,330 --> 00:35:12,670
what's caused this in the first place
and is it doing anything to the rest of
591
00:35:12,670 --> 00:35:13,670
the galaxy that it's in?
592
00:35:14,210 --> 00:35:15,710
So Okay. Are you finding anything?
593
00:35:16,610 --> 00:35:19,430
I'm finding, I'm finding some
stuff. I'm finding some stuff.
594
00:35:19,690 --> 00:35:24,470
But the science process you like you
can't claim anything until you've had it
595
00:35:24,770 --> 00:35:29,070
looked at a lot. So got some
ideas but we're getting there.
596
00:35:29,730 --> 00:35:32,310
You said we don't really
know how they form,
597
00:35:32,750 --> 00:35:34,790
I thought we did know
how black holes form.
598
00:35:35,050 --> 00:35:39,600
We dunno why that type of like process
is triggered in the first place.
599
00:35:40,220 --> 00:35:45,040
So we don't know how all of the sort
of like dust and things started falling
600
00:35:45,070 --> 00:35:47,240
onto the black hole. Um,
601
00:35:47,300 --> 00:35:51,280
and why it isn't happening to every galaxy
or why they exist in the first place.
602
00:35:51,300 --> 00:35:56,080
So that's what's going on. Um, we know
why black holes form mostly, mostly.
603
00:35:56,510 --> 00:35:58,440
Does Ai does chat G P T,
604
00:35:58,440 --> 00:36:00,880
does any of this stuff come into
what you are doing at the moment?
605
00:36:01,420 --> 00:36:04,680
So I don't use it uh,
in my work specifically,
606
00:36:04,980 --> 00:36:08,840
but it has started coming up actually
quite a lot. Just in general. Um,
607
00:36:08,840 --> 00:36:13,000
in my department we actually had a
seminar the other week all a all about it,
608
00:36:13,350 --> 00:36:17,440
like what it could possibly mean for
the future and things like that. Um,
609
00:36:17,940 --> 00:36:18,560
cuz uh,
610
00:36:18,560 --> 00:36:23,440
things like this sort of sort of
common waves cuz prior to this the
611
00:36:23,460 --> 00:36:27,720
new sort of like shiny thing is machine
learning that a lot of people are now
612
00:36:27,960 --> 00:36:32,160
actually using in their research.
Um, so it was thinking about like ah,
613
00:36:32,780 --> 00:36:34,680
in, you know, in the next like five,
614
00:36:34,700 --> 00:36:37,800
10 years are people gonna be
just using AI as you know,
615
00:36:37,800 --> 00:36:40,920
as much as they'd use
any kind of coding, um,
616
00:36:42,430 --> 00:36:43,320
forgotten the word,
617
00:36:43,350 --> 00:36:48,320
sort of any anything that they use to
code or are they going to try and like cut
618
00:36:48,320 --> 00:36:49,800
it out completely and things like that.
619
00:36:50,170 --> 00:36:53,440
Which has been quite interesting
to see happen live almost.
620
00:36:53,900 --> 00:36:57,840
In the seminar. Is are people concerned
about it? Is there excitement about it?
621
00:36:58,100 --> 00:37:01,760
Uh, it's definitely like a bit
of both. There is the concern,
622
00:37:02,020 --> 00:37:05,080
the big thing that I sort of
noticed straight away is how um,
623
00:37:05,270 --> 00:37:09,080
with sort of undergraduates
and their work and things,
624
00:37:09,360 --> 00:37:12,080
there's already been, there's like a um,
625
00:37:12,270 --> 00:37:17,040
sort of plagiarism checker thing called
Turn It In that a lot of undergraduates
626
00:37:17,040 --> 00:37:20,680
might have heard of and I've heard how
that's already been like updated to
627
00:37:20,680 --> 00:37:24,520
recognize like chat G P T
and things like that. Um,
628
00:37:25,300 --> 00:37:27,350
so there is uh,
629
00:37:27,350 --> 00:37:32,190
worries about actually like assessing
people and seeing like what do they know,
630
00:37:32,190 --> 00:37:36,830
what don't they know? But then there's
also really like how useful it could be.
631
00:37:36,830 --> 00:37:40,510
Like people are seeing how it could
be, um, really helpful in the future.
632
00:37:40,510 --> 00:37:43,230
Especially with like
writing papers and writing,
633
00:37:43,230 --> 00:37:45,990
especially the introduction
of papers where you are,
634
00:37:46,450 --> 00:37:50,710
you can't plagiarize other people but
you have to say essentially the same
635
00:37:50,710 --> 00:37:52,190
things in a different way anyway.
636
00:37:52,810 --> 00:37:57,430
So like for my work I'd have to
define what a black hole is and
637
00:37:57,790 --> 00:37:59,870
motivate the reason for studying it.
638
00:37:59,890 --> 00:38:02,310
But I'm not the first person
ever to study a black hole.
639
00:38:02,890 --> 00:38:05,030
So there is like words of just like,
640
00:38:05,030 --> 00:38:07,750
oh why don't you try
like writing it yourself,
641
00:38:08,120 --> 00:38:10,710
doing things like that and then like, um,
642
00:38:10,710 --> 00:38:14,510
putting it into chat G P T and seeing if
it can help you write it in a bit of a
643
00:38:14,510 --> 00:38:18,870
like more concise way and things like
that, which is really interesting to see.
644
00:38:19,250 --> 00:38:21,670
So I mean are you using it
yourself in that way or?
645
00:38:22,030 --> 00:38:25,150
Hmm. I, so I haven't yet
but I think I'm going to,
646
00:38:25,270 --> 00:38:27,310
I think I'm definitely going to um,
647
00:38:27,380 --> 00:38:31,190
I've definitely got colleagues of mine
who've used it to just think of titles
648
00:38:31,190 --> 00:38:32,230
for their papers as well.
649
00:38:32,370 --> 00:38:37,170
That's one sort of hard thing
trying to make it relevant to your
650
00:38:37,170 --> 00:38:40,610
field but actually have it be catchy so
people are likely to read it and things
651
00:38:40,610 --> 00:38:44,970
like that. Um, and as the
sort of saying always goes,
652
00:38:45,260 --> 00:38:48,250
especially in like the
more mathematical sciences,
653
00:38:48,250 --> 00:38:50,170
we're not necessarily
that great at writing.
654
00:38:50,870 --> 00:38:55,010
So spent years and years learning all
this hard maths and then suddenly you've
655
00:38:55,010 --> 00:38:57,810
gotta now start writing
papers which can be, you know,
656
00:38:57,830 --> 00:39:01,930
any kind of length but you know,
10, 15 pages worth of writing is,
657
00:39:02,120 --> 00:39:06,810
doesn't come naturally to most of us.
So it's very helpful in being like,
658
00:39:06,810 --> 00:39:09,210
well this is the hardest bit, can we,
659
00:39:09,270 --> 00:39:13,930
can we get an AI to sort of
supplement the help with it and how
660
00:39:14,150 --> 00:39:18,610
how much can we, you know, do before we
have to be careful about obviously, um,
661
00:39:18,610 --> 00:39:22,170
plagiarizing or writing things that maybe
aren't a hundred percent like correct
662
00:39:22,190 --> 00:39:23,970
in the way that we mean
and things like that.
663
00:39:24,510 --> 00:39:29,130
So I specifically would use it to help
me like with the actual like language and
664
00:39:29,180 --> 00:39:29,870
prose,
665
00:39:29,870 --> 00:39:34,650
but when I'm using citations I need
to specifically go through other like
666
00:39:35,040 --> 00:39:38,650
peer reviewed papers that are
published formally read those,
667
00:39:38,650 --> 00:39:41,410
understand what they're saying and
be like, yes, that's what I mean,
668
00:39:41,440 --> 00:39:44,130
this paper backs it
up, I'm using that. Um,
669
00:39:44,270 --> 00:39:47,050
cuz something that's actually
very interesting is that like, uh,
670
00:39:47,050 --> 00:39:51,730
just in general we've had a play
around with it and say like, okay this,
671
00:39:51,910 --> 00:39:54,690
you know, this is a
fact the sun is a star,
672
00:39:54,880 --> 00:39:59,330
find like a a paper that
like I could use to cite it
673
00:39:59,750 --> 00:40:04,570
and it can spit out a link that looks
like a paper citation but if you
674
00:40:04,570 --> 00:40:06,370
click it it either goes to nowhere,
675
00:40:06,560 --> 00:40:10,650
it's like a fake link or it goes to
a paper that's completely different.
676
00:40:11,110 --> 00:40:14,610
So sort of recognize the kind of links
that you want if you're looking at
677
00:40:14,810 --> 00:40:18,850
academic sources. But it's not
like actually understanding them,
678
00:40:18,850 --> 00:40:22,170
it just knows how to sort of
mirror it on a page. So yeah,
679
00:40:22,170 --> 00:40:26,970
you'd very much not want to use it to
as a citation manager or anything like
680
00:40:26,970 --> 00:40:28,330
that. Yes. Yeah. Okay.
681
00:40:28,670 --> 00:40:32,850
It, I'm tempted to think of it as
a sort of a calculator for writing.
682
00:40:32,920 --> 00:40:36,730
Like we've had calculators for maths
for a very long time, you know,
683
00:40:36,730 --> 00:40:38,090
people doing maths have had Yeah.
684
00:40:38,090 --> 00:40:42,210
Calculators and computers to help them
with that maths and this is really a
685
00:40:42,210 --> 00:40:44,330
calculator in a way to
help you with writing.
686
00:40:45,110 --> 00:40:48,290
The difference is a calculator
is as far as I know always right,
687
00:40:48,750 --> 00:40:51,490
but church isn't always
right with its writing.
688
00:40:52,000 --> 00:40:54,170
Yeah, exactly. And um,
689
00:40:54,980 --> 00:40:57,970
other things as well is
that it, it can help,
690
00:40:58,290 --> 00:41:02,330
I I think it's very much leaning towards
it can help with writing but you can't
691
00:41:02,330 --> 00:41:05,570
really use it to do the science. Um,
692
00:41:05,620 --> 00:41:09,850
we've seen it like we, there are
other more sort of specialized
693
00:41:11,720 --> 00:41:14,210
UMIs that maybe do a little
bit better, but you could be,
694
00:41:14,210 --> 00:41:18,810
you can't go and be like, oh can you code
me up uh, a code that does x, y and z?
695
00:41:19,430 --> 00:41:23,670
Um, like it can do a few things
very in like a sort of simple way,
696
00:41:23,690 --> 00:41:25,190
but you do have to go in and be like,
697
00:41:25,190 --> 00:41:29,190
okay it's not quite getting that right
but I can go in and like edit this a bit
698
00:41:29,190 --> 00:41:30,023
and things like that.
699
00:41:30,050 --> 00:41:34,320
So it hasn't gotten very much
to the sort of like coding
700
00:41:34,850 --> 00:41:36,160
mathi side yet,
701
00:41:36,580 --> 00:41:41,560
but I think it's well on its way cuz
I know that it has access to the um,
702
00:41:42,000 --> 00:41:45,720
anything that's on Git, which is where
you can upload like code two for free.
703
00:41:45,960 --> 00:41:46,793
Anything that's um,
704
00:41:46,900 --> 00:41:50,560
public it's got like access to
and is sort of like learning from.
705
00:41:51,180 --> 00:41:56,120
So we're not that far away from it
sort of getting there. Very much so.
706
00:41:56,910 --> 00:41:59,880
Yeah, I, my my feeling is I
don't think it's gonna go away.
707
00:42:00,240 --> 00:42:05,240
I think it's definitely going to become
a part of like science and academia and
708
00:42:05,240 --> 00:42:09,120
things like that, but it's gonna
have to be done in a careful way.
709
00:42:09,140 --> 00:42:13,320
So you're not just, if you are not
just going in and being like, okay,
710
00:42:13,320 --> 00:42:16,840
write me a code that does
that, um, here's the data, uh,
711
00:42:16,840 --> 00:42:18,720
it's done it all perfectly. Um,
712
00:42:19,180 --> 00:42:23,600
let me go ahead with it in that way
cuz then it's like have you done it?
713
00:42:23,660 --> 00:42:25,840
Can you claim that you've done that? But.
714
00:42:25,840 --> 00:42:27,560
What's to stop people doing that? Is it,
715
00:42:27,560 --> 00:42:32,520
I mean is that a concern for you
that effectively it might remove the
716
00:42:32,550 --> 00:42:34,360
need for people to do coding?
717
00:42:34,670 --> 00:42:38,720
Yeah, yeah I think that is sort of
like a concern that's coming up, um,
718
00:42:39,050 --> 00:42:40,880
being like, oh no, it's
gonna sort of take,
719
00:42:41,430 --> 00:42:46,080
take people's positions and
jobs and things like that
and you'd have a room full
720
00:42:46,080 --> 00:42:50,520
of people typing into chat G P t instead
of a room full of people typing into
721
00:42:51,000 --> 00:42:55,840
whatever IDE they're using and things
like that. But um, maybe I have,
722
00:42:55,850 --> 00:42:59,840
maybe I have a bit of a cynical
approach but I also think yes,
723
00:42:59,840 --> 00:43:02,000
this is all happening. Yes this is all um,
724
00:43:02,390 --> 00:43:07,320
sort of open source and new and shiny
and people are having a look and going
725
00:43:07,380 --> 00:43:11,160
in but you like it is made by a company.
726
00:43:11,750 --> 00:43:16,560
They aren't going to just like have
it completely go free reign and things
727
00:43:16,560 --> 00:43:21,360
like that. I feel like monetization
for it is coming fairly quickly.
728
00:43:21,720 --> 00:43:24,960
I don't know if you'd get
away with just being like, oh,
729
00:43:24,960 --> 00:43:29,040
instead of hiring , you know,
a master student or PhD students,
730
00:43:29,040 --> 00:43:33,000
I'll get somebody to write this
into chat g PT for me in the future.
731
00:43:33,380 --> 00:43:38,280
So I, I'm not sure how how far it's
gonna get before stuff starts coming up.
732
00:43:38,280 --> 00:43:39,360
Like, oh pay us,
733
00:43:39,660 --> 00:43:43,600
pay us to use it or you only
get three free searches today,
734
00:43:43,690 --> 00:43:46,200
audit and things like that. Mm-hmm.
735
00:43:46,240 --> 00:43:48,800
. Oh yeah. So are
you, but are you seeing, you know,
736
00:43:48,930 --> 00:43:53,600
other PhD students, other undergraduates
using it in a way which concerns you?
737
00:43:53,960 --> 00:43:54,140
I.
738
00:43:54,140 --> 00:43:55,960
Do believe that that
is starting to happen.
739
00:43:56,600 --> 00:44:01,360
I do think people are definitely
getting homework questions or coursework
740
00:44:01,680 --> 00:44:02,470
questions,
741
00:44:02,470 --> 00:44:06,400
copying and pasting it into chat e p t
and being like answer this question and
742
00:44:06,400 --> 00:44:09,920
then just taking out wherever it
says whether they've done it or not.
743
00:44:10,580 --> 00:44:12,240
And that is, that is a problem,
744
00:44:12,350 --> 00:44:15,200
that is a problem for universities
because you know, you need to,
745
00:44:15,580 --> 00:44:19,240
you do need to know these things if you
want to get the qualification that they
746
00:44:19,480 --> 00:44:21,160
are offering you. Um,
747
00:44:21,740 --> 00:44:26,460
and it's hard because you can't
always know what the answer will be
748
00:44:26,720 --> 00:44:29,900
cuz right now maybe it's quite in its
infancy and you can be like, okay,
749
00:44:29,900 --> 00:44:34,560
it spits out the sort of five different
versions of this and those are the ones
750
00:44:34,560 --> 00:44:38,880
that we know have come from Chap G P T
and so they have to be like filtered out
751
00:44:39,260 --> 00:44:41,640
but it won't, it's very quickly, you know,
752
00:44:41,840 --> 00:44:46,800
evolving beyond that and it might
it might get a lot harder to do v
753
00:44:46,900 --> 00:44:48,080
uh, very soon. I.
754
00:44:48,180 --> 00:44:52,560
My my understanding is that quite soon
it will be have an update where it's
755
00:44:52,560 --> 00:44:56,240
gonna start being able to check and
the veracity of what it's saying in,
756
00:44:56,300 --> 00:44:57,840
in the next additional.
757
00:44:58,310 --> 00:45:03,120
Yeah, yeah. I I've, I do
think that is coming and yeah,
758
00:45:03,120 --> 00:45:06,840
it's, it's not, it's not gonna
be a trivial thing to sort of,
759
00:45:07,340 --> 00:45:10,880
to sort of manage because you know, we're,
760
00:45:11,140 --> 00:45:13,880
we are scientists, we are
supposed to, you know,
761
00:45:14,010 --> 00:45:17,600
enjoy innovation and things
like that and there is,
762
00:45:18,360 --> 00:45:23,160
I can't really see a a an ultra
valid reason to not use it.
763
00:45:23,220 --> 00:45:27,920
For example, like I said in
paper intros where you are like,
764
00:45:28,070 --> 00:45:33,000
okay these six papers are all
saying different permutations of the
765
00:45:33,000 --> 00:45:35,840
same thing and I need to be the
seventh paper that says the different
766
00:45:35,840 --> 00:45:37,120
permutation of the same thing.
767
00:45:37,980 --> 00:45:42,960
But then it's that versus I'm gonna
use this to like do my research or
768
00:45:43,020 --> 00:45:47,200
get agree and you know, not
actually do any of it myself.
769
00:45:47,460 --> 00:45:52,440
So the knowledge is not there and
therefore can you like claim fully
770
00:45:52,510 --> 00:45:56,440
that you've uh, done that and
things like that. So it's a,
771
00:45:56,510 --> 00:45:59,400
it's it's, it's gonna be
difficult cuz you wanna,
772
00:45:59,900 --> 00:46:03,800
you wanna take advantage of the good
parts of it without just relying on it
773
00:46:03,800 --> 00:46:05,040
completely kind of thing.
774
00:46:05,390 --> 00:46:06,140
Yeah.
775
00:46:06,140 --> 00:46:09,920
So what what you want is humanity to look
at something and take the best from it
776
00:46:09,980 --> 00:46:11,440
and, and leave the rest behind.
777
00:46:12,140 --> 00:46:16,920
Yep. Yep. If only , if only
we could do that , oh god,
778
00:46:17,400 --> 00:46:17,880
.
779
00:46:17,880 --> 00:46:21,560
I mean so sure certain aspects of society
will do that but it, it, you know,
780
00:46:22,580 --> 00:46:25,720
we can't just hope that that will
happen. Right. And, and yeah,
781
00:46:26,080 --> 00:46:30,320
I wonder whether there's anything
that could be done or do we need to do
782
00:46:30,520 --> 00:46:33,560
anything from your point of view to, um,
783
00:46:33,770 --> 00:46:38,760
apart from monetizing it and
restricting access to it,
784
00:46:38,780 --> 00:46:41,560
is there anything that
needs to be done to,
785
00:46:41,980 --> 00:46:46,000
to stop those people
taking advantage of the,
786
00:46:47,780 --> 00:46:52,240
the ease that it brings
or the possibly less
787
00:46:53,080 --> 00:46:54,560
pleasant ways of using it?
788
00:46:54,830 --> 00:46:59,360
Yeah, I, I, hmm, I don't
know cuz it's like,
789
00:46:59,660 --> 00:47:02,760
um, it's gotta be mixed with the,
790
00:47:02,900 --> 00:47:07,160
the bureaucracy even of like
higher education in universities.
791
00:47:07,260 --> 00:47:11,880
Cuz most academia in, at least that
I know is done in universities.
792
00:47:12,540 --> 00:47:15,360
Yes there's things like ESO and
stuff like that but like, you know,
793
00:47:15,360 --> 00:47:16,440
if you look around the globe,
794
00:47:16,440 --> 00:47:21,200
they're all just people lecturers in
different universities working on research
795
00:47:21,980 --> 00:47:26,000
and like trying to sort of
blend it through cuz you know,
796
00:47:26,000 --> 00:47:28,560
there's been loads of
innovations we've gone from like,
797
00:47:28,670 --> 00:47:33,240
just even in my lifetime you've gone to
smartphones then I was doing exams and
798
00:47:33,240 --> 00:47:37,280
they were just like, smartphones and
smart watches need to be off. Um,
799
00:47:37,540 --> 00:47:42,320
and then like covid happened and they
had to redo all of exams so people could
800
00:47:42,320 --> 00:47:43,153
do them at home.
801
00:47:43,700 --> 00:47:48,080
So obviously the questions became very
different and also a lot harder because
802
00:47:48,180 --> 00:47:48,880
you know,
803
00:47:48,880 --> 00:47:53,080
a lot of them you could like Google
prior to Covid and they're just like,
804
00:47:53,080 --> 00:47:54,960
well if they're at home they
can't just Google the answers.
805
00:47:55,040 --> 00:47:59,160
I need like more long form
thinking from them I guess.
806
00:48:00,020 --> 00:48:02,920
Um, and, and now the
next one is like with,
807
00:48:02,920 --> 00:48:07,800
with AI and just typing something
into literally just a chat
808
00:48:07,820 --> 00:48:11,520
bot and getting an
answer out and deciding,
809
00:48:11,670 --> 00:48:15,760
well actually in this
case it's fine and in,
810
00:48:15,900 --> 00:48:20,680
and in this case we can't really
like use it very much but I
811
00:48:20,680 --> 00:48:22,040
dunno like what, what,
812
00:48:22,270 --> 00:48:26,000
what could you do if it gets to like,
813
00:48:26,000 --> 00:48:29,000
it gets in intergen enough that
you could sit down and be like,
814
00:48:29,000 --> 00:48:32,080
okay here's like some data I have, uh,
815
00:48:32,310 --> 00:48:36,640
code up and analyze it and then tell
me if there are any differences or
816
00:48:36,680 --> 00:48:38,200
similarities or whatever else.
817
00:48:38,580 --> 00:48:41,760
And if it produces a reasonable piece
of work and you've gone through and
818
00:48:41,760 --> 00:48:45,680
checked it, then would
you not use it, you know?
819
00:48:46,340 --> 00:48:48,680
Mm-hmm. ,
well you, yeah, it's,
820
00:48:49,070 --> 00:48:51,280
it's a hard thing to say that
you wouldn't use it, isn't.
821
00:48:51,280 --> 00:48:54,880
It? Yeah. Yeah, because I, cuz
that would, that would be like,
822
00:48:54,880 --> 00:48:58,760
well what are we doing if we
didn't use it actually so, mm.
823
00:48:59,140 --> 00:49:04,000
But if you were considering
the world before smartphones as
824
00:49:04,000 --> 00:49:07,280
you say, and how smartphones
have changed the world,
825
00:49:07,300 --> 00:49:11,920
do you think chat G P T and other things
are gonna change the world in a similar
826
00:49:12,060 --> 00:49:12,740
way.
827
00:49:12,740 --> 00:49:15,720
In the way that it just
becomes very sort of common?
828
00:49:16,160 --> 00:49:19,920
I can very easily imagine it
in search engines and like,
829
00:49:20,200 --> 00:49:23,000
I just don't think search engines
are that different from it, right.
830
00:49:23,190 --> 00:49:26,320
Whether you like it or not,
it seems to be here to stay.
831
00:49:26,800 --> 00:49:31,640
I think it's gonna just become more
and more used in different things and
832
00:49:31,640 --> 00:49:33,440
you'll have to adapt to deal with it.
833
00:49:33,480 --> 00:49:36,440
I think I'm wary of it but I
don't think it's gonna go away.
834
00:49:36,860 --> 00:49:41,680
And I'm interested to see how
it's dealt with because as
835
00:49:41,680 --> 00:49:46,160
much as it's sort of coming into
our lives, I'm not like a, you know,
836
00:49:46,280 --> 00:49:47,720
a head of school,
837
00:49:47,880 --> 00:49:52,800
a head of department or head of
like science at a university or in
838
00:49:52,800 --> 00:49:54,160
like a massive collaboration.
839
00:49:54,860 --> 00:49:59,040
And so I think they are thinking about
it and they've def I've definitely seen
840
00:49:59,040 --> 00:50:01,520
it. I definitely know undergrad
to use it. I mean I would,
841
00:50:02,180 --> 00:50:04,800
if I was still doing an undergrad
degree, I'd be like, damn,
842
00:50:05,190 --> 00:50:06,160
this is gonna help me.
843
00:50:06,220 --> 00:50:09,360
Let me open up a chat g p t
account and see what I can do.
844
00:50:09,940 --> 00:50:14,490
So I wanna see how it sort of goes in
the next couple years and if they'll ever
845
00:50:14,490 --> 00:50:18,370
embrace it or sort of try and
filter it out as much as possible.
846
00:50:19,130 --> 00:50:23,930
I, i if chat G B T and others
evolve to the point or are
847
00:50:23,930 --> 00:50:27,440
updated to the point where
they can do essentially the,
848
00:50:28,100 --> 00:50:31,760
the work that you want, the science,
the coding, the what, the equations,
849
00:50:31,960 --> 00:50:34,160
whatever it is that you are
doing for your research,
850
00:50:35,230 --> 00:50:37,320
what role do you see for yourself in.
851
00:50:37,320 --> 00:50:41,480
That? I think it's
difficult because if a a,
852
00:50:41,520 --> 00:50:45,320
a department or a collaboration
or something like that re like
853
00:50:46,070 --> 00:50:51,040
changes the roles so that they get rid
of all their PhD students and replace
854
00:50:51,040 --> 00:50:52,200
them with chat ai.
855
00:50:52,840 --> 00:50:57,720
I do think that they would be fine
for like a while but then I think
856
00:50:57,720 --> 00:51:00,440
they just run into the problem
that's kind of happening already,
857
00:51:00,660 --> 00:51:02,840
but a lot faster in that
like there's a lot of a,
858
00:51:03,360 --> 00:51:08,360
a mass in academia sort of finish your
PhD, don't want to become a postdoc,
859
00:51:09,010 --> 00:51:13,080
don't want to become a
lecturer for whatever various
reasons there are and then
860
00:51:13,080 --> 00:51:17,640
you'll sort of run out of people
to, to to keep things going.
861
00:51:18,340 --> 00:51:19,060
Um,
862
00:51:19,060 --> 00:51:23,440
so I don't think it's going to
literally just get rid of us completely,
863
00:51:24,140 --> 00:51:27,640
but I do think they need to be
careful about it cuz you know,
864
00:51:27,640 --> 00:51:30,520
at the end of the day you need somebody
to lecture, you need somebody to mark,
865
00:51:30,990 --> 00:51:35,640
need people to teach. And
if you aren't ,
866
00:51:35,640 --> 00:51:38,560
if you haven't got PhD students
to become those people,
867
00:51:38,590 --> 00:51:39,960
then you're just gonna lose 'em all.
868
00:51:40,500 --> 00:51:43,880
If you could ask AI at,
869
00:51:43,940 --> 00:51:47,200
at any form to answer
any question in physics,
870
00:51:48,310 --> 00:51:49,200
what would you ask it?
871
00:51:49,540 --> 00:51:53,560
Oh, that is hard. .
That is very hard.
872
00:51:54,360 --> 00:51:58,120
I what initially comes to mind
is maybe something that's like,
873
00:51:58,310 --> 00:52:01,440
like Nobel Prize worthy. I think just for,
874
00:52:02,340 --> 00:52:07,060
for like my personal, for my, you know,
875
00:52:07,340 --> 00:52:10,580
personal curiosity. Um, but if I, if I,
876
00:52:10,680 --> 00:52:15,260
if I could ask it any questions and I
knew with like a hundred percent certainty
877
00:52:15,260 --> 00:52:16,093
it would be right.
878
00:52:16,460 --> 00:52:21,180
I think actually I'd ask it the like
most Vegas question that I could
879
00:52:21,180 --> 00:52:23,940
possibly think. So it could
just solve everything like, oh,
880
00:52:23,940 --> 00:52:28,380
what's the grand unified theory?
And then it just spits out ,
881
00:52:28,430 --> 00:52:30,820
everything that we're looking into. But,
882
00:52:31,560 --> 00:52:33,860
but I would only do that
after I finished my PhD.
883
00:52:34,220 --> 00:52:35,780
I want to be a doctor first ,
884
00:52:36,650 --> 00:52:40,980
then I'd like collapse it for the rest
of everybody else. , I'd wait,
885
00:52:41,180 --> 00:52:43,260
I would literally wait
until I'm done ,
886
00:52:45,960 --> 00:52:47,300
that's it. But that would be it.
887
00:52:47,660 --> 00:52:50,100
I put that same question to Phil Moti.
888
00:52:50,800 --> 00:52:53,340
Oh, what's the nature of the wave
function? What is the wave function?
889
00:52:54,240 --> 00:52:57,740
So we still, after a hundred
years or so of quantum mechanics,
890
00:52:57,740 --> 00:52:59,740
right at the core of it is this
thing called the wave function.
891
00:52:59,960 --> 00:53:01,460
So it's this mathematical object.
892
00:53:01,690 --> 00:53:04,060
It's not an experimental
observable as you say.
893
00:53:04,060 --> 00:53:08,380
And there's lots and lots of debates
between what's called the ontological camp
894
00:53:08,380 --> 00:53:11,780
without going too philosophical and
the epistemological camp in terms of,
895
00:53:12,000 --> 00:53:16,820
is the wave function telling us
about the nature of reality itself?
896
00:53:16,840 --> 00:53:21,660
Or is it telling us about her information
on the, on reality? What is the,
897
00:53:21,660 --> 00:53:24,260
the nature of the way function That
would, that would that, that would be it.
898
00:53:24,260 --> 00:53:24,900
Yeah.
899
00:53:24,900 --> 00:53:25,800
for me, ,
900
00:53:26,220 --> 00:53:30,040
if chat g p t wants to replace
physicists in the real world,
901
00:53:30,430 --> 00:53:33,360
it's gonna have to develop
that kind of curiosity.
902
00:53:34,820 --> 00:53:39,440
I'd like to thank Carol Green and Philip
Moti for joining me for this episode of
903
00:53:39,440 --> 00:53:40,140
the podcast.
904
00:53:40,140 --> 00:53:44,040
And I'd like to thank you for sticking
with me with my hay fever voice.
905
00:53:44,650 --> 00:53:48,400
We'll be back next month when hopefully
my voice will be somewhere back to
906
00:53:48,400 --> 00:53:50,960
normal. And we'll be
exploring Moore's Law.
907
00:53:51,500 --> 00:53:53,160
And thank you very much for listening.
908
00:53:58,030 --> 00:53:59,160
Physics world.