China’s AI Rulebook

WEBVTT
1
00:00:00.760 --> 00:00:04.599
The Tech Policy Design Institute acknowledges and pays our respects
2
00:00:04.799 --> 00:00:09.839
to all First Nations people. We recognise and celebrate that
3
00:00:09.900 --> 00:00:15.060
among many things, Indigenous people were Australia's first tech innovators.
4
00:00:17.440 --> 00:00:20.519
One of the things I think is so important to
5
00:00:20.640 --> 00:00:23.280
understand about Chinese tech policymaking is.
6
00:00:23.670 --> 00:00:26.190
Is the power of the process here.
7
00:00:26.410 --> 00:00:33.420
Guidance, recommendations, all kinds of policy instruments in an incredibly agile, flexible,
8
00:00:34.100 --> 00:00:37.840
fast way. It's about thinking, what are the problems that
9
00:00:37.880 --> 00:00:41.890
we have? Lifting 800 million people out of poverty over
10
00:00:41.929 --> 00:00:44.869
the last 25 years, that's a big problem.
11
00:00:45.549 --> 00:00:46.850
And there was a lot.
12
00:00:46.710 --> 00:00:51.159
Of technology that went into addressing that problem They don't
13
00:00:51.310 --> 00:00:54.479
use that language of a race with the United States.
14
00:01:01.700 --> 00:01:05.079
Hello and welcome to another episode of Tech Mirror. My
15
00:01:05.120 --> 00:01:08.049
name is Johanna Weaver and I'm your host. This is
16
00:01:08.090 --> 00:01:11.769
the podcast where we talk about how humans are shaping
17
00:01:11.849 --> 00:01:14.890
technology and how we can make sure that it's shaping
18
00:01:15.030 --> 00:01:17.909
a world that we all want to live in. And
19
00:01:17.950 --> 00:01:23.000
today my guest is Madeline Carr. Madeline is an Australian, importantly,
20
00:01:23.099 --> 00:01:26.099
but she is also a professor of global politics and
21
00:01:26.159 --> 00:01:30.640
cybersecurity at the University College London. And Madeline was one
22
00:01:30.659 --> 00:01:33.980
of the first people that I encountered when I started
23
00:01:34.040 --> 00:01:37.060
working in this field that was looking at the intersection
24
00:01:37.400 --> 00:01:42.719
of emerging technologies, cybersecurity and geopolitics. So Madeline is actually
25
00:01:42.769 --> 00:01:45.950
in part responsible for my career. It's wonderful to have
26
00:01:46.010 --> 00:01:47.370
you on the pod today, Madeline.
27
00:01:47.709 --> 00:01:49.030
Thank you so much, Joanna.
28
00:01:49.180 --> 00:01:52.780
And wonderful to learn that I had any part to
29
00:01:52.819 --> 00:02:00.209
play whatsoever in spawning the great Policy Design Institute.
30
00:02:00.250 --> 00:02:03.609
Yes, well, I mean, it was definitely, you were looking
31
00:02:03.650 --> 00:02:06.349
at these issues before they were cool. That is definitely
32
00:02:06.549 --> 00:02:10.659
fair to say. And you're currently in Beijing as the
33
00:02:10.699 --> 00:02:14.900
Visiting Chair of Technology at Swartzman's College at Qinghao University
34
00:02:15.060 --> 00:02:18.060
in Beijing, which for those who are not super familiar
35
00:02:18.099 --> 00:02:21.199
with Chinese universities, it's kind of the equivalent of Harvard
36
00:02:21.340 --> 00:02:24.129
University in the US. So it's like the most prestigious
37
00:02:24.229 --> 00:02:27.750
law university in What led me to reach out to you, Madeline,
38
00:02:27.789 --> 00:02:31.000
was an article that you had published in The Conversation
39
00:02:31.180 --> 00:02:35.439
about US tech leaders are urgently calling for rules for AI,
40
00:02:35.539 --> 00:02:38.800
but China already has them. And so we've invited you
41
00:02:38.840 --> 00:02:41.460
on today to talk about that article, to look at
42
00:02:41.659 --> 00:02:44.900
what China is doing in terms of regulating AI, what
43
00:02:45.000 --> 00:02:47.379
we should be learning from that, what we shouldn't be
44
00:02:47.460 --> 00:02:50.780
replicating in the Chinese model. We'll have a look from
45
00:02:50.819 --> 00:02:54.319
a geopolitical perspective and compare what the US and China
46
00:02:54.539 --> 00:02:57.740
are doing, and probably also touch on the concept of
47
00:02:57.780 --> 00:03:02.469
liability as well. But before we delve down into that
48
00:03:02.569 --> 00:03:05.349
level of detail, Madeline, can you just give us a
49
00:03:05.389 --> 00:03:08.909
bit of context in terms of recording this from Beijing?
50
00:03:09.550 --> 00:03:11.889
So what are you working on at the moment? How
51
00:03:11.969 --> 00:03:14.389
have you come to be sitting in Beijing on this podcast?
52
00:03:15.330 --> 00:03:16.490
Yes, how indeed.
53
00:03:16.509 --> 00:03:19.270
I mean, I guess I've been coming to China basically
54
00:03:20.090 --> 00:03:23.469
over the last decade or more for track 1.5 and
55
00:03:23.509 --> 00:03:28.409
track two dialogues between the UK and China on cybersecurity
56
00:03:28.590 --> 00:03:32.669
and cybercrime. But a few years ago, I was approached
57
00:03:32.969 --> 00:03:36.620
to join Schwarzman College as a visiting chair.
58
00:03:36.759 --> 00:03:38.319
And that has.
59
00:03:38.199 --> 00:03:42.699
Been just a really incredible opportunity for me because I
60
00:03:42.780 --> 00:03:45.280
spend sort of three or four months of every year
61
00:03:45.360 --> 00:03:50.139
here teaching in the program, but also participating in talks
62
00:03:50.219 --> 00:03:54.840
and forums and meeting Chinese colleagues. And, you know, for
63
00:03:54.879 --> 00:04:01.360
someone who's interested in tech policy and particularly Chinese tech policy,
64
00:04:01.479 --> 00:04:06.879
it's just so, so helpful to be here, to get
65
00:04:06.939 --> 00:04:10.099
a deeper understanding of how Chinese people and the Chinese
66
00:04:10.139 --> 00:04:17.110
government approaches technology, how they approach technology policy. and how
67
00:04:17.149 --> 00:04:21.129
they bring together their kind of objectives and ambitions into
68
00:04:21.170 --> 00:04:22.209
governance frameworks.
69
00:04:22.250 --> 00:04:24.269
Because for you and I, it's.
70
00:04:24.189 --> 00:04:30.199
That governance piece that is so, so powerful and so important.
71
00:04:31.139 --> 00:04:36.540
And yeah, it's been an extraordinary experience and privilege, I
72
00:04:36.579 --> 00:04:39.220
would say, to be able to spend so much time
73
00:04:39.300 --> 00:04:42.240
here understanding those things from a Chinese perspective.
74
00:04:42.870 --> 00:04:46.110
And I guess one thing that I always find fascinating
75
00:04:46.189 --> 00:04:49.370
is reflecting on the atmospherics of being in Beijing at
76
00:04:49.389 --> 00:04:52.230
the moment. There still aren't a lot of Australians in
77
00:04:52.269 --> 00:04:55.319
Beijing traveling. There are definitely more than there were a
78
00:04:55.360 --> 00:04:57.560
few years ago, but can you just help paint a
79
00:04:57.579 --> 00:05:00.120
picture of what it's like on the ground in Beijing
80
00:05:00.160 --> 00:05:03.220
right now, how technology is sort of interacting with day-to-day
81
00:05:03.259 --> 00:05:06.990
life before we get into the nerdy regulation part?
82
00:05:07.310 --> 00:05:08.389
Yeah, of course.
83
00:05:08.709 --> 00:05:11.029
I'm very fond of Beijing because I spend a lot
84
00:05:11.069 --> 00:05:14.389
of time here. It's true that there are a ton
85
00:05:14.430 --> 00:05:17.060
of Westerners here, but, you know, there are quite a
86
00:05:17.149 --> 00:05:22.529
few Western students at the universities on exchange programs. And
87
00:05:22.670 --> 00:05:25.870
I think less than there were before COVID. But one
88
00:05:25.889 --> 00:05:27.829
of the things that has made a big difference in
89
00:05:27.910 --> 00:05:32.149
China over the last year is they have visa-free travel
90
00:05:32.370 --> 00:05:35.100
for a lot of countries. So where it used to
91
00:05:35.129 --> 00:05:37.800
be quite painful to apply even for a tourist visa
92
00:05:37.829 --> 00:05:40.379
to China, you don't have to anymore. There's a visa
93
00:05:40.399 --> 00:05:42.459
waiver system in place. And I think that's made a
94
00:05:42.480 --> 00:05:45.699
big difference to the volume of people that come on
95
00:05:45.740 --> 00:05:49.439
a holiday to visit this place. In terms of just
96
00:05:49.579 --> 00:05:54.410
day-to-day interaction with technology, I mean, I just came back
97
00:05:54.490 --> 00:05:59.589
from a visit to quite a rural area in the southwest.
98
00:05:59.050 --> 00:06:00.930
Of China, Yunnan province.
99
00:06:01.329 --> 00:06:04.810
And even in a place like that, you know, if
100
00:06:04.870 --> 00:06:07.350
I wanted to buy a, you know, a banana from
101
00:06:07.439 --> 00:06:11.620
someone on a street side store, I could pay with
102
00:06:11.680 --> 00:06:16.500
my payment app. So the place is, it's incredibly integrated
103
00:06:17.459 --> 00:06:22.899
and wired up and connected for this kind of extreme
104
00:06:22.959 --> 00:06:27.160
level of convenience and sort of seamlessness that makes it, yeah,
105
00:06:27.319 --> 00:06:29.930
it's a very easy place in a strange way because
106
00:06:29.939 --> 00:06:32.410
the language is difficult for Westerners, but it's a very
107
00:06:32.610 --> 00:06:35.850
easy place. Once you have those elements in place, it's
108
00:06:35.889 --> 00:06:38.689
very easy to travel. And a colleague at dinner last
109
00:06:38.730 --> 00:06:41.680
night was telling me that she was out for dinner
110
00:06:41.720 --> 00:06:43.800
with a friend and he offered to give her a
111
00:06:43.819 --> 00:06:46.459
lift home. And he said, I have a car, I'll
112
00:06:46.519 --> 00:06:49.399
call my car. And he called it, this was in Beijing,
113
00:06:49.560 --> 00:06:52.250
and it was a driverless car that pulled up to
114
00:06:52.310 --> 00:06:54.490
take them home from dinner, which she was, you know,
115
00:06:54.529 --> 00:06:55.569
wildly excited about.
116
00:06:55.769 --> 00:06:56.009
Yeah.
117
00:06:56.149 --> 00:06:58.990
And he told her that car cost about 20,000 US dollars.
118
00:06:59.050 --> 00:07:03.189
So there's a lot of optimism about technology here that
119
00:07:03.250 --> 00:07:07.100
I don't see at home in the U.K., Or in
120
00:07:07.120 --> 00:07:10.019
the West generally, there's much more optimism about it, much
121
00:07:10.060 --> 00:07:12.860
more excitement about what it will do to improve the
122
00:07:12.899 --> 00:07:15.199
human condition rather than degrade it.
123
00:07:15.500 --> 00:07:18.910
And is there concern about, you know, when we think
124
00:07:18.970 --> 00:07:22.269
about technology and China, people often think, OK, yes, it
125
00:07:22.329 --> 00:07:25.629
gives that convenience, but it's also the sort of surveillance
126
00:07:25.709 --> 00:07:28.879
and monitoring that comes with it. Is that something that
127
00:07:28.939 --> 00:07:30.819
is there sort of on the surface?
128
00:07:31.040 --> 00:07:32.279
Is it overt, covert?
129
00:07:32.639 --> 00:07:34.759
How much does that play into conversations?
130
00:07:35.100 --> 00:07:38.399
I can't sort of speak for the 1.5 billion people here,
131
00:07:38.519 --> 00:07:44.089
but I don't encounter that in my day-to-day exchanges with people.
132
00:07:44.170 --> 00:07:48.329
They don't have that anxiety about surveillance. And it was
133
00:07:48.389 --> 00:07:52.050
really eye-opening when I came here because, of course, that's
134
00:07:52.089 --> 00:07:56.990
something that we are very alive to and sensitive about.
135
00:07:58.579 --> 00:08:02.810
I don't see that sensitivity here. And on the contrary,
136
00:08:04.129 --> 00:08:07.810
what I've come to recognize over the last four years
137
00:08:07.889 --> 00:08:11.509
or so of being embedded in this system, even if
138
00:08:11.569 --> 00:08:15.920
temporarily and sporadically, is that particularly when you look at
139
00:08:16.040 --> 00:08:20.980
issues like cybercrime, those things are much less of a problem.
140
00:08:21.660 --> 00:08:24.279
They don't have burner phones here, for example. If you
141
00:08:24.319 --> 00:08:26.939
want a phone, you have to show your ID. And
142
00:08:26.980 --> 00:08:29.310
that phone is linked to your ID and to your
143
00:08:29.350 --> 00:08:33.070
bank account. The flip side of that is, yes, there's
144
00:08:33.149 --> 00:08:36.990
not a lot of, you know, you can't do things secretly.
145
00:08:37.070 --> 00:08:37.789
That's true.
146
00:08:39.289 --> 00:08:41.090
But the flip side of that is that there's a
147
00:08:41.190 --> 00:08:44.429
kind of high level of personal accountability. So if something
148
00:08:44.470 --> 00:08:48.580
goes wrong, for example, if I should buy something in
149
00:08:48.620 --> 00:08:50.700
a shop and I see they have charged me twice,
150
00:08:51.340 --> 00:08:54.899
they would reverse that themselves when they saw it because
151
00:08:54.919 --> 00:08:58.110
they there would be no escaping that. You know, everything
152
00:08:58.169 --> 00:09:02.490
is traceable, everything is trackable. And from a kind of
153
00:09:02.649 --> 00:09:08.950
privacy perspective, we wouldn't probably tolerate that. But if you're
154
00:09:09.049 --> 00:09:15.610
focusing on security and crime prevention, we sort of in
155
00:09:15.669 --> 00:09:18.700
some ways look longingly, you know, at what can be
156
00:09:18.750 --> 00:09:21.700
done to prevent those kind of online harms.
157
00:09:22.220 --> 00:09:24.279
Things like bullying, online bullying.
158
00:09:23.960 --> 00:09:26.820
Is a absolutely can't happen because you have to put
159
00:09:26.840 --> 00:09:32.159
your name on your social media accounts and yeah just
160
00:09:32.419 --> 00:09:34.659
kind of something to be said for that I'm not
161
00:09:34.700 --> 00:09:38.370
suggesting we could transplant that to Australia or the UK
162
00:09:38.429 --> 00:09:40.730
not at all but it does show you the other
163
00:09:40.769 --> 00:09:41.370
side of it.
164
00:09:48.830 --> 00:09:52.750
Maybe let's talk about some of the laws first. So
165
00:09:52.769 --> 00:09:57.039
the article that you wrote took the view of following
166
00:09:57.179 --> 00:10:02.799
how China has progressively regulated artificial intelligence, not necessarily through,
167
00:10:02.820 --> 00:10:06.620
you know, one overarching piece of regulation, but progressively as
168
00:10:06.679 --> 00:10:09.539
the technology has evolved and kind of challenging this assumption
169
00:10:09.580 --> 00:10:12.809
that the technology is evolving too fast for regulation to
170
00:10:12.850 --> 00:10:17.259
keep up because actually China has progressively kept up. So
171
00:10:17.279 --> 00:10:19.919
can you take us through a bit of a timeline
172
00:10:19.980 --> 00:10:24.059
in terms of the key pieces of regulation or legislation
173
00:10:24.600 --> 00:10:28.259
that is governing artificial intelligence in China and what the
174
00:10:28.299 --> 00:10:32.950
particular parts that they are focusing on? But maybe first,
175
00:10:33.370 --> 00:10:35.629
if you can comment on that sequencing piece and the
176
00:10:35.669 --> 00:10:38.529
challenging that it's not possible to regulate and to keep up.
177
00:10:39.049 --> 00:10:40.049
Yeah, absolutely.
178
00:10:40.090 --> 00:10:44.200
I mean, I wrote that article out of frustration with
179
00:10:44.289 --> 00:10:48.879
this kind of global discourse that emerged after Dario Amode's post,
180
00:10:50.299 --> 00:10:56.539
which was basically premised on this notion that governments have
181
00:10:56.580 --> 00:10:58.799
to do something and they have to do something quickly
182
00:10:58.860 --> 00:11:05.679
to stop this out-of-control sector from causing harm. And having
183
00:11:05.779 --> 00:11:10.830
spent time here studying those Chinese instruments, I felt that
184
00:11:10.870 --> 00:11:15.870
there was a real lack of understanding of how we
185
00:11:15.879 --> 00:11:20.590
could say the other big AI superpower, is approaching those
186
00:11:20.870 --> 00:11:25.110
challenges and how they have done this incrementally. And, you know,
187
00:11:25.230 --> 00:11:28.570
I'll run through that sort of sequence of, you know,
188
00:11:28.649 --> 00:11:32.100
major pieces. But one of the things I think is
189
00:11:32.220 --> 00:11:37.940
so important to understand about Chinese tech policymaking is the
190
00:11:38.039 --> 00:11:42.480
power of the process here. And it's almost, I would
191
00:11:42.539 --> 00:11:47.750
say it's a asymmetric advantage that China is able to
192
00:11:48.309 --> 00:11:54.710
draft kind of guidance, recommendations, all kinds of policy instruments
193
00:11:55.190 --> 00:12:00.830
in an incredibly agile, flexible, fast way. They're very innovative,
194
00:12:01.149 --> 00:12:03.330
not just in terms of technology, but in terms of
195
00:12:03.389 --> 00:12:08.009
technology policy. They're very innovative in the way they experiment,
196
00:12:08.690 --> 00:12:13.389
in the way they will reverse things that don't work
197
00:12:13.529 --> 00:12:17.580
out It's really extraordinary to see what can be done
198
00:12:17.789 --> 00:12:23.200
with the right mechanisms in place. And again, I'm not
199
00:12:23.259 --> 00:12:26.419
suggesting that the way they do that can be picked
200
00:12:26.500 --> 00:12:29.480
up and transplanted to another place because there are all
201
00:12:29.500 --> 00:12:31.860
kinds of structural reasons why they can do that. But
202
00:12:31.899 --> 00:12:34.000
I do think there are some lessons to be learned
203
00:12:34.659 --> 00:12:37.679
or at least to be examined of how China does this.
204
00:12:38.460 --> 00:12:40.190
So basically....
205
00:12:40.570 --> 00:12:46.250
Back in 2022, they developed these provisions on algorithmic recommendations.
206
00:12:46.610 --> 00:12:51.509
And this kind of preceded AI, but it links into AI.
207
00:12:51.570 --> 00:12:57.029
So they wanted to ensure that users had some sense
208
00:12:57.070 --> 00:13:02.029
of choice in the algorithmic recommendations that were presented to them.
209
00:13:02.250 --> 00:13:05.649
And that there was also a level of transparency around
210
00:13:05.690 --> 00:13:10.100
this and accountability for platforms. and service providers in how
211
00:13:10.139 --> 00:13:16.309
those algorithmic recommendations were being developed and presented. A year later,
212
00:13:16.409 --> 00:13:21.629
basically a year later in 2023, they developed these provisions
213
00:13:21.750 --> 00:13:25.690
on deep synthesis internet information services, they call it. And
214
00:13:25.710 --> 00:13:29.799
this basically covered synthetic or AI generated text. That was
215
00:13:29.960 --> 00:13:36.269
really important because we see this rise of synthetically generated
216
00:13:36.330 --> 00:13:40.000
or AI generated images and audio and video. These things
217
00:13:40.080 --> 00:13:45.419
are very fundamental risks to cybersecurity, and they're also risks
218
00:13:45.500 --> 00:13:49.799
to personal security. So they have, in addition to providing
219
00:13:49.840 --> 00:13:55.250
that all AI generated data is clearly labeled as AI generated,
220
00:13:56.129 --> 00:13:58.990
they also took steps around the same time to ensure
221
00:13:59.029 --> 00:14:03.549
that anyone online offering advice, medical advice or financial advice,
222
00:14:03.600 --> 00:14:06.840
for example, has to be fully qualified. So you can't
223
00:14:06.879 --> 00:14:10.870
just now go, go on, you know, TikTok or whatever and,
224
00:14:11.370 --> 00:14:14.529
and give investment advice and what's worked for you and
225
00:14:14.549 --> 00:14:18.129
growing your portfolio if you're not a financial advisor and
226
00:14:18.190 --> 00:14:19.370
same with medical advice.
227
00:14:19.649 --> 00:14:22.529
So that was part of a, you could.
228
00:14:21.970 --> 00:14:28.320
Say validating of online information that people were, were absorbing
229
00:14:28.580 --> 00:14:34.690
and engaging with. Um, A few months later, in August 2023,
230
00:14:34.690 --> 00:14:38.330
they developed these interim measures for the management of generative
231
00:14:38.669 --> 00:14:44.190
artificial intelligence services. And there they basically address things like
232
00:14:44.269 --> 00:14:47.750
training data. So we know that the data that's used
233
00:14:47.830 --> 00:14:51.129
to train models is very important. And you hear this
234
00:14:51.190 --> 00:14:56.149
often from African countries, for example, pointing out that the
235
00:14:56.210 --> 00:14:59.389
models that are being made available to them are not
236
00:14:59.490 --> 00:15:04.159
trained unrelevant data for them, and hence they're not particularly useful.
237
00:15:04.740 --> 00:15:09.179
But they also need to respect intellectual property, which of
238
00:15:09.279 --> 00:15:12.360
course has been a huge debate in the West, and
239
00:15:12.419 --> 00:15:17.399
particularly in the UK, where artists have taken exception to
240
00:15:17.440 --> 00:15:21.159
their work being used in training these models. But then
241
00:15:21.210 --> 00:15:24.490
just most recently, there was the latest update of the
242
00:15:24.590 --> 00:15:27.610
AI safety governance framework. And this is, I guess, the
243
00:15:27.649 --> 00:15:31.480
most to me, the most interesting document because it really
244
00:15:31.620 --> 00:15:35.480
lays out how the Chinese government is thinking about this,
245
00:15:35.519 --> 00:15:38.100
how it's approaching it. And they're up to version three
246
00:15:38.139 --> 00:15:40.779
of this, just came out a few weeks ago. First
247
00:15:40.799 --> 00:15:43.370
of all, it classifies a whole lot of risks that
248
00:15:43.850 --> 00:15:49.230
are being iteratively understood or incrementally understood. And it also
249
00:15:49.350 --> 00:15:52.549
lays out what's very interesting for people like you and me,
250
00:15:53.159 --> 00:15:56.919
It lays out not only technical mitigations against these risks,
251
00:15:57.039 --> 00:15:59.159
but it lays out governance mechanisms.
252
00:15:59.730 --> 00:16:01.210
It lays out proposals for them.
253
00:16:01.710 --> 00:16:05.149
So, yeah, I think that's kind of the trajectory that
254
00:16:05.169 --> 00:16:07.649
I outlined in the piece and that I've observed here.
255
00:16:08.009 --> 00:16:09.990
And we'll put a link to the piece in the
256
00:16:10.029 --> 00:16:11.950
show notes as well as to some of these pieces,
257
00:16:12.029 --> 00:16:14.830
some of the English translations to some of these instruments
258
00:16:14.929 --> 00:16:17.750
as you refer to them. But it is quite extraordinary
259
00:16:17.789 --> 00:16:21.120
to reflect on that, right? So we're in Australia focused
260
00:16:21.169 --> 00:16:23.240
on digital duty of care at the moment. A big
261
00:16:23.279 --> 00:16:25.740
focus on that is fix our feeds and the algorithms,
262
00:16:25.799 --> 00:16:28.559
which in my view is important, but way too narrow.
263
00:16:28.620 --> 00:16:30.080
We need to be looking at it in a much
264
00:16:30.120 --> 00:16:34.509
more expansive way. But the Chinese approach back in 2022
265
00:16:34.509 --> 00:16:38.970
had that covered. And then, you know, the conversation around
266
00:16:39.029 --> 00:16:42.500
synthetic content, deepfakes, et cetera, is one that's always ongoing
267
00:16:42.649 --> 00:16:46.460
and continues to be ongoing here in Australia, again, having
268
00:16:46.659 --> 00:16:51.159
dealt with that years ago. I'm really intrigued and interested
269
00:16:51.259 --> 00:16:55.360
by some of the substance of these proposals, but also
270
00:16:55.720 --> 00:16:59.080
your comment about the process of Because a lot of
271
00:16:59.120 --> 00:17:07.240
these are published as rules or provisional draft instruments, interim measures, guidelines.
272
00:17:07.759 --> 00:17:11.180
Is there anything in that process, appreciating that it sits
273
00:17:11.220 --> 00:17:16.240
within a much larger Chinese bureaucracy, are there parts of
274
00:17:16.289 --> 00:17:20.430
that that we should be looking to transplant into Australian
275
00:17:20.509 --> 00:17:24.509
or UK structures? And given that you've got that familiarity
276
00:17:24.970 --> 00:17:29.720
with both sets of systems... Is it possible to bring
277
00:17:29.779 --> 00:17:32.259
some of that innovation, do you think, to the way
278
00:17:32.299 --> 00:17:34.579
that Australia is considering regulating, for example?
279
00:17:34.900 --> 00:17:36.140
Yes, I do think so.
280
00:17:36.569 --> 00:17:38.509
And I think when you look at some of the,
281
00:17:39.569 --> 00:17:44.849
especially the technical mitigations in the latest update of the framework,
282
00:17:45.269 --> 00:17:49.940
for example, the proposal that AI agents be given a
283
00:17:50.079 --> 00:17:56.099
unique identifier. I've heard proposals along the lines of, Can
284
00:17:56.119 --> 00:18:00.299
they be assigned an IP address or some other unique identifier?
285
00:18:00.460 --> 00:18:04.299
And I think something like that could really be quite
286
00:18:04.359 --> 00:18:09.619
game-changing because, you know, getting back to what we were
287
00:18:09.660 --> 00:18:13.839
saying earlier about the relative merits of not being able
288
00:18:13.880 --> 00:18:16.960
to buy a burner phone and use it to commit crime,
289
00:18:17.720 --> 00:18:21.220
being able to trace an agent and know where it
290
00:18:21.259 --> 00:18:23.980
has been, I mean, this is pertinent for Australia, right,
291
00:18:24.019 --> 00:18:28.380
for the recent Medicare intrusion, being able to know and
292
00:18:28.460 --> 00:18:32.640
identify where has that agent come from and where has
293
00:18:32.660 --> 00:18:35.559
that been, I think is going to be pretty fundamental.
294
00:18:36.119 --> 00:18:40.220
And I think there's real value in considering how we
295
00:18:40.240 --> 00:18:45.230
can collaborate with the Chinese as they develop that technology
296
00:18:45.670 --> 00:18:48.670
to do something like that. There's no reason why Australia
297
00:18:48.690 --> 00:18:51.990
needs to just wait and kind of accept that from
298
00:18:52.049 --> 00:18:55.690
another country. But I think there's real potential and value
299
00:18:55.849 --> 00:18:59.670
in looking at, we all have that interest. We all
300
00:18:59.710 --> 00:19:02.269
have that interest in being able to identify agents online.
301
00:19:02.329 --> 00:19:03.529
How can we do that together?
302
00:19:04.569 --> 00:19:09.109
So we've spoken there through the, I guess, the timeline
303
00:19:09.329 --> 00:19:12.730
of how China has approached the regulation. We've spoken a
304
00:19:12.750 --> 00:19:15.450
little bit about the processes that they use and also
305
00:19:15.490 --> 00:19:17.940
their willingness to say this doesn't work and we need
306
00:19:18.019 --> 00:19:21.480
to do something else, which I think is actually quite unique.
307
00:19:21.559 --> 00:19:23.690
There's not a lot of governments that have that. And
308
00:19:23.710 --> 00:19:26.230
that's partly because they don't have an opposition that's going
309
00:19:26.250 --> 00:19:29.170
to say you got it wrong. So let's acknowledge that.
310
00:19:30.049 --> 00:19:30.470
That's right.
311
00:19:31.150 --> 00:19:35.440
But what do you think countries like Australia should learn
312
00:19:35.480 --> 00:19:40.059
about regulating iteratively based on the Chinese approach? Is that
313
00:19:40.140 --> 00:19:43.509
the key lesson out of here? When you wrote that article,
314
00:19:43.869 --> 00:19:45.890
what was the sort of key take out that you
315
00:19:45.930 --> 00:19:49.109
wanted people to have from this? Is it the iterative nature?
316
00:19:49.289 --> 00:19:52.140
Is it that it's possible to do it? What's the focus?
317
00:19:52.619 --> 00:19:56.799
I think there are a couple of takeaways from what
318
00:19:56.839 --> 00:20:01.190
I'm seeing here. One is that we can't pick up
319
00:20:01.289 --> 00:20:04.190
wholesale and replicate what China does.
320
00:20:04.289 --> 00:20:05.730
And even China would say that.
321
00:20:05.910 --> 00:20:11.740
China is not proselytizing or evangelistic about its approach to policy.
322
00:20:11.839 --> 00:20:16.099
It never really seeks.
323
00:20:15.819 --> 00:20:21.130
To impose that anywhere else because they honestly don't believe
324
00:20:21.170 --> 00:20:23.549
that it's possible for anyone else to do what they
325
00:20:23.569 --> 00:20:24.490
do in China.
326
00:20:25.490 --> 00:20:26.009
And it is possible.
327
00:20:26.549 --> 00:20:30.109
Certainly the fact that there's not an opposition to answer
328
00:20:30.170 --> 00:20:33.509
to when things don't work out, it does give them
329
00:20:33.670 --> 00:20:37.380
freedom to turn around on things that haven't worked or
330
00:20:37.460 --> 00:20:41.859
to issue correctives when policy experiments don't pan out the
331
00:20:41.900 --> 00:20:42.819
way they expect them to.
332
00:20:43.339 --> 00:20:44.400
But I think that.
333
00:20:44.299 --> 00:20:50.829
There's value in looking closely at individual instruments and the
334
00:20:51.609 --> 00:20:55.819
clauses and proposals within those instruments to see, Well, what
335
00:20:55.920 --> 00:21:00.940
could we do? For example, the unique identifiers for agents.
336
00:21:01.279 --> 00:21:04.720
That's not a Chinese problem and it's not a Chinese solution.
337
00:21:04.740 --> 00:21:06.240
That is a global problem.
338
00:21:06.960 --> 00:21:09.660
And that's something that everyone would have a vested interest
339
00:21:09.720 --> 00:21:14.599
in understanding. I don't think you would find many policymakers
340
00:21:15.180 --> 00:21:19.230
who would argue for let's allow these agents to just
341
00:21:19.309 --> 00:21:23.710
roam through our systems anonymously and do as they will.
342
00:21:24.019 --> 00:21:27.579
I mean, another proposal is that those agents must be
343
00:21:27.660 --> 00:21:32.829
programmed to comply with relevant laws. And why they haven't
344
00:21:32.869 --> 00:21:37.049
been is a good question, particularly the proprietary models that
345
00:21:37.089 --> 00:21:39.630
are coming out of America with open source models coming
346
00:21:39.670 --> 00:21:42.650
out of China. That becomes more of an individual problem.
347
00:21:43.250 --> 00:21:47.569
But I think, yeah, looking at those instruments very carefully
348
00:21:47.769 --> 00:21:51.599
and considering what could be useful, what might we be
349
00:21:51.650 --> 00:21:56.029
able to adapt? not in whole, maybe in part. I mean,
350
00:21:56.049 --> 00:21:58.230
what I'm still trying to get to grips with here,
351
00:21:58.289 --> 00:22:00.390
and when I have, I'll check back in with you,
352
00:22:01.049 --> 00:22:05.259
is how do they do this so flexibly and so agile,
353
00:22:05.410 --> 00:22:09.380
leaving aside the kind of structure of their governance system here,
354
00:22:09.440 --> 00:22:11.599
but how do they do that? And I know that
355
00:22:11.640 --> 00:22:17.160
they have slightly different relationships with think tanks, with academics.
356
00:22:17.359 --> 00:22:21.700
There's a more, let's say a more joined up line
357
00:22:21.740 --> 00:22:29.329
of communication and research from researchers in universities, industry, think tanks,
358
00:22:29.549 --> 00:22:33.289
and what the government needs. And this is something that
359
00:22:33.309 --> 00:22:37.589
we've worked hard on in the UK to try and
360
00:22:38.210 --> 00:22:42.079
bring a bit more synthesis and coherence to those conversations.
361
00:22:42.559 --> 00:22:45.039
And we continue to struggle with it as well, though.
362
00:22:45.519 --> 00:22:46.579
How do you get that.
363
00:22:46.500 --> 00:22:51.750
Information out of academic conference papers and journal articles? and
364
00:22:51.890 --> 00:22:54.690
into the hands of people who need it. It doesn't
365
00:22:54.910 --> 00:22:58.619
happen as smoothly as it needs to in the UK.
366
00:22:59.099 --> 00:23:03.240
I bet it's similar in Australia and somehow here they
367
00:23:03.279 --> 00:23:05.700
seem to be able to get what they need. And
368
00:23:05.720 --> 00:23:09.480
I think there's something to understand there about how do
369
00:23:09.519 --> 00:23:13.309
they extract that support from the rest of their research
370
00:23:13.549 --> 00:23:15.029
community and infrastructure.
371
00:23:15.589 --> 00:23:19.750
And I think the research community but also the relationships
372
00:23:19.789 --> 00:23:23.930
with industry as well and that sort of really close connection,
373
00:23:23.970 --> 00:23:28.430
which inevitably is what enables the government to be able
374
00:23:28.490 --> 00:23:31.930
to understand, you know, where the frontier is that they
375
00:23:31.990 --> 00:23:35.200
need to be regulating in a way that just doesn't
376
00:23:35.420 --> 00:23:39.039
exist with the separation between Silicon Valley and the rest
377
00:23:39.099 --> 00:23:42.559
of the world, perhaps not even Silicon Valley and Washington.
378
00:23:42.579 --> 00:23:52.819
Music That's an interesting segue, perhaps, to think about the
379
00:23:52.859 --> 00:23:56.960
different ways that China and the US think about global
380
00:23:57.000 --> 00:24:03.490
governance of AI. So China recently established the Shanghai World
381
00:24:03.769 --> 00:24:05.990
AI Cooperation Organization.
382
00:24:06.009 --> 00:24:06.970
That's right.
383
00:24:07.150 --> 00:24:11.130
When you reflect on the way that China's approaching AI,
384
00:24:11.700 --> 00:24:15.319
engaging globally, taking your point that they're not proselytizing the
385
00:24:15.359 --> 00:24:21.890
particular method of governance process-wise, but they are quite proactive
386
00:24:22.089 --> 00:24:26.109
in saying we are being responsible, our AI is being
387
00:24:26.150 --> 00:24:28.650
built in a way that can be controlled and can
388
00:24:28.710 --> 00:24:32.779
be regulated. What's your reflection on that contrast between that
389
00:24:32.900 --> 00:24:33.579
and the US?
390
00:24:34.200 --> 00:24:38.079
Yeah, I mean, this is quite fundamental to different approaches
391
00:24:38.220 --> 00:24:44.230
to global governance and to the rest of world slash
392
00:24:44.250 --> 00:24:48.869
global south, however we would think about it. I guess
393
00:24:49.269 --> 00:24:54.950
in America, the last, what maybe we could say 70
394
00:24:54.950 --> 00:24:59.859
years or more, has been driven by a particular belief
395
00:25:00.740 --> 00:25:06.160
in innovation, which is that governments should play very little role.
396
00:25:06.920 --> 00:25:07.400
They should.
397
00:25:08.980 --> 00:25:10.960
You know, stay out of the way, leave it to
398
00:25:10.980 --> 00:25:11.920
the private sector.
399
00:25:12.440 --> 00:25:17.500
And that's the best way to propel innovation. And then
400
00:25:17.839 --> 00:25:21.119
the benefits of what happens in the private sector will
401
00:25:21.180 --> 00:25:26.680
trickle down either financially or in other ways without a
402
00:25:26.740 --> 00:25:29.980
great deal of direction, that there's this kind of market
403
00:25:30.200 --> 00:25:35.460
drivers for innovation that will benefit everyone. That's not really
404
00:25:36.569 --> 00:25:42.690
what you see here. And Here, I think the approach
405
00:25:42.759 --> 00:25:49.319
is much more about problem solving and using technology for
406
00:25:49.380 --> 00:25:52.460
problem solving. So what is the problem? The problem is
407
00:25:52.640 --> 00:25:56.359
our energy supply. So we need technology to manage the
408
00:25:56.440 --> 00:26:01.619
energy grid. What is another problem? The manufacturing base, as
409
00:26:01.900 --> 00:26:05.680
China's middle class grows, the manufacturing base looks perhaps at
410
00:26:05.759 --> 00:26:09.210
risk of moving to countries like Vietnam or What to do? Well,
411
00:26:09.329 --> 00:26:14.630
advanced robotics, advanced manufacturing, these types of things. And I
412
00:26:16.039 --> 00:26:20.759
think for China, that's a fundamental difference. It's not just
413
00:26:20.819 --> 00:26:24.460
about making money and then hoping that will trickle down
414
00:26:24.500 --> 00:26:27.779
to everyone else. It's about thinking, what are the problems
415
00:26:27.839 --> 00:26:32.259
that we have? Lifting 800 million people out of poverty
416
00:26:32.319 --> 00:26:36.839
over the last, whatever it's been, 20, 25 years, that's
417
00:26:36.900 --> 00:26:40.130
a big problem. And there was a lot of technology
418
00:26:40.150 --> 00:26:44.299
that went into addressing that problem. And that continues to
419
00:26:44.339 --> 00:26:51.740
be the thinking. So now, in 2026, as China, I guess,
420
00:26:51.799 --> 00:26:57.890
is on the doorstep of this new global leadership role,
421
00:26:58.430 --> 00:27:04.299
that is very much structured around supporting the global south
422
00:27:04.359 --> 00:27:06.690
and supporting the developing world in those countries that are
423
00:27:06.740 --> 00:27:10.599
still coming up behind it to escape poverty. Now, you can,
424
00:27:11.319 --> 00:27:16.519
of course, see that somewhat cynically or instrumentally, as Kishore
425
00:27:16.559 --> 00:27:21.579
Mahbubani points out, the former foreign minister of Singapore, he's,
426
00:27:21.880 --> 00:27:23.819
you know, he's, he's famous for this line, he says,
427
00:27:24.259 --> 00:27:27.400
only 13% of the world is in the West. And
428
00:27:27.440 --> 00:27:30.480
that 13% has a particular view of China, the other
429
00:27:30.480 --> 00:27:34.960
87% has quite a different view of engagement with China.
430
00:27:35.160 --> 00:27:39.960
And China also understands that. So you could see that
431
00:27:40.059 --> 00:27:43.099
quite cynically and instrumentally that this is a good.
432
00:27:43.599 --> 00:27:44.880
Profile of global governance.
433
00:27:46.309 --> 00:27:49.130
But you could also see it as that's their approach.
434
00:27:49.349 --> 00:27:53.289
They would say that they feel a responsibility to support
435
00:27:53.349 --> 00:27:56.490
those countries that are coming up behind them. And they also,
436
00:27:56.509 --> 00:27:59.470
I mean, this is one of the interesting things about
437
00:27:59.490 --> 00:28:01.630
the global governance work they're doing. They have a great
438
00:28:02.289 --> 00:28:06.430
a great Institute set up at Fudan University. And I've
439
00:28:06.470 --> 00:28:09.380
engaged with those, those scholars quite a bit. And they,
440
00:28:09.519 --> 00:28:13.940
and they pointed out that people in the global South,
441
00:28:13.980 --> 00:28:19.140
when they hear this discourse about AI safety, they really
442
00:28:19.539 --> 00:28:23.859
respond negatively to that because for them, it sounds a
443
00:28:24.039 --> 00:28:32.809
lot like, um, like climate change discourse. now we've industrialized,
444
00:28:32.849 --> 00:28:37.009
we'd like everyone to be more conscious of environmental degradation
445
00:28:37.470 --> 00:28:40.869
while they still struggling to kind of get their economies
446
00:28:40.930 --> 00:28:43.710
and their social systems off the ground. So for them,
447
00:28:44.069 --> 00:28:46.369
it can be received in the global South as the
448
00:28:46.430 --> 00:28:51.869
West again, the developed wealthy West again, saying, be careful now,
449
00:28:51.890 --> 00:28:55.130
let's not go too far and too fast, just slow
450
00:28:55.150 --> 00:28:56.509
down with AI.
451
00:28:56.630 --> 00:29:01.000
And that That's not received all that well in the
452
00:29:01.039 --> 00:29:01.740
rest of the world.
453
00:29:02.109 --> 00:29:04.029
And I think it's when you combine that with what
454
00:29:04.069 --> 00:29:07.710
you had said before about the focus of Chinese AI
455
00:29:07.789 --> 00:29:10.890
on problem solving, and then you combine it with that
456
00:29:10.990 --> 00:29:13.740
history of don't tell us to slow down. Now you
457
00:29:13.799 --> 00:29:17.779
have the advantage. It is very powerful. And I attended
458
00:29:17.819 --> 00:29:19.720
a conference in India a couple of years ago and
459
00:29:19.779 --> 00:29:25.410
came away just so struck by how Every Western person
460
00:29:25.490 --> 00:29:29.450
was talking about the risks, the dangers, and everyone from
461
00:29:29.490 --> 00:29:32.539
the rest of the world was talking about the opportunity.
462
00:29:32.710 --> 00:29:35.180
And even if you look at, you know, the AI
463
00:29:35.359 --> 00:29:38.859
impact summit held by India versus the AI safety summits
464
00:29:38.900 --> 00:29:43.420
that have been held in the UK or France, Seoul,
465
00:29:43.839 --> 00:29:47.619
it's such a different focus. In the conversations that you're
466
00:29:47.640 --> 00:29:51.960
having in China, is there a conceptualization of being in
467
00:29:52.019 --> 00:29:54.359
a race with the US? You know, because that's how
468
00:29:54.420 --> 00:29:57.869
the US is always perceiving this is, right? We have
469
00:29:57.890 --> 00:29:59.910
to beat China. We have to be in that race.
470
00:30:01.089 --> 00:30:03.619
How is that conversation unfold?
471
00:30:04.119 --> 00:30:07.440
I never hear that here. And that's not to say
472
00:30:07.480 --> 00:30:09.960
it never happens, but I never hear it. They don't
473
00:30:10.140 --> 00:30:13.240
use that language of a race with the United States.
474
00:30:13.420 --> 00:30:17.670
It may be surprising the extent to which China still
475
00:30:18.410 --> 00:30:23.710
admires and respects the United States, and particularly big tech
476
00:30:23.750 --> 00:30:27.279
companies and great universities and The other thing I never
477
00:30:27.339 --> 00:30:32.700
hear here, in addition to racing to the US, I
478
00:30:32.759 --> 00:30:36.079
don't hear them talk about being number one. And in fact,
479
00:30:36.839 --> 00:30:42.890
there's a kind of saying in Chinese where basically it's
480
00:30:42.910 --> 00:30:47.289
better to be number two. And I see that reflected
481
00:30:47.349 --> 00:30:50.769
a lot in China. Number one is a dangerous position.
482
00:30:51.890 --> 00:30:55.599
Number two is a safer position and it's still great.
483
00:30:56.269 --> 00:30:59.750
And that sort of resonates with me a lot when
484
00:30:59.789 --> 00:31:03.519
I hear Chinese people talk about their relationship with the US,
485
00:31:03.640 --> 00:31:07.500
the tech competition with the US. It gets back to
486
00:31:07.940 --> 00:31:10.319
that thing of problem solving. So they have their own
487
00:31:10.380 --> 00:31:13.660
set of problems that they're directing AI towards, and that's
488
00:31:13.680 --> 00:31:16.619
what they're focused on. So if America does great as well,
489
00:31:16.859 --> 00:31:19.849
it's fine. The thing that they do object to, of course,
490
00:31:20.200 --> 00:31:24.349
is the export bans on things like chips and chip
491
00:31:24.390 --> 00:31:26.670
fab equipment and that type of thing. Of course, they
492
00:31:26.690 --> 00:31:30.190
object to that and they feel that's a very kind
493
00:31:30.210 --> 00:31:35.559
of blunt instrument to try and slow China's progress. But
494
00:31:35.599 --> 00:31:39.359
I don't hear that race. It just doesn't come up.
495
00:31:39.579 --> 00:31:43.579
They're very focused on what they're doing and they don't
496
00:31:43.619 --> 00:31:46.480
like the U.S. getting in the way. But yeah, that's
497
00:31:46.519 --> 00:31:47.299
how it seems to me.
498
00:31:47.910 --> 00:31:52.740
There's the recent meeting between the Chinese president, the US president,
499
00:31:53.160 --> 00:31:58.720
and putting in place the startings of conversations around emergency
500
00:31:58.759 --> 00:32:03.829
communications during an AI crisis, for example. Is that something
501
00:32:03.869 --> 00:32:06.630
that is also perceived as posturing or is that seen
502
00:32:06.789 --> 00:32:08.910
as necessary leadership?
503
00:32:09.630 --> 00:32:13.339
I mean, I think it's seen as necessary and desirable.
504
00:32:13.519 --> 00:32:19.440
I think with any kind of potential crisis, global crisis
505
00:32:19.460 --> 00:32:25.720
or anything that has the potential to escalate, there's generally
506
00:32:25.759 --> 00:32:30.539
a preference for good communication and communication channels. The kind
507
00:32:30.559 --> 00:32:34.869
of feedback that I got about the recent meeting in
508
00:32:34.900 --> 00:32:38.329
the US and the earlier one, earlier in the year,
509
00:32:39.130 --> 00:32:43.069
was that, and this was not from Chinese colleagues, actually,
510
00:32:43.109 --> 00:32:47.170
it was from American colleagues, Well, there was a frustration
511
00:32:47.230 --> 00:32:50.980
on the Chinese side that there was insufficient preparation for
512
00:32:51.000 --> 00:32:53.440
those meetings in both cases, that there was a lack
513
00:32:53.480 --> 00:32:56.539
of preparatory meetings. There was a lack of discussion about
514
00:32:56.579 --> 00:33:02.650
what could be possible. And American colleagues pointed out to
515
00:33:02.690 --> 00:33:07.309
me that this is partly a consequence of the absence
516
00:33:07.349 --> 00:33:12.289
of professional diplomats in the State Department. So without that
517
00:33:12.589 --> 00:33:16.769
real professional class of experienced diplomats that know how to
518
00:33:16.829 --> 00:33:20.190
prepare for these meetings, that know how to get things
519
00:33:20.250 --> 00:33:23.990
in place, that know how to lay the pathway for
520
00:33:24.029 --> 00:33:28.509
something substantial to come out, nothing comes out. And I
521
00:33:29.369 --> 00:33:32.910
think that was a real frustration on the Chinese side,
522
00:33:32.950 --> 00:33:37.380
that there just wasn't this preparatory work to make the
523
00:33:37.440 --> 00:33:41.809
most of those two meetings. But Not to say that
524
00:33:41.839 --> 00:33:45.740
they weren't, look, they're still useful and it's good to
525
00:33:45.819 --> 00:33:51.240
see that those leaders can meet. But it was frustrating
526
00:33:51.259 --> 00:33:54.500
that there wasn't more of substance that happened during those meetings.
527
00:33:55.059 --> 00:33:57.730
One thing that I found myself saying a lot recently
528
00:33:57.970 --> 00:34:00.650
actually in respect to copyright but applies here as well
529
00:34:00.710 --> 00:34:04.670
is that negotiation isn't magic, right? It requires a huge
530
00:34:04.710 --> 00:34:08.090
amount of hard work and preparation to really understand the
531
00:34:08.150 --> 00:34:11.929
other person's perspective and where the commonality is. And then,
532
00:34:12.159 --> 00:34:14.539
you know, to actually get sensible people in a room
533
00:34:14.599 --> 00:34:17.679
to have the conversation, to make the concessions because there's
534
00:34:17.739 --> 00:34:22.940
always concessions in that type of conversation. And yeah, I mean,
535
00:34:23.000 --> 00:34:25.800
the hollowing out of the State Department is having huge
536
00:34:25.840 --> 00:34:30.199
implications across every domain. Think about development assistance or engagement
537
00:34:30.260 --> 00:34:34.449
bilaterally with different countries. It really, it's not something that
538
00:34:34.489 --> 00:34:36.710
I think people focus on, but you're absolutely right in
539
00:34:36.750 --> 00:34:39.929
terms of that impact and the ability to drive and
540
00:34:39.989 --> 00:34:45.130
to get good outcomes. So you close your article talking
541
00:34:45.170 --> 00:34:48.650
about liability and the fact that, you know, if the
542
00:34:48.769 --> 00:34:51.989
US continues perhaps to take an approach that is sort
543
00:34:52.030 --> 00:34:54.489
of hands off, let it rip, let the companies do
544
00:34:54.510 --> 00:34:56.440
what they like, despite the fact that the companies are
545
00:34:56.480 --> 00:34:59.199
calling for this regulation, although sometimes we take that with
546
00:34:59.239 --> 00:35:00.300
a grain of salt too because.
547
00:35:00.159 --> 00:35:01.079
We've heard that before.
548
00:35:02.469 --> 00:35:07.730
That liability might be a way that the world globally
549
00:35:07.869 --> 00:35:11.260
responds and puts in place a regime that regardless of
550
00:35:11.320 --> 00:35:13.639
what the US government may or may not want to do,
551
00:35:14.019 --> 00:35:18.900
holds companies accountable. Do you see something like that happening?
552
00:35:19.260 --> 00:35:24.230
There is that liability that exists for companies. Is it possible,
553
00:35:24.269 --> 00:35:27.630
do you think, to have that type of regime? And
554
00:35:27.670 --> 00:35:29.630
would it be something that you think we would need
555
00:35:29.750 --> 00:35:32.869
the US government participation in? Is it something that a
556
00:35:33.309 --> 00:35:36.429
middle power like Australia could build a coalition of countries
557
00:35:36.590 --> 00:35:39.960
to sort of implement something like this? This concept of
558
00:35:39.980 --> 00:35:42.039
liability is one I'm focused on a lot.
559
00:35:41.940 --> 00:35:43.159
At the moment. Okay.
560
00:35:43.340 --> 00:35:43.639
Yeah.
561
00:35:44.139 --> 00:35:44.980
Not surprising.
562
00:35:46.760 --> 00:35:50.880
I guess from years of working in cybersecurity, I felt
563
00:35:50.940 --> 00:35:54.809
that Essentially, liability really sharpens the mind, doesn't it?
564
00:35:55.679 --> 00:35:57.920
And in some ways... Changes the incentives.
565
00:35:57.980 --> 00:35:58.980
That's actually the reality.
566
00:35:59.000 --> 00:36:02.050
It changes the incentives. That's the reality, Joanna.
567
00:36:02.409 --> 00:36:08.429
Australia's produced this fabulous guidance for corporate governance of cyber.
568
00:36:08.989 --> 00:36:10.849
We looked at a lot of these because we developed
569
00:36:10.949 --> 00:36:13.809
one in the UK as well. And the Australian one
570
00:36:13.869 --> 00:36:18.110
is just incredibly sophisticated, valuable document in my mind.
571
00:36:18.610 --> 00:36:20.800
And essentially....
572
00:36:20.949 --> 00:36:23.150
What we're seeing in countries, I don't know if this
573
00:36:23.210 --> 00:36:26.829
is the next step in Australia, but in the UK,
574
00:36:26.869 --> 00:36:30.369
there's a discussion now that we've issued this voluntary guidance
575
00:36:31.170 --> 00:36:35.510
that in some not too distant future that this will
576
00:36:35.530 --> 00:36:38.699
be written into the corporate code. In other words, directors
577
00:36:38.739 --> 00:36:42.699
of companies will become liable for the cybersecurity of their organizations,
578
00:36:43.039 --> 00:36:46.380
not just fiscally liable, liable for the cybersecurity as well.
579
00:36:46.969 --> 00:36:50.739
And this is at the end of decades of trying
580
00:36:50.800 --> 00:36:55.960
to get business leaders to apply themselves more enthusiastically and
581
00:36:56.019 --> 00:36:59.440
energetically to cybersecurity and realizing at the end of the day,
582
00:36:59.960 --> 00:37:03.179
there are not other incentives that are working. There are
583
00:37:03.239 --> 00:37:06.860
not financial incentives. Business leaders are very good at evaluating
584
00:37:06.960 --> 00:37:10.159
risk and costing that risk. And that's what they do.
585
00:37:10.489 --> 00:37:14.969
And so I think it's really from that experience in
586
00:37:15.010 --> 00:37:21.739
cybersecurity of seeing governments struggle for years to encourage, plead,
587
00:37:21.760 --> 00:37:26.579
incentivize businesses to get more serious about cybersecurity, that now
588
00:37:26.619 --> 00:37:30.300
we land on this issue of corporate liability or director liability.
589
00:37:30.900 --> 00:37:36.389
And I think with.
590
00:37:34.670 --> 00:37:37.389
AI, this will be fundamental.
591
00:37:37.429 --> 00:37:39.769
But the problem is we won't have decades to talk
592
00:37:39.829 --> 00:37:42.849
about this and think about it. We're going to have
593
00:37:42.909 --> 00:37:47.320
to build on past experience and act more quickly. to
594
00:37:47.719 --> 00:37:54.300
work out how can this be implemented and what kind
595
00:37:54.340 --> 00:37:58.329
of models will work for liability. Because essentially you could
596
00:37:58.369 --> 00:38:04.949
say this is maybe perhaps controversial or provocative, but you
597
00:38:04.969 --> 00:38:08.340
could say that, you know, these AI companies are essentially
598
00:38:08.420 --> 00:38:12.500
releasing malware into the system. And then when something happens,
599
00:38:12.539 --> 00:38:15.679
they apologize some months later in the case of Australia.
600
00:38:16.199 --> 00:38:18.139
Or I don't even know if there was an apology, actually.
601
00:38:18.199 --> 00:38:20.679
I read the Prime Minister's... There has since been an apology.
602
00:38:20.699 --> 00:38:22.000
There has been an apology.
603
00:38:22.219 --> 00:38:22.380
Yeah.
604
00:38:22.420 --> 00:38:24.449
But that's not really sufficient, is it?
605
00:38:24.659 --> 00:38:27.090
Because that's not how you can develop things in a
606
00:38:27.309 --> 00:38:31.949
global system upon which our critical infrastructure is built and runs.
607
00:38:32.070 --> 00:38:38.309
That's not appropriate. And if they feel they need some
608
00:38:38.349 --> 00:38:38.789
kind of.
609
00:38:40.570 --> 00:38:43.500
Governance to prevent them from doing that and they're not
610
00:38:43.579 --> 00:38:47.409
getting it in their home country... then I guess that
611
00:38:47.530 --> 00:38:51.380
that's one option for countries like Australia to consider how
612
00:38:51.409 --> 00:38:54.019
can a liability framework work. And I think, you know,
613
00:38:54.099 --> 00:38:59.880
Australia is an amazing example. It's really internationally when we
614
00:38:59.920 --> 00:39:04.239
have these discussions about tech policy, often looking at Australia,
615
00:39:04.400 --> 00:39:08.360
it's just for relatively small countries, it really punches above
616
00:39:08.409 --> 00:39:11.469
its weight. And in part due to the work that
617
00:39:11.530 --> 00:39:14.389
you guys do, but also the eSafety Commissioner, you see
618
00:39:14.429 --> 00:39:19.239
these Things like the under-16 ban being very quickly replicated
619
00:39:19.260 --> 00:39:22.219
around the world. I think something like 40 countries now
620
00:39:22.429 --> 00:39:26.530
are trialing that. And I think Australia, somehow you have
621
00:39:26.630 --> 00:39:31.219
some kind of magic mix where you're able to it
622
00:39:31.280 --> 00:39:35.119
doesn't always work, but you're able to put together approaches
623
00:39:35.179 --> 00:39:40.920
to these problems that are quite bold and innovative. And yeah,
624
00:39:41.019 --> 00:39:44.030
I guess that's what would be nice to see in
625
00:39:44.070 --> 00:39:46.710
this context as well. To a large extent, it is
626
00:39:46.750 --> 00:39:51.090
a middle power problem, isn't it, Joanna? Because For most countries,
627
00:39:52.570 --> 00:39:55.489
the US and China maybe being the exception, we're going
628
00:39:55.510 --> 00:40:00.380
to have foreign technology in our critical infrastructure, in everything.
629
00:40:00.440 --> 00:40:03.440
We are going to have foreign owned and operated technology
630
00:40:04.179 --> 00:40:07.139
deep in our infrastructure. And the idea that we will
631
00:40:07.800 --> 00:40:12.449
create our own or have domestic alternatives in bits and
632
00:40:12.530 --> 00:40:17.190
pieces maybe, but fundamentally, no, that will not happen. And
633
00:40:17.230 --> 00:40:21.639
so the great question of the next chapter, I guess,
634
00:40:22.139 --> 00:40:25.809
is exactly what you're doing. How can you then govern
635
00:40:26.469 --> 00:40:30.409
foreign technology in a way that it allows you to
636
00:40:30.489 --> 00:40:34.989
extract the maximum benefits and mitigate against the worst risks?
637
00:40:35.530 --> 00:40:38.110
And that is a governance problem. There might be some
638
00:40:38.170 --> 00:40:41.239
technological solutions as well, but in large part, that is
639
00:40:41.280 --> 00:40:45.340
a governance problem. But I think for Australia in particular,
640
00:40:45.559 --> 00:40:48.989
and the UK is in the same position, We had
641
00:40:49.030 --> 00:40:54.440
this long kind of easy ride where the best technology
642
00:40:55.179 --> 00:40:58.139
for the last many decades has been coming from our
643
00:40:58.260 --> 00:41:03.400
very close strategic partner. Now, we need to be planning
644
00:41:03.440 --> 00:41:05.949
for the future where that is not happening. The best
645
00:41:05.989 --> 00:41:10.389
technology in every domain is not coming from the United
646
00:41:10.429 --> 00:41:13.349
States anymore. A lot of the best technology is coming
647
00:41:13.389 --> 00:41:15.730
from China and it will continue to do so. And
648
00:41:15.789 --> 00:41:18.650
we want that technology. And we don't have a close
649
00:41:18.670 --> 00:41:21.769
strategic relationship with China. So how can we do that
650
00:41:23.409 --> 00:41:28.000
without just excluding ourselves from the future? How can we
651
00:41:28.059 --> 00:41:31.260
build those governance structures so that we can get what
652
00:41:31.320 --> 00:41:31.780
we need?
653
00:41:32.139 --> 00:41:33.760
You know, I was in a meeting yesterday actually where
654
00:41:33.800 --> 00:41:37.840
someone was talking about, you know, how powerful the Chinese
655
00:41:37.920 --> 00:41:41.690
open source, open weight models are and that we need
656
00:41:41.710 --> 00:41:44.190
to take a similar approach to Australia took with 5G
657
00:41:44.250 --> 00:41:47.170
and just exclude all of those models. And You know,
658
00:41:47.190 --> 00:41:50.409
that is one potential option, but I think it is
659
00:41:50.760 --> 00:41:54.800
a small subset of the options that we have available
660
00:41:54.860 --> 00:41:57.599
to us. But we have to be able to provide
661
00:41:57.800 --> 00:42:01.579
assurance that we actually are able to control that technology
662
00:42:01.659 --> 00:42:05.090
and also that it's not having a negative security implication.
663
00:42:05.829 --> 00:42:08.800
And the irony is, and so much of this conversation
664
00:42:08.840 --> 00:42:12.280
is actually uncomfortable, right? But at the moment, the Chinese
665
00:42:12.340 --> 00:42:16.059
are actually much better at controlling their models because they've
666
00:42:16.099 --> 00:42:24.599
got all of these regulations. Madeline, I want to ask
667
00:42:24.739 --> 00:42:27.909
one last question of you before we let you go.
668
00:42:28.389 --> 00:42:32.670
What do you think is the biggest misperception about Chinese
669
00:42:32.710 --> 00:42:36.670
approaches to technology or something that you find yourself constantly
670
00:42:36.949 --> 00:42:41.139
correcting or itching to set the record straight on?
671
00:42:42.159 --> 00:42:49.699
Perhaps that everything here in the tech sector is top-down managed.
672
00:42:50.400 --> 00:42:54.409
It's one of the most interesting questions misconceptions that I've
673
00:42:54.469 --> 00:42:55.630
come to understand.
674
00:42:56.469 --> 00:42:58.329
Certainly there is a kind of.
675
00:42:58.510 --> 00:43:03.300
Overall centrally set framework for where the country is going
676
00:43:03.480 --> 00:43:07.539
and what the objectives are, but the extent to which
677
00:43:07.599 --> 00:43:11.159
those are developed locally at a provincial level or a
678
00:43:11.199 --> 00:43:17.090
municipal level is really extraordinary. And the competition, it's an
679
00:43:17.150 --> 00:43:23.210
incredibly competitive ecosystem here. So the people here are very entrepreneurial,
680
00:43:24.690 --> 00:43:28.900
but it's incredibly competitive. And the Chinese government has no
681
00:43:28.960 --> 00:43:34.039
problem allowing competition to fall by the wayside and the
682
00:43:34.280 --> 00:43:38.760
survival of the strongest, basically. I mean, there's a story
683
00:43:38.820 --> 00:43:43.489
about how they invited Tesla into China, that there were many,
684
00:43:43.550 --> 00:43:47.670
many EV companies at one point, and they really wanted
685
00:43:47.750 --> 00:43:51.320
to kind of cull them and They invited Tesla in
686
00:43:51.840 --> 00:43:57.780
as the world's best EV company. And that eliminated a
687
00:43:57.880 --> 00:44:00.920
lot of the Chinese competition because they couldn't measure up
688
00:44:00.940 --> 00:44:05.650
to Tesla. And so there's a very competitive tech landscape here.
689
00:44:06.369 --> 00:44:12.889
There's plenty of government subsidies and provincial level support to
690
00:44:12.949 --> 00:44:17.130
attract tech companies, that's for sure. It's not the case
691
00:44:17.269 --> 00:44:21.190
that these companies are just propped up and they don't
692
00:44:21.210 --> 00:44:24.429
have to succeed. They're very competitive and the ones that
693
00:44:24.489 --> 00:44:28.610
succeed are excellent. And that's why these kind of comments like,
694
00:44:29.809 --> 00:44:32.489
let's just not have Chinese technology, I think are so
695
00:44:32.570 --> 00:44:37.780
short-sighted for Australia because the world is changing and Australia
696
00:44:37.840 --> 00:44:40.679
does not want to be left on the wrong side
697
00:44:41.360 --> 00:44:42.500
of what's happening now.
698
00:44:43.179 --> 00:44:45.219
It doesn't want to shut itself out because.
699
00:44:46.070 --> 00:44:50.110
of some ideological attachment to America.
700
00:44:54.090 --> 00:44:57.150
Thank you so much for taking the time to have
701
00:44:57.210 --> 00:44:59.170
this conversation. Not at all.
702
00:44:59.190 --> 00:45:01.170
It's just a real pleasure to talk to you, Joanna.
703
00:45:01.570 --> 00:45:12.610
Thank you. Well, that's it for this episode of Tech Mirror,
704
00:45:12.750 --> 00:45:15.590
which is brought to you by the Tech Policy Design Institute.
705
00:45:16.280 --> 00:45:18.519
We're based here in Canberra on the lands of the
706
00:45:18.539 --> 00:45:23.590
Ngunnawal Ngambri people. If you found today's conversation useful or thought-provoking,
707
00:45:24.030 --> 00:45:26.349
please do share it with a friend or a colleague
708
00:45:26.690 --> 00:45:29.989
or leave a review and subscribe wherever you get your podcasts.
709
00:45:30.630 --> 00:45:33.550
If you're watching, please don't forget to like and follow.
710
00:45:34.090 --> 00:45:38.210
For show notes, you can visit techpolicy.au forward slash podcast.
711
00:45:38.730 --> 00:45:41.409
And this podcast was made possible with thanks to the
712
00:45:41.550 --> 00:45:46.059
generous contributions from government, industry and philanthropy, to the Tech
713
00:45:46.079 --> 00:45:49.820
Policy Design Fund, the full details of which are available
714
00:45:49.980 --> 00:45:53.820
on our website. The team at Audiocraft produced this pod
715
00:45:53.980 --> 00:45:56.639
on the lands of the Gadigal people of the Eora Nation.
716
00:45:56.880 --> 00:46:00.670
Music is by Talia Skopelis. A big thank you also
717
00:46:00.690 --> 00:46:03.869
to the team at the Tech Policy Design Institute without
718
00:46:03.949 --> 00:46:07.309
whom this pod would not be possible. Thank you for
719
00:46:07.349 --> 00:46:11.409
joining us. And as always, get in touch and get involved.