WEBVTT
1
00:00:00.760 --> 00:00:05.080
In this world of technology is ever change rearranging that
2
00:00:05.799 --> 00:00:07.200
you need someone to help you out.
3
00:00:08.000 --> 00:00:08.720
I know someone that.
4
00:00:08.839 --> 00:00:10.039
Can be alone.
5
00:00:10.119 --> 00:00:11.160
You'll be with Paul.
6
00:00:11.279 --> 00:00:12.720
I'm a day Lane.
7
00:00:12.599 --> 00:00:18.760
In the Tech Zone. Welcome back to the show me, Paul,
8
00:00:18.800 --> 00:00:19.760
I'm Dave Slaine.
9
00:00:19.800 --> 00:00:22.359
Thank you so much. This is our second segment. If
10
00:00:22.399 --> 00:00:25.839
you missed our first segment, check it out. And talked
11
00:00:25.839 --> 00:00:29.800
to Steve Yule from the Consumer Technology Association Foundation, and
12
00:00:29.839 --> 00:00:32.439
I also talked about social media and.
13
00:00:34.079 --> 00:00:37.119
Just how we have to be smarter with it.
14
00:00:37.679 --> 00:00:40.240
I mean, people are making money off of it because
15
00:00:40.280 --> 00:00:42.799
of our anger, and to me, that doesn't make any
16
00:00:42.840 --> 00:00:45.359
sense at Allso if you miss my rant, please please
17
00:00:45.399 --> 00:00:48.280
please check it out. Here in California, there is a
18
00:00:48.320 --> 00:00:51.159
bill that's on the governor's desk that's ready for him
19
00:00:51.159 --> 00:00:56.600
to sign that Some who agree with this bill are
20
00:00:56.640 --> 00:00:59.719
saying that it's needed to make sure there's from guard
21
00:01:00.520 --> 00:01:04.560
on AI and protections of their but there are also
22
00:01:04.799 --> 00:01:09.920
some who feel that this may steiny innovation. So we're
23
00:01:09.959 --> 00:01:15.040
going to talk to an organization that is urging Governor
24
00:01:15.159 --> 00:01:19.920
Newsom to veto this bill that is on his desk,
25
00:01:19.959 --> 00:01:22.319
and we're going to find out why. And also this
26
00:01:22.439 --> 00:01:26.959
Organization A. They were not alone in reaching out to
27
00:01:27.519 --> 00:01:31.879
the governor. The Consumer Technology Association was too. We're going
28
00:01:31.959 --> 00:01:35.400
to find out about this and more with my next guest.
29
00:01:36.760 --> 00:01:39.519
I am so delighted to welcome back to the program.
30
00:01:39.719 --> 00:01:43.079
Our good friend, Paul Leckis is the senior vice president
31
00:01:43.079 --> 00:01:44.760
over at si A.
32
00:01:45.560 --> 00:01:48.280
Paul, what's going on? How are you, Paul?
33
00:01:48.359 --> 00:01:50.560
I'm doing great. It is such a pleasure to be
34
00:01:50.640 --> 00:01:51.359
back here with you.
35
00:01:51.879 --> 00:01:52.959
Hey, it's great to chat with you.
36
00:01:53.120 --> 00:01:57.560
I always enjoy our chats because we talk about some
37
00:01:57.719 --> 00:02:01.159
things that we should all be aware of, especially when
38
00:02:01.159 --> 00:02:04.040
it comes to the not the Internet of things, but
39
00:02:04.159 --> 00:02:06.120
the intelligence.
40
00:02:05.760 --> 00:02:08.400
Of things AI, AI and everything.
41
00:02:08.479 --> 00:02:11.000
Plus I really like your name too, so you know,
42
00:02:11.039 --> 00:02:14.599
I gotta gotta give you problems with that, Paul. Let's
43
00:02:14.599 --> 00:02:18.439
get down to, as we say, not business but business,
44
00:02:18.439 --> 00:02:24.439
all right, Paul. So, there is a new I guess
45
00:02:24.599 --> 00:02:27.919
bill that's out here in California. So it affects me
46
00:02:28.240 --> 00:02:32.680
and for our listeners who are regional here in California,
47
00:02:32.719 --> 00:02:37.479
but also our listeners who are worldwide and even here domestically.
48
00:02:38.240 --> 00:02:41.199
Why don't we give them a little background about this
49
00:02:41.360 --> 00:02:46.680
bill and what it's aimed to do, and why many
50
00:02:46.759 --> 00:02:51.199
are urging Governor Newsom to veto this.
51
00:02:52.159 --> 00:02:56.960
Yeah, absolutely so. I We're talking about a b called
52
00:02:57.159 --> 00:03:03.199
SBE which has now past the California legislature. It is
53
00:03:03.280 --> 00:03:08.520
on Governor Newsom's desk to sign or to veto, and
54
00:03:08.840 --> 00:03:13.800
it is titled the Safe and Secure Innovation for Printier
55
00:03:14.000 --> 00:03:19.800
Artificial Intelligence Models Act. It is quite a mouthful, but
56
00:03:20.199 --> 00:03:23.919
essentially what this bill aims to do is provide a
57
00:03:23.960 --> 00:03:29.039
framework to address concerns around the safety and security of
58
00:03:29.120 --> 00:03:33.319
the largest models out there. We call them frontier AI models,
59
00:03:33.360 --> 00:03:37.599
basically models that are on the frontier of the most
60
00:03:37.680 --> 00:03:41.360
advanced AI models today, the models that are powering things
61
00:03:41.400 --> 00:03:47.319
like CHATGPT, Gemini, and other services that people use regularly.
62
00:03:48.759 --> 00:03:52.719
What the bill would do is impose a number of requirements,
63
00:03:53.000 --> 00:03:57.919
primarily on the developers of these models, requirements that they
64
00:03:57.960 --> 00:04:02.159
would need to undertake before the developing these models, after
65
00:04:02.240 --> 00:04:05.240
developing these models, before they're put into service or made
66
00:04:05.240 --> 00:04:08.879
available to the public, and it would focus on a
67
00:04:08.919 --> 00:04:12.120
couple of different types of risks that are associated with
68
00:04:12.360 --> 00:04:15.599
these sorts of models. Some of them are described as
69
00:04:15.639 --> 00:04:20.160
critical harms, and these are things that include the creation
70
00:04:20.480 --> 00:04:25.399
of chemical, biological, radiological, or nuclear weapons using these models,
71
00:04:26.560 --> 00:04:30.319
the potential for mass casualties of at least five hundred
72
00:04:30.360 --> 00:04:36.920
million dollars, the potential for potential for other grave harms
73
00:04:36.959 --> 00:04:40.120
to public safety and security, and I'm quoting a little
74
00:04:40.120 --> 00:04:45.120
bit from the bill itself. There are also requirements around
75
00:04:45.480 --> 00:04:50.360
what are called AI safety incidents, and these are a
76
00:04:50.439 --> 00:04:53.759
variety of different things that could happen by using an
77
00:04:53.759 --> 00:05:00.199
AI model, including malicious use in indvertain release, unauthorized access,
78
00:05:01.160 --> 00:05:05.839
the behavior that is not intended from the models, either
79
00:05:05.920 --> 00:05:08.959
because these models act in ways that are unpredictable or
80
00:05:09.000 --> 00:05:14.240
because they're misused. And in addition to putting these requirements
81
00:05:14.240 --> 00:05:18.959
on developers, including doing third party audits, conducting assessments of
82
00:05:19.040 --> 00:05:25.639
the potential for risks, reporting any incidents that a developer
83
00:05:25.680 --> 00:05:29.879
is aware of, it would also establish a framework for
84
00:05:30.240 --> 00:05:34.759
the Attorney General to bring suits against developers. And I
85
00:05:34.759 --> 00:05:37.720
think this is one of the reasons why the bill
86
00:05:37.759 --> 00:05:41.759
has been so contentious, and not to get to ahead
87
00:05:41.759 --> 00:05:43.720
of myself, the bill has a number of other things,
88
00:05:43.800 --> 00:05:49.439
but you know, we are talking about really advanced AI
89
00:05:49.560 --> 00:05:54.759
models where the techniques and the technologies to accurately measure
90
00:05:55.360 --> 00:05:59.319
the risk of harm of misuse of misappropriation and so
91
00:05:59.399 --> 00:06:04.519
forth are still being developed. We don't have any concrete
92
00:06:04.560 --> 00:06:10.639
standards that developers can use, and out of the industry community,
93
00:06:10.800 --> 00:06:14.399
there's grave concern that if a bill like this were
94
00:06:14.439 --> 00:06:19.720
to be enacted into law, it could essentially halt innovation
95
00:06:20.040 --> 00:06:24.639
because the specter of liability is too great, it's too ambiguous,
96
00:06:24.839 --> 00:06:31.519
and developers don't have a sense of, you know, what
97
00:06:31.759 --> 00:06:35.519
would be appropriate here. There's a number of other concerns
98
00:06:35.560 --> 00:06:38.959
I think with the general structure. But one other thing
99
00:06:39.000 --> 00:06:42.160
to note is, you know, this is the sort of
100
00:06:42.240 --> 00:06:45.480
thing that the federal government is looking at very closely
101
00:06:45.560 --> 00:06:49.800
right now. They are looking at ways to work collectively
102
00:06:49.959 --> 00:06:54.439
collaboratively with industry to try to develop the right kind
103
00:06:54.439 --> 00:06:59.560
of standards that we need to investigate, interrogate these models
104
00:06:59.560 --> 00:07:02.800
and see what are potentially capable of. It's something that's
105
00:07:03.040 --> 00:07:07.199
happening at the federal level, something that's happening internationally, and
106
00:07:08.639 --> 00:07:10.680
you know, there is a view that it is too
107
00:07:10.839 --> 00:07:16.639
premature to build a new system of liability here while
108
00:07:16.639 --> 00:07:19.600
we're still kind of building the airplane.
109
00:07:20.319 --> 00:07:21.680
That definitely makes sense, Paul.
110
00:07:22.839 --> 00:07:25.399
If we have a standard here in California, and let's
111
00:07:25.439 --> 00:07:30.759
say that the federal government passes a law, then the
112
00:07:30.800 --> 00:07:34.160
federal law will supersede the state law.
113
00:07:34.399 --> 00:07:37.040
I assume there are to be a lot of confusion
114
00:07:37.800 --> 00:07:39.759
right when it comes when it comes to a lot
115
00:07:39.759 --> 00:07:40.040
of things.
116
00:07:40.040 --> 00:07:44.279
And and looking at at this letter that was sent
117
00:07:44.360 --> 00:07:49.120
to Governor Newsom, it was not only your organization, but
118
00:07:49.199 --> 00:07:53.800
also a powerhouse of the Consumer Technology Association too, And
119
00:07:54.399 --> 00:07:55.519
what was it like to.
120
00:07:56.040 --> 00:08:00.319
Kind of have them get on board with with so
121
00:08:00.360 --> 00:08:05.000
with SIIA, to really come together and collaborate to talk
122
00:08:05.040 --> 00:08:09.319
about some things that needed to be put in front
123
00:08:09.360 --> 00:08:12.120
of the governor before he made a decision on what
124
00:08:12.199 --> 00:08:13.439
to do with this actual bill.
125
00:08:13.959 --> 00:08:18.279
Yeah. So, the Conservative Technology Association CTA is one of
126
00:08:18.279 --> 00:08:21.600
the organizations in the tech space that we collaborate with
127
00:08:21.720 --> 00:08:23.879
a lot. We work with them a lot on issues
128
00:08:24.839 --> 00:08:29.199
and what they're hearing from their members. There's some overlapping
129
00:08:29.319 --> 00:08:31.680
members of our two associations, but they have a number
130
00:08:31.720 --> 00:08:34.360
of members that are not part of SIA and vice versa.
131
00:08:35.480 --> 00:08:37.559
They're hearing the same sort of things that we're hearing
132
00:08:38.200 --> 00:08:40.879
that there is concern that this sort of a bill
133
00:08:41.080 --> 00:08:45.240
being enacted at a state level could really stand in
134
00:08:45.279 --> 00:08:48.679
the way of the kind of innovation we need. But
135
00:08:48.879 --> 00:08:50.519
another thing that we raised in the letter that I
136
00:08:50.519 --> 00:08:53.519
think is really important is that nobody is going into
137
00:08:53.559 --> 00:08:59.960
this thinking that AI models are inherently safe or don't
138
00:09:00.120 --> 00:09:04.279
need some sort of guard rails. Frankly, I think the
139
00:09:04.399 --> 00:09:10.320
safety and security questions that we're talking about at least
140
00:09:10.320 --> 00:09:13.679
as significant is kind of the immediate issues that people
141
00:09:13.679 --> 00:09:16.879
are addressing right now and looking at right now around
142
00:09:16.879 --> 00:09:21.200
the socio economic impact of AI models and discrete situations.
143
00:09:22.159 --> 00:09:27.639
But safety and security is of paramount importance, and if
144
00:09:27.679 --> 00:09:29.639
we are going to develop a framework, we need to
145
00:09:29.679 --> 00:09:34.240
do this in the right way. It is I think
146
00:09:34.279 --> 00:09:40.399
that contrasting this with the area of privacy, and I'm
147
00:09:40.440 --> 00:09:43.960
sure many of your listeners are familiar with the fact
148
00:09:44.000 --> 00:09:46.960
that California has a privacy law and the federal government
149
00:09:47.000 --> 00:09:51.960
does not, and California has really helped to move forward
150
00:09:52.080 --> 00:09:54.600
the debate in the United States around privacy. We are
151
00:09:54.720 --> 00:09:58.039
very hopeful there would be a federal law, but when
152
00:09:58.080 --> 00:10:00.799
we're in the realm of AI safety, this is something
153
00:10:00.879 --> 00:10:05.360
very different. It is it makes a lot of sense
154
00:10:05.440 --> 00:10:09.559
for states and other bodies to be looking at ways
155
00:10:09.600 --> 00:10:12.480
in which AI is used in their jurisdiction, and the
156
00:10:12.480 --> 00:10:16.720
states and counties and cities, and you know, to put
157
00:10:16.759 --> 00:10:20.320
guardrails around things like the use of AI in making
158
00:10:20.360 --> 00:10:23.600
employment decisions or criminal justice or those sorts of things
159
00:10:23.600 --> 00:10:27.399
that people touch every day. But here we're talking about
160
00:10:27.440 --> 00:10:30.240
something that is really at the threshold of where science
161
00:10:30.279 --> 00:10:34.480
and research is and we are really concerned about the
162
00:10:34.519 --> 00:10:38.639
impact of this on innovation. I think that will be
163
00:10:38.720 --> 00:10:44.200
felt more acutely in California, given that this is a
164
00:10:44.240 --> 00:10:48.000
bill coming out of California. But it's also hard to
165
00:10:48.039 --> 00:10:50.279
see how the impact of a bill like this would
166
00:10:50.320 --> 00:10:54.159
be limited to just California. In some ways, it would
167
00:10:54.159 --> 00:10:58.039
be a de facto regulation of AI across the country
168
00:10:59.799 --> 00:11:04.919
in number of cases. We're concerned that this is something
169
00:11:04.960 --> 00:11:08.440
that CTA agrees with as well, that this is going
170
00:11:08.519 --> 00:11:16.240
to lead to really hurt our own industry, and our
171
00:11:16.320 --> 00:11:20.399
industry is working some There may be some naysayers out there,
172
00:11:20.440 --> 00:11:22.279
but the industry is really working to try to be
173
00:11:22.320 --> 00:11:25.879
as safe and secure as possible. And we can't say
174
00:11:25.879 --> 00:11:30.240
the same for AI models that are being developed potentially
175
00:11:30.240 --> 00:11:35.240
in other countries that will probably swoop in and fill
176
00:11:35.279 --> 00:11:38.360
the gap that is created if if we're not able
177
00:11:38.399 --> 00:11:44.240
to continue to release advanced models out of US developers.
178
00:11:45.440 --> 00:11:47.480
You know, Paul that it makes a lot of sense.
179
00:11:47.519 --> 00:11:52.799
Here's here's kind of the I guess concern when it
180
00:11:52.840 --> 00:11:55.759
comes to all all things that AI that we you
181
00:11:55.799 --> 00:11:59.600
know that that we have now we all know that
182
00:11:59.679 --> 00:12:01.519
something needs to be done. You know, you can look
183
00:12:01.559 --> 00:12:06.159
at even though there's this election cycle, how AI has
184
00:12:06.240 --> 00:12:11.440
been used to just show so much disinformation, deep fakes,
185
00:12:11.879 --> 00:12:13.200
all these other things.
186
00:12:13.000 --> 00:12:14.039
That that are going on.
187
00:12:14.759 --> 00:12:19.080
And what can be done I guess for the public
188
00:12:20.039 --> 00:12:25.480
now to kind of, I guess, you know, protect themselves
189
00:12:25.879 --> 00:12:28.919
while while the governments are trying to figure out a
190
00:12:28.960 --> 00:12:34.480
good policy that can number one, won't stifle innovation, but
191
00:12:34.600 --> 00:12:38.080
number two, we'll have some some guard rails and safety,
192
00:12:38.440 --> 00:12:41.720
you know, for for consumers, businesses, and and so forth.
193
00:12:42.240 --> 00:12:49.000
What can consumers, businesses, organizations do now to make sure
194
00:12:49.559 --> 00:12:52.840
that that that they're really having some self accountability to
195
00:12:52.960 --> 00:12:55.639
make sure that they are protecting themselves.
196
00:12:55.919 --> 00:12:59.720
Yeah, and that's such a great question. And you know
197
00:12:59.759 --> 00:13:04.000
that we of the legislatures outside of California tend to
198
00:13:04.039 --> 00:13:10.840
move slowly. And you know, Congress, I think there's a
199
00:13:10.840 --> 00:13:15.080
lot of proposals before Congress that would be really helpful here.
200
00:13:16.559 --> 00:13:19.600
But you know, California is moving very fast. California has
201
00:13:19.600 --> 00:13:21.679
actually taken a number of steps that have been really
202
00:13:21.720 --> 00:13:25.440
positive in the AI space. But you know, what can
203
00:13:25.480 --> 00:13:32.000
we do beyond relying on our elected officials to actually
204
00:13:33.320 --> 00:13:37.879
protect ourselves individually, try to make use of these tools
205
00:13:38.000 --> 00:13:42.639
in a positive way. You'd mentioned misinformation. I think that
206
00:13:42.639 --> 00:13:45.320
that is a concern that people are seeing all the time,
207
00:13:45.600 --> 00:13:51.200
and you know it is I think on that front,
208
00:13:52.519 --> 00:13:55.759
people have to be smart consumers, they have to be
209
00:13:56.759 --> 00:14:00.080
critical thinkers of what they see on the Internet. I
210
00:14:00.120 --> 00:14:03.600
think it's it's really hard to it's really hard for
211
00:14:03.679 --> 00:14:08.440
legislators to try to stop the spread of misinformation online,
212
00:14:08.480 --> 00:14:11.240
even AI enabled misinformation. I think that we've seen some
213
00:14:11.279 --> 00:14:15.879
progress in a few areas and I'm hopeful that will continue.
214
00:14:16.200 --> 00:14:22.240
And that includes deep fakes of non consensual intimate imagery.
215
00:14:22.559 --> 00:14:25.039
We've seen a number of state laws, including in California,
216
00:14:25.840 --> 00:14:29.799
that are really focused on people who are uh uh,
217
00:14:31.159 --> 00:14:33.360
who are the bad actors here, the people who are
218
00:14:34.039 --> 00:14:39.960
generating and disseminating those kind of harmful images. Uh Regulation
219
00:14:40.000 --> 00:14:42.039
of content is something that's very difficult for the government
220
00:14:42.039 --> 00:14:44.879
to do though, because of the First Amendment, Except in
221
00:14:45.120 --> 00:14:49.519
certain discrete uh, situations where there's a grave threat to
222
00:14:49.559 --> 00:14:54.799
public to public interests. So people need to be spark consumers.
223
00:14:55.039 --> 00:15:01.320
And I think, you know, you don't want to put
224
00:15:01.360 --> 00:15:04.519
the onus on individuals, but we need to look with
225
00:15:04.600 --> 00:15:07.279
a good degree of skepticism at information that we're seeing
226
00:15:07.320 --> 00:15:10.600
out there. And I think there is some value in
227
00:15:11.519 --> 00:15:17.600
relying on, you know, reputable sources of information to verify
228
00:15:17.799 --> 00:15:21.840
anything that you see that seems to be shocking or
229
00:15:22.440 --> 00:15:26.639
concerning or you know, out of the norm. I think
230
00:15:26.679 --> 00:15:29.879
that you know, you're seeing in the news space that
231
00:15:30.519 --> 00:15:34.519
reputable news publications are probably seen as more authoritative now,
232
00:15:36.000 --> 00:15:38.679
and that's because there's a lot of information floating out
233
00:15:38.720 --> 00:15:41.000
there that's not correct, and some of it is being
234
00:15:41.039 --> 00:15:44.399
circulated inadvertently, and some of it is intentional. There are
235
00:15:44.519 --> 00:15:49.559
malicious actors out there, for sure. I think that you know,
236
00:15:49.679 --> 00:15:52.600
businesses that are looking to adopt AI can take a
237
00:15:52.679 --> 00:15:55.039
number of steps, and there are frameworks out there to
238
00:15:55.320 --> 00:15:58.559
assess the risk of the tools that they're using to
239
00:15:58.639 --> 00:16:00.799
make sure that they are using them the right way.
240
00:16:01.919 --> 00:16:09.240
I think everyone needs to be careful in adopting AI tools.
241
00:16:09.320 --> 00:16:11.480
I think they can do a lot of good, but
242
00:16:12.240 --> 00:16:15.639
we need to approach them in a reasonable way. Some
243
00:16:15.679 --> 00:16:17.720
of the guidances coming out of the National Institute of
244
00:16:17.720 --> 00:16:22.519
Standards of Technology is NISSED is incredibly useful for those
245
00:16:22.559 --> 00:16:25.279
who are developing AI as well as for those who
246
00:16:25.360 --> 00:16:29.039
are looking to purchase AI or apply AI tools in
247
00:16:29.080 --> 00:16:33.320
different places. And I think that that kind of guidance,
248
00:16:33.879 --> 00:16:39.200
which is being developed in collaboration with academia, civil society,
249
00:16:39.279 --> 00:16:44.200
and industry, is really useful. There is you know, we've
250
00:16:44.240 --> 00:16:47.159
done a lot of work about how to safely incorporate
251
00:16:47.200 --> 00:16:51.360
AI and classrooms, and I think that that the bar
252
00:16:51.480 --> 00:16:54.399
is being raised in that space, as in a number
253
00:16:54.480 --> 00:16:59.720
of other spaces, and I'm optimistic that we're actually from
254
00:16:59.759 --> 00:17:06.759
an industry perspective, creating tools and applications that are safer
255
00:17:06.799 --> 00:17:12.039
and more secure and avoid a number of the concerns
256
00:17:12.119 --> 00:17:15.680
or mitigate the concerns that were present in earlier AI models.
257
00:17:16.559 --> 00:17:20.519
Now we're not totally free of things like algorithmic bias,
258
00:17:21.680 --> 00:17:24.519
but many of the applications that are out there are
259
00:17:24.519 --> 00:17:27.359
doing a much better job with those sorts of things.
260
00:17:27.359 --> 00:17:30.559
Now that we have a track record of how AI
261
00:17:30.680 --> 00:17:34.000
is being used in the field. I still do think
262
00:17:34.039 --> 00:17:37.240
that we need we need government to take some action here,
263
00:17:37.960 --> 00:17:43.279
but we need policy that's going to be that's good,
264
00:17:43.519 --> 00:17:46.119
that's good policy that is going to support innovation and
265
00:17:46.119 --> 00:17:50.200
also provide the kind of guardrails. You know, one issue
266
00:17:50.480 --> 00:17:53.960
in ten forty seven that I think hasn't gotten enough
267
00:17:54.000 --> 00:17:59.440
attention is that it, you know, seeks on the one hand,
268
00:17:59.599 --> 00:18:05.480
to control the kind of grave potential harms that could
269
00:18:05.960 --> 00:18:09.400
come out of these most powerful AI models, But in
270
00:18:09.480 --> 00:18:12.240
other aspects of the bill, it's looking at things that
271
00:18:12.279 --> 00:18:18.559
are more routine, guard of variety and things that that
272
00:18:18.759 --> 00:18:22.880
can be expected from AI models. And I think that's
273
00:18:22.960 --> 00:18:25.960
where you're really seeing some of the pushback from industry
274
00:18:26.160 --> 00:18:32.119
that you know, an obligation to report any incident that arises,
275
00:18:32.480 --> 00:18:38.559
or an obligation to avoid any sort of impact on
276
00:18:38.640 --> 00:18:43.200
public safety where that term isn't even defined, could lead
277
00:18:43.279 --> 00:18:49.200
to liability, is really threatening those who are looking to
278
00:18:49.400 --> 00:18:53.000
build the more, more powerful AI models that are capable
279
00:18:53.000 --> 00:18:55.000
of doing a lot of things, a lot of really
280
00:18:55.079 --> 00:19:01.359
positive things for our society right now. And and that's why,
281
00:19:01.480 --> 00:19:04.160
you know, an effort like this needs needs more attention,
282
00:19:04.480 --> 00:19:08.440
it needs more more focus, and we really have to
283
00:19:08.880 --> 00:19:10.880
think about whether it's the right approach to go.
284
00:19:11.720 --> 00:19:16.079
And just curious, Paul, the legislatures here in our in
285
00:19:16.119 --> 00:19:20.799
our state, California, have they reached out to like S I,
286
00:19:20.799 --> 00:19:23.839
I A or other you know, thought leaders out there
287
00:19:23.880 --> 00:19:26.839
in AI to kind of get their input when they
288
00:19:27.519 --> 00:19:30.480
kind of crafted this bill and then put it together.
289
00:19:30.640 --> 00:19:32.079
And or do they just.
290
00:19:33.519 --> 00:19:37.599
You know, rest on their own understanding of AI maybe
291
00:19:37.640 --> 00:19:41.400
even misinformation out there. Do they try to kind of
292
00:19:42.039 --> 00:19:45.200
get the community involved.
293
00:19:45.039 --> 00:19:47.400
In these decision makers decision making them.
294
00:19:47.440 --> 00:19:51.640
Sorry, Yeah, I think it's you know, there's certainly been
295
00:19:51.680 --> 00:19:57.119
outreach from industry and from other thought leaders, and this
296
00:19:57.160 --> 00:20:00.880
is the kind of bill that reading it does reflect
297
00:20:01.000 --> 00:20:05.039
engagement with folks out there who are really thinking seriously
298
00:20:05.079 --> 00:20:10.680
about this. But I also think it reflects an approach
299
00:20:12.000 --> 00:20:19.279
that doesn't account for it doesn't really account for the
300
00:20:19.319 --> 00:20:29.279
industry perspective, and reflects a view that really there's that
301
00:20:29.319 --> 00:20:35.000
the industry is always oppositional to legislation, and and for
302
00:20:35.039 --> 00:20:38.799
that reason, I think it is somewhat concerning. You know,
303
00:20:39.000 --> 00:20:44.119
certainly I represent industry, I work with companies in industry,
304
00:20:44.240 --> 00:20:50.559
leading AI developers, leading AI deployers. But I don't think
305
00:20:50.599 --> 00:20:53.119
that there are just two ways to go about everything.
306
00:20:54.160 --> 00:20:55.920
I think there are parts of this build that are
307
00:20:56.039 --> 00:20:59.440
very positive, but then there are parts that that really aren't.
308
00:21:00.400 --> 00:21:02.359
I've said before that I think the right way to
309
00:21:02.400 --> 00:21:04.880
look at this set of issues is at the federal
310
00:21:04.960 --> 00:21:09.880
level in a unified manner, in a manner where you
311
00:21:09.920 --> 00:21:14.720
also have the infrastructure, the expertise, and the compute resources
312
00:21:14.759 --> 00:21:19.279
necessary to work collaboratively with developers to try to find
313
00:21:19.319 --> 00:21:23.759
better ways and find ways to mitigate the risks associating
314
00:21:23.759 --> 00:21:28.359
with those models. But we are in the AI debate today,
315
00:21:29.519 --> 00:21:32.880
it is developed into something that seems more oppositional than
316
00:21:32.920 --> 00:21:36.279
I think it really should. I think there are there
317
00:21:36.279 --> 00:21:39.920
are a number of different ways to go about developing policy,
318
00:21:41.160 --> 00:21:45.039
and it's not a black and white situation, but this
319
00:21:45.119 --> 00:21:51.640
bill reflects a pretty far reflects a view that is
320
00:21:51.680 --> 00:21:54.680
supported by one end of that spectrum in the debate.
321
00:21:55.839 --> 00:21:56.359
Yeah, I.
322
00:21:57.920 --> 00:22:02.200
Definitely concur with that all looking at it and everything,
323
00:22:02.279 --> 00:22:06.200
and that's why we're glad we have organizations as as
324
00:22:06.319 --> 00:22:10.400
as yours, the s i A and to really keep
325
00:22:10.440 --> 00:22:13.519
us in form of what's going on with policy, but
326
00:22:13.559 --> 00:22:17.200
also innovation out there and really kind of you know,
327
00:22:17.359 --> 00:22:21.319
walking walking that that that great balance of innovation but
328
00:22:21.440 --> 00:22:25.079
also also safety as well, Paul, Before I let you go,
329
00:22:25.880 --> 00:22:27.599
is there anything else out there you'd like to share
330
00:22:27.880 --> 00:22:30.319
with the audience who are watching us and who are
331
00:22:30.319 --> 00:22:31.640
listening to us.
332
00:22:34.480 --> 00:22:37.599
You know, I think that we've covered it. This is
333
00:22:37.839 --> 00:22:42.519
a you know, California being so active in AI, and
334
00:22:42.599 --> 00:22:45.839
I think it's been very helpful for further in the debate.
335
00:22:46.720 --> 00:22:49.400
But we need to take care. We need to take
336
00:22:49.440 --> 00:22:54.559
care here. I think when we're talking about UH, we're
337
00:22:54.599 --> 00:22:59.359
talking about regulating things where we don't actually have the
338
00:22:59.440 --> 00:23:03.440
science in place and the technology in place and the
339
00:23:03.480 --> 00:23:06.599
standards in place to achieve some of the goals that
340
00:23:06.640 --> 00:23:08.480
we lay out, and we have to be careful there.
341
00:23:09.200 --> 00:23:15.000
We can't just we can't just demand that industry takes
342
00:23:15.000 --> 00:23:19.279
certain steps where those steps aren't possible. And the spectrum
343
00:23:19.279 --> 00:23:23.079
of liability here is so great that this kind of
344
00:23:23.079 --> 00:23:26.039
bill is going to have a real lasting effect on
345
00:23:27.119 --> 00:23:32.000
the USAI industry. And I really think we need to
346
00:23:32.000 --> 00:23:35.319
be looking at these things in a collaborative manner and
347
00:23:35.359 --> 00:23:39.920
trying to figure out what the best solutions are. I
348
00:23:40.559 --> 00:23:44.000
you know, would credit the California legislature for taking some
349
00:23:44.039 --> 00:23:49.039
really positive steps on other AI bills, and I think
350
00:23:49.119 --> 00:23:52.680
that it's great that they are so focused on this area,
351
00:23:55.079 --> 00:23:59.240
and I think I basically set my piece Paul.
352
00:23:59.599 --> 00:24:02.359
All right, what sounds good. We'll call that a mic
353
00:24:02.440 --> 00:24:05.119
drop moment. And thank you so much, Paul.
354
00:24:05.200 --> 00:24:07.720
And if ones like to know more information about some
355
00:24:07.720 --> 00:24:11.519
of the great things that they ask I A is doing,
356
00:24:11.599 --> 00:24:12.640
how can they keep up?
357
00:24:13.279 --> 00:24:16.319
Yeah, well, follow us. We are on We are on
358
00:24:16.319 --> 00:24:22.039
Twitter now known as x si A Policy. Check out
359
00:24:22.039 --> 00:24:27.039
our website SAA dot net. We are very active in
360
00:24:27.079 --> 00:24:34.000
the space and always issuing new thoughts and hopefully constructive recommendations.
361
00:24:33.200 --> 00:24:34.440
On the way forward.
362
00:24:35.079 --> 00:24:37.880
Well, I appreciate I appreciate the time, Paul, Thank you
363
00:24:37.920 --> 00:24:39.200
again for having.
364
00:24:39.000 --> 00:24:41.720
Me back on a huge shout out to my friend
365
00:24:41.720 --> 00:24:44.440
Paul Leis. I really enjoyed chatting with him. When it
366
00:24:44.480 --> 00:24:47.119
comes to AI and what it means for business is
367
00:24:47.119 --> 00:24:51.359
just a really balance of views of what's going on
368
00:24:51.960 --> 00:24:55.200
here in California and Paul is in d C. We're
369
00:24:55.240 --> 00:24:58.039
doing a lot of great things to make sure not
370
00:24:58.119 --> 00:25:01.519
only things are safe, but also the innovation is not steiny.
371
00:25:01.640 --> 00:25:04.359
So you want to hear your thoughts. Where do you
372
00:25:04.400 --> 00:25:07.599
come down on this? Please share with me. We would
373
00:25:07.640 --> 00:25:12.880
love to hear your view, all right, folks. It's time
374
00:25:12.920 --> 00:25:15.359
for me to make like a tree and get out
375
00:25:15.359 --> 00:25:17.799
of here. But until next time, do me a favor.
376
00:25:17.920 --> 00:25:21.799
Stay healthy, stay safe, and remember I love you guys
377
00:25:21.839 --> 00:25:22.359
your life.
378
00:25:22.599 --> 00:25:25.440
CEUs is right around the corner. Can wait to see
379
00:25:25.440 --> 00:25:26.599
everyone here. Take care.
380
00:25:28.920 --> 00:25:33.119
In this world of technology, things are ever changing, rearranging.
381
00:25:34.000 --> 00:25:36.799
You need someone to help you out. I know someone
382
00:25:36.799 --> 00:25:39.319
that can be alone. You'll be with Paul.
383
00:25:39.440 --> 00:25:40.079
I'm a day
384
00:25:40.200 --> 00:25:43.440
Slaine in the tech Zone.
1
00:00:00.760 --> 00:00:05.080
In this world of technology is ever change rearranging that
2
00:00:05.799 --> 00:00:07.200
you need someone to help you out.
3
00:00:08.000 --> 00:00:08.720
I know someone that.
4
00:00:08.839 --> 00:00:10.039
Can be alone.
5
00:00:10.119 --> 00:00:11.160
You'll be with Paul.
6
00:00:11.279 --> 00:00:12.720
I'm a day Lane.
7
00:00:12.599 --> 00:00:18.760
In the Tech Zone. Welcome back to the show me, Paul,
8
00:00:18.800 --> 00:00:19.760
I'm Dave Slaine.
9
00:00:19.800 --> 00:00:22.359
Thank you so much. This is our second segment. If
10
00:00:22.399 --> 00:00:25.839
you missed our first segment, check it out. And talked
11
00:00:25.839 --> 00:00:29.800
to Steve Yule from the Consumer Technology Association Foundation, and
12
00:00:29.839 --> 00:00:32.439
I also talked about social media and.
13
00:00:34.079 --> 00:00:37.119
Just how we have to be smarter with it.
14
00:00:37.679 --> 00:00:40.240
I mean, people are making money off of it because
15
00:00:40.280 --> 00:00:42.799
of our anger, and to me, that doesn't make any
16
00:00:42.840 --> 00:00:45.359
sense at Allso if you miss my rant, please please
17
00:00:45.399 --> 00:00:48.280
please check it out. Here in California, there is a
18
00:00:48.320 --> 00:00:51.159
bill that's on the governor's desk that's ready for him
19
00:00:51.159 --> 00:00:56.600
to sign that Some who agree with this bill are
20
00:00:56.640 --> 00:00:59.719
saying that it's needed to make sure there's from guard
21
00:01:00.520 --> 00:01:04.560
on AI and protections of their but there are also
22
00:01:04.799 --> 00:01:09.920
some who feel that this may steiny innovation. So we're
23
00:01:09.959 --> 00:01:15.040
going to talk to an organization that is urging Governor
24
00:01:15.159 --> 00:01:19.920
Newsom to veto this bill that is on his desk,
25
00:01:19.959 --> 00:01:22.319
and we're going to find out why. And also this
26
00:01:22.439 --> 00:01:26.959
Organization A. They were not alone in reaching out to
27
00:01:27.519 --> 00:01:31.879
the governor. The Consumer Technology Association was too. We're going
28
00:01:31.959 --> 00:01:35.400
to find out about this and more with my next guest.
29
00:01:36.760 --> 00:01:39.519
I am so delighted to welcome back to the program.
30
00:01:39.719 --> 00:01:43.079
Our good friend, Paul Leckis is the senior vice president
31
00:01:43.079 --> 00:01:44.760
over at si A.
32
00:01:45.560 --> 00:01:48.280
Paul, what's going on? How are you, Paul?
33
00:01:48.359 --> 00:01:50.560
I'm doing great. It is such a pleasure to be
34
00:01:50.640 --> 00:01:51.359
back here with you.
35
00:01:51.879 --> 00:01:52.959
Hey, it's great to chat with you.
36
00:01:53.120 --> 00:01:57.560
I always enjoy our chats because we talk about some
37
00:01:57.719 --> 00:02:01.159
things that we should all be aware of, especially when
38
00:02:01.159 --> 00:02:04.040
it comes to the not the Internet of things, but
39
00:02:04.159 --> 00:02:06.120
the intelligence.
40
00:02:05.760 --> 00:02:08.400
Of things AI, AI and everything.
41
00:02:08.479 --> 00:02:11.000
Plus I really like your name too, so you know,
42
00:02:11.039 --> 00:02:14.599
I gotta gotta give you problems with that, Paul. Let's
43
00:02:14.599 --> 00:02:18.439
get down to, as we say, not business but business,
44
00:02:18.439 --> 00:02:24.439
all right, Paul. So, there is a new I guess
45
00:02:24.599 --> 00:02:27.919
bill that's out here in California. So it affects me
46
00:02:28.240 --> 00:02:32.680
and for our listeners who are regional here in California,
47
00:02:32.719 --> 00:02:37.479
but also our listeners who are worldwide and even here domestically.
48
00:02:38.240 --> 00:02:41.199
Why don't we give them a little background about this
49
00:02:41.360 --> 00:02:46.680
bill and what it's aimed to do, and why many
50
00:02:46.759 --> 00:02:51.199
are urging Governor Newsom to veto this.
51
00:02:52.159 --> 00:02:56.960
Yeah, absolutely so. I We're talking about a b called
52
00:02:57.159 --> 00:03:03.199
SBE which has now past the California legislature. It is
53
00:03:03.280 --> 00:03:08.520
on Governor Newsom's desk to sign or to veto, and
54
00:03:08.840 --> 00:03:13.800
it is titled the Safe and Secure Innovation for Printier
55
00:03:14.000 --> 00:03:19.800
Artificial Intelligence Models Act. It is quite a mouthful, but
56
00:03:20.199 --> 00:03:23.919
essentially what this bill aims to do is provide a
57
00:03:23.960 --> 00:03:29.039
framework to address concerns around the safety and security of
58
00:03:29.120 --> 00:03:33.319
the largest models out there. We call them frontier AI models,
59
00:03:33.360 --> 00:03:37.599
basically models that are on the frontier of the most
60
00:03:37.680 --> 00:03:41.360
advanced AI models today, the models that are powering things
61
00:03:41.400 --> 00:03:47.319
like CHATGPT, Gemini, and other services that people use regularly.
62
00:03:48.759 --> 00:03:52.719
What the bill would do is impose a number of requirements,
63
00:03:53.000 --> 00:03:57.919
primarily on the developers of these models, requirements that they
64
00:03:57.960 --> 00:04:02.159
would need to undertake before the developing these models, after
65
00:04:02.240 --> 00:04:05.240
developing these models, before they're put into service or made
66
00:04:05.240 --> 00:04:08.879
available to the public, and it would focus on a
67
00:04:08.919 --> 00:04:12.120
couple of different types of risks that are associated with
68
00:04:12.360 --> 00:04:15.599
these sorts of models. Some of them are described as
69
00:04:15.639 --> 00:04:20.160
critical harms, and these are things that include the creation
70
00:04:20.480 --> 00:04:25.399
of chemical, biological, radiological, or nuclear weapons using these models,
71
00:04:26.560 --> 00:04:30.319
the potential for mass casualties of at least five hundred
72
00:04:30.360 --> 00:04:36.920
million dollars, the potential for potential for other grave harms
73
00:04:36.959 --> 00:04:40.120
to public safety and security, and I'm quoting a little
74
00:04:40.120 --> 00:04:45.120
bit from the bill itself. There are also requirements around
75
00:04:45.480 --> 00:04:50.360
what are called AI safety incidents, and these are a
76
00:04:50.439 --> 00:04:53.759
variety of different things that could happen by using an
77
00:04:53.759 --> 00:05:00.199
AI model, including malicious use in indvertain release, unauthorized access,
78
00:05:01.160 --> 00:05:05.839
the behavior that is not intended from the models, either
79
00:05:05.920 --> 00:05:08.959
because these models act in ways that are unpredictable or
80
00:05:09.000 --> 00:05:14.240
because they're misused. And in addition to putting these requirements
81
00:05:14.240 --> 00:05:18.959
on developers, including doing third party audits, conducting assessments of
82
00:05:19.040 --> 00:05:25.639
the potential for risks, reporting any incidents that a developer
83
00:05:25.680 --> 00:05:29.879
is aware of, it would also establish a framework for
84
00:05:30.240 --> 00:05:34.759
the Attorney General to bring suits against developers. And I
85
00:05:34.759 --> 00:05:37.720
think this is one of the reasons why the bill
86
00:05:37.759 --> 00:05:41.759
has been so contentious, and not to get to ahead
87
00:05:41.759 --> 00:05:43.720
of myself, the bill has a number of other things,
88
00:05:43.800 --> 00:05:49.439
but you know, we are talking about really advanced AI
89
00:05:49.560 --> 00:05:54.759
models where the techniques and the technologies to accurately measure
90
00:05:55.360 --> 00:05:59.319
the risk of harm of misuse of misappropriation and so
91
00:05:59.399 --> 00:06:04.519
forth are still being developed. We don't have any concrete
92
00:06:04.560 --> 00:06:10.639
standards that developers can use, and out of the industry community,
93
00:06:10.800 --> 00:06:14.399
there's grave concern that if a bill like this were
94
00:06:14.439 --> 00:06:19.720
to be enacted into law, it could essentially halt innovation
95
00:06:20.040 --> 00:06:24.639
because the specter of liability is too great, it's too ambiguous,
96
00:06:24.839 --> 00:06:31.519
and developers don't have a sense of, you know, what
97
00:06:31.759 --> 00:06:35.519
would be appropriate here. There's a number of other concerns
98
00:06:35.560 --> 00:06:38.959
I think with the general structure. But one other thing
99
00:06:39.000 --> 00:06:42.160
to note is, you know, this is the sort of
100
00:06:42.240 --> 00:06:45.480
thing that the federal government is looking at very closely
101
00:06:45.560 --> 00:06:49.800
right now. They are looking at ways to work collectively
102
00:06:49.959 --> 00:06:54.439
collaboratively with industry to try to develop the right kind
103
00:06:54.439 --> 00:06:59.560
of standards that we need to investigate, interrogate these models
104
00:06:59.560 --> 00:07:02.800
and see what are potentially capable of. It's something that's
105
00:07:03.040 --> 00:07:07.199
happening at the federal level, something that's happening internationally, and
106
00:07:08.639 --> 00:07:10.680
you know, there is a view that it is too
107
00:07:10.839 --> 00:07:16.639
premature to build a new system of liability here while
108
00:07:16.639 --> 00:07:19.600
we're still kind of building the airplane.
109
00:07:20.319 --> 00:07:21.680
That definitely makes sense, Paul.
110
00:07:22.839 --> 00:07:25.399
If we have a standard here in California, and let's
111
00:07:25.439 --> 00:07:30.759
say that the federal government passes a law, then the
112
00:07:30.800 --> 00:07:34.160
federal law will supersede the state law.
113
00:07:34.399 --> 00:07:37.040
I assume there are to be a lot of confusion
114
00:07:37.800 --> 00:07:39.759
right when it comes when it comes to a lot
115
00:07:39.759 --> 00:07:40.040
of things.
116
00:07:40.040 --> 00:07:44.279
And and looking at at this letter that was sent
117
00:07:44.360 --> 00:07:49.120
to Governor Newsom, it was not only your organization, but
118
00:07:49.199 --> 00:07:53.800
also a powerhouse of the Consumer Technology Association too, And
119
00:07:54.399 --> 00:07:55.519
what was it like to.
120
00:07:56.040 --> 00:08:00.319
Kind of have them get on board with with so
121
00:08:00.360 --> 00:08:05.000
with SIIA, to really come together and collaborate to talk
122
00:08:05.040 --> 00:08:09.319
about some things that needed to be put in front
123
00:08:09.360 --> 00:08:12.120
of the governor before he made a decision on what
124
00:08:12.199 --> 00:08:13.439
to do with this actual bill.
125
00:08:13.959 --> 00:08:18.279
Yeah. So, the Conservative Technology Association CTA is one of
126
00:08:18.279 --> 00:08:21.600
the organizations in the tech space that we collaborate with
127
00:08:21.720 --> 00:08:23.879
a lot. We work with them a lot on issues
128
00:08:24.839 --> 00:08:29.199
and what they're hearing from their members. There's some overlapping
129
00:08:29.319 --> 00:08:31.680
members of our two associations, but they have a number
130
00:08:31.720 --> 00:08:34.360
of members that are not part of SIA and vice versa.
131
00:08:35.480 --> 00:08:37.559
They're hearing the same sort of things that we're hearing
132
00:08:38.200 --> 00:08:40.879
that there is concern that this sort of a bill
133
00:08:41.080 --> 00:08:45.240
being enacted at a state level could really stand in
134
00:08:45.279 --> 00:08:48.679
the way of the kind of innovation we need. But
135
00:08:48.879 --> 00:08:50.519
another thing that we raised in the letter that I
136
00:08:50.519 --> 00:08:53.519
think is really important is that nobody is going into
137
00:08:53.559 --> 00:08:59.960
this thinking that AI models are inherently safe or don't
138
00:09:00.120 --> 00:09:04.279
need some sort of guard rails. Frankly, I think the
139
00:09:04.399 --> 00:09:10.320
safety and security questions that we're talking about at least
140
00:09:10.320 --> 00:09:13.679
as significant is kind of the immediate issues that people
141
00:09:13.679 --> 00:09:16.879
are addressing right now and looking at right now around
142
00:09:16.879 --> 00:09:21.200
the socio economic impact of AI models and discrete situations.
143
00:09:22.159 --> 00:09:27.639
But safety and security is of paramount importance, and if
144
00:09:27.679 --> 00:09:29.639
we are going to develop a framework, we need to
145
00:09:29.679 --> 00:09:34.240
do this in the right way. It is I think
146
00:09:34.279 --> 00:09:40.399
that contrasting this with the area of privacy, and I'm
147
00:09:40.440 --> 00:09:43.960
sure many of your listeners are familiar with the fact
148
00:09:44.000 --> 00:09:46.960
that California has a privacy law and the federal government
149
00:09:47.000 --> 00:09:51.960
does not, and California has really helped to move forward
150
00:09:52.080 --> 00:09:54.600
the debate in the United States around privacy. We are
151
00:09:54.720 --> 00:09:58.039
very hopeful there would be a federal law, but when
152
00:09:58.080 --> 00:10:00.799
we're in the realm of AI safety, this is something
153
00:10:00.879 --> 00:10:05.360
very different. It is it makes a lot of sense
154
00:10:05.440 --> 00:10:09.559
for states and other bodies to be looking at ways
155
00:10:09.600 --> 00:10:12.480
in which AI is used in their jurisdiction, and the
156
00:10:12.480 --> 00:10:16.720
states and counties and cities, and you know, to put
157
00:10:16.759 --> 00:10:20.320
guardrails around things like the use of AI in making
158
00:10:20.360 --> 00:10:23.600
employment decisions or criminal justice or those sorts of things
159
00:10:23.600 --> 00:10:27.399
that people touch every day. But here we're talking about
160
00:10:27.440 --> 00:10:30.240
something that is really at the threshold of where science
161
00:10:30.279 --> 00:10:34.480
and research is and we are really concerned about the
162
00:10:34.519 --> 00:10:38.639
impact of this on innovation. I think that will be
163
00:10:38.720 --> 00:10:44.200
felt more acutely in California, given that this is a
164
00:10:44.240 --> 00:10:48.000
bill coming out of California. But it's also hard to
165
00:10:48.039 --> 00:10:50.279
see how the impact of a bill like this would
166
00:10:50.320 --> 00:10:54.159
be limited to just California. In some ways, it would
167
00:10:54.159 --> 00:10:58.039
be a de facto regulation of AI across the country
168
00:10:59.799 --> 00:11:04.919
in number of cases. We're concerned that this is something
169
00:11:04.960 --> 00:11:08.440
that CTA agrees with as well, that this is going
170
00:11:08.519 --> 00:11:16.240
to lead to really hurt our own industry, and our
171
00:11:16.320 --> 00:11:20.399
industry is working some There may be some naysayers out there,
172
00:11:20.440 --> 00:11:22.279
but the industry is really working to try to be
173
00:11:22.320 --> 00:11:25.879
as safe and secure as possible. And we can't say
174
00:11:25.879 --> 00:11:30.240
the same for AI models that are being developed potentially
175
00:11:30.240 --> 00:11:35.240
in other countries that will probably swoop in and fill
176
00:11:35.279 --> 00:11:38.360
the gap that is created if if we're not able
177
00:11:38.399 --> 00:11:44.240
to continue to release advanced models out of US developers.
178
00:11:45.440 --> 00:11:47.480
You know, Paul that it makes a lot of sense.
179
00:11:47.519 --> 00:11:52.799
Here's here's kind of the I guess concern when it
180
00:11:52.840 --> 00:11:55.759
comes to all all things that AI that we you
181
00:11:55.799 --> 00:11:59.600
know that that we have now we all know that
182
00:11:59.679 --> 00:12:01.519
something needs to be done. You know, you can look
183
00:12:01.559 --> 00:12:06.159
at even though there's this election cycle, how AI has
184
00:12:06.240 --> 00:12:11.440
been used to just show so much disinformation, deep fakes,
185
00:12:11.879 --> 00:12:13.200
all these other things.
186
00:12:13.000 --> 00:12:14.039
That that are going on.
187
00:12:14.759 --> 00:12:19.080
And what can be done I guess for the public
188
00:12:20.039 --> 00:12:25.480
now to kind of, I guess, you know, protect themselves
189
00:12:25.879 --> 00:12:28.919
while while the governments are trying to figure out a
190
00:12:28.960 --> 00:12:34.480
good policy that can number one, won't stifle innovation, but
191
00:12:34.600 --> 00:12:38.080
number two, we'll have some some guard rails and safety,
192
00:12:38.440 --> 00:12:41.720
you know, for for consumers, businesses, and and so forth.
193
00:12:42.240 --> 00:12:49.000
What can consumers, businesses, organizations do now to make sure
194
00:12:49.559 --> 00:12:52.840
that that that they're really having some self accountability to
195
00:12:52.960 --> 00:12:55.639
make sure that they are protecting themselves.
196
00:12:55.919 --> 00:12:59.720
Yeah, and that's such a great question. And you know
197
00:12:59.759 --> 00:13:04.000
that we of the legislatures outside of California tend to
198
00:13:04.039 --> 00:13:10.840
move slowly. And you know, Congress, I think there's a
199
00:13:10.840 --> 00:13:15.080
lot of proposals before Congress that would be really helpful here.
200
00:13:16.559 --> 00:13:19.600
But you know, California is moving very fast. California has
201
00:13:19.600 --> 00:13:21.679
actually taken a number of steps that have been really
202
00:13:21.720 --> 00:13:25.440
positive in the AI space. But you know, what can
203
00:13:25.480 --> 00:13:32.000
we do beyond relying on our elected officials to actually
204
00:13:33.320 --> 00:13:37.879
protect ourselves individually, try to make use of these tools
205
00:13:38.000 --> 00:13:42.639
in a positive way. You'd mentioned misinformation. I think that
206
00:13:42.639 --> 00:13:45.320
that is a concern that people are seeing all the time,
207
00:13:45.600 --> 00:13:51.200
and you know it is I think on that front,
208
00:13:52.519 --> 00:13:55.759
people have to be smart consumers, they have to be
209
00:13:56.759 --> 00:14:00.080
critical thinkers of what they see on the Internet. I
210
00:14:00.120 --> 00:14:03.600
think it's it's really hard to it's really hard for
211
00:14:03.679 --> 00:14:08.440
legislators to try to stop the spread of misinformation online,
212
00:14:08.480 --> 00:14:11.240
even AI enabled misinformation. I think that we've seen some
213
00:14:11.279 --> 00:14:15.879
progress in a few areas and I'm hopeful that will continue.
214
00:14:16.200 --> 00:14:22.240
And that includes deep fakes of non consensual intimate imagery.
215
00:14:22.559 --> 00:14:25.039
We've seen a number of state laws, including in California,
216
00:14:25.840 --> 00:14:29.799
that are really focused on people who are uh uh,
217
00:14:31.159 --> 00:14:33.360
who are the bad actors here, the people who are
218
00:14:34.039 --> 00:14:39.960
generating and disseminating those kind of harmful images. Uh Regulation
219
00:14:40.000 --> 00:14:42.039
of content is something that's very difficult for the government
220
00:14:42.039 --> 00:14:44.879
to do though, because of the First Amendment, Except in
221
00:14:45.120 --> 00:14:49.519
certain discrete uh, situations where there's a grave threat to
222
00:14:49.559 --> 00:14:54.799
public to public interests. So people need to be spark consumers.
223
00:14:55.039 --> 00:15:01.320
And I think, you know, you don't want to put
224
00:15:01.360 --> 00:15:04.519
the onus on individuals, but we need to look with
225
00:15:04.600 --> 00:15:07.279
a good degree of skepticism at information that we're seeing
226
00:15:07.320 --> 00:15:10.600
out there. And I think there is some value in
227
00:15:11.519 --> 00:15:17.600
relying on, you know, reputable sources of information to verify
228
00:15:17.799 --> 00:15:21.840
anything that you see that seems to be shocking or
229
00:15:22.440 --> 00:15:26.639
concerning or you know, out of the norm. I think
230
00:15:26.679 --> 00:15:29.879
that you know, you're seeing in the news space that
231
00:15:30.519 --> 00:15:34.519
reputable news publications are probably seen as more authoritative now,
232
00:15:36.000 --> 00:15:38.679
and that's because there's a lot of information floating out
233
00:15:38.720 --> 00:15:41.000
there that's not correct, and some of it is being
234
00:15:41.039 --> 00:15:44.399
circulated inadvertently, and some of it is intentional. There are
235
00:15:44.519 --> 00:15:49.559
malicious actors out there, for sure. I think that you know,
236
00:15:49.679 --> 00:15:52.600
businesses that are looking to adopt AI can take a
237
00:15:52.679 --> 00:15:55.039
number of steps, and there are frameworks out there to
238
00:15:55.320 --> 00:15:58.559
assess the risk of the tools that they're using to
239
00:15:58.639 --> 00:16:00.799
make sure that they are using them the right way.
240
00:16:01.919 --> 00:16:09.240
I think everyone needs to be careful in adopting AI tools.
241
00:16:09.320 --> 00:16:11.480
I think they can do a lot of good, but
242
00:16:12.240 --> 00:16:15.639
we need to approach them in a reasonable way. Some
243
00:16:15.679 --> 00:16:17.720
of the guidances coming out of the National Institute of
244
00:16:17.720 --> 00:16:22.519
Standards of Technology is NISSED is incredibly useful for those
245
00:16:22.559 --> 00:16:25.279
who are developing AI as well as for those who
246
00:16:25.360 --> 00:16:29.039
are looking to purchase AI or apply AI tools in
247
00:16:29.080 --> 00:16:33.320
different places. And I think that that kind of guidance,
248
00:16:33.879 --> 00:16:39.200
which is being developed in collaboration with academia, civil society,
249
00:16:39.279 --> 00:16:44.200
and industry, is really useful. There is you know, we've
250
00:16:44.240 --> 00:16:47.159
done a lot of work about how to safely incorporate
251
00:16:47.200 --> 00:16:51.360
AI and classrooms, and I think that that the bar
252
00:16:51.480 --> 00:16:54.399
is being raised in that space, as in a number
253
00:16:54.480 --> 00:16:59.720
of other spaces, and I'm optimistic that we're actually from
254
00:16:59.759 --> 00:17:06.759
an industry perspective, creating tools and applications that are safer
255
00:17:06.799 --> 00:17:12.039
and more secure and avoid a number of the concerns
256
00:17:12.119 --> 00:17:15.680
or mitigate the concerns that were present in earlier AI models.
257
00:17:16.559 --> 00:17:20.519
Now we're not totally free of things like algorithmic bias,
258
00:17:21.680 --> 00:17:24.519
but many of the applications that are out there are
259
00:17:24.519 --> 00:17:27.359
doing a much better job with those sorts of things.
260
00:17:27.359 --> 00:17:30.559
Now that we have a track record of how AI
261
00:17:30.680 --> 00:17:34.000
is being used in the field. I still do think
262
00:17:34.039 --> 00:17:37.240
that we need we need government to take some action here,
263
00:17:37.960 --> 00:17:43.279
but we need policy that's going to be that's good,
264
00:17:43.519 --> 00:17:46.119
that's good policy that is going to support innovation and
265
00:17:46.119 --> 00:17:50.200
also provide the kind of guardrails. You know, one issue
266
00:17:50.480 --> 00:17:53.960
in ten forty seven that I think hasn't gotten enough
267
00:17:54.000 --> 00:17:59.440
attention is that it, you know, seeks on the one hand,
268
00:17:59.599 --> 00:18:05.480
to control the kind of grave potential harms that could
269
00:18:05.960 --> 00:18:09.400
come out of these most powerful AI models, But in
270
00:18:09.480 --> 00:18:12.240
other aspects of the bill, it's looking at things that
271
00:18:12.279 --> 00:18:18.559
are more routine, guard of variety and things that that
272
00:18:18.759 --> 00:18:22.880
can be expected from AI models. And I think that's
273
00:18:22.960 --> 00:18:25.960
where you're really seeing some of the pushback from industry
274
00:18:26.160 --> 00:18:32.119
that you know, an obligation to report any incident that arises,
275
00:18:32.480 --> 00:18:38.559
or an obligation to avoid any sort of impact on
276
00:18:38.640 --> 00:18:43.200
public safety where that term isn't even defined, could lead
277
00:18:43.279 --> 00:18:49.200
to liability, is really threatening those who are looking to
278
00:18:49.400 --> 00:18:53.000
build the more, more powerful AI models that are capable
279
00:18:53.000 --> 00:18:55.000
of doing a lot of things, a lot of really
280
00:18:55.079 --> 00:19:01.359
positive things for our society right now. And and that's why,
281
00:19:01.480 --> 00:19:04.160
you know, an effort like this needs needs more attention,
282
00:19:04.480 --> 00:19:08.440
it needs more more focus, and we really have to
283
00:19:08.880 --> 00:19:10.880
think about whether it's the right approach to go.
284
00:19:11.720 --> 00:19:16.079
And just curious, Paul, the legislatures here in our in
285
00:19:16.119 --> 00:19:20.799
our state, California, have they reached out to like S I,
286
00:19:20.799 --> 00:19:23.839
I A or other you know, thought leaders out there
287
00:19:23.880 --> 00:19:26.839
in AI to kind of get their input when they
288
00:19:27.519 --> 00:19:30.480
kind of crafted this bill and then put it together.
289
00:19:30.640 --> 00:19:32.079
And or do they just.
290
00:19:33.519 --> 00:19:37.599
You know, rest on their own understanding of AI maybe
291
00:19:37.640 --> 00:19:41.400
even misinformation out there. Do they try to kind of
292
00:19:42.039 --> 00:19:45.200
get the community involved.
293
00:19:45.039 --> 00:19:47.400
In these decision makers decision making them.
294
00:19:47.440 --> 00:19:51.640
Sorry, Yeah, I think it's you know, there's certainly been
295
00:19:51.680 --> 00:19:57.119
outreach from industry and from other thought leaders, and this
296
00:19:57.160 --> 00:20:00.880
is the kind of bill that reading it does reflect
297
00:20:01.000 --> 00:20:05.039
engagement with folks out there who are really thinking seriously
298
00:20:05.079 --> 00:20:10.680
about this. But I also think it reflects an approach
299
00:20:12.000 --> 00:20:19.279
that doesn't account for it doesn't really account for the
300
00:20:19.319 --> 00:20:29.279
industry perspective, and reflects a view that really there's that
301
00:20:29.319 --> 00:20:35.000
the industry is always oppositional to legislation, and and for
302
00:20:35.039 --> 00:20:38.799
that reason, I think it is somewhat concerning. You know,
303
00:20:39.000 --> 00:20:44.119
certainly I represent industry, I work with companies in industry,
304
00:20:44.240 --> 00:20:50.559
leading AI developers, leading AI deployers. But I don't think
305
00:20:50.599 --> 00:20:53.119
that there are just two ways to go about everything.
306
00:20:54.160 --> 00:20:55.920
I think there are parts of this build that are
307
00:20:56.039 --> 00:20:59.440
very positive, but then there are parts that that really aren't.
308
00:21:00.400 --> 00:21:02.359
I've said before that I think the right way to
309
00:21:02.400 --> 00:21:04.880
look at this set of issues is at the federal
310
00:21:04.960 --> 00:21:09.880
level in a unified manner, in a manner where you
311
00:21:09.920 --> 00:21:14.720
also have the infrastructure, the expertise, and the compute resources
312
00:21:14.759 --> 00:21:19.279
necessary to work collaboratively with developers to try to find
313
00:21:19.319 --> 00:21:23.759
better ways and find ways to mitigate the risks associating
314
00:21:23.759 --> 00:21:28.359
with those models. But we are in the AI debate today,
315
00:21:29.519 --> 00:21:32.880
it is developed into something that seems more oppositional than
316
00:21:32.920 --> 00:21:36.279
I think it really should. I think there are there
317
00:21:36.279 --> 00:21:39.920
are a number of different ways to go about developing policy,
318
00:21:41.160 --> 00:21:45.039
and it's not a black and white situation, but this
319
00:21:45.119 --> 00:21:51.640
bill reflects a pretty far reflects a view that is
320
00:21:51.680 --> 00:21:54.680
supported by one end of that spectrum in the debate.
321
00:21:55.839 --> 00:21:56.359
Yeah, I.
322
00:21:57.920 --> 00:22:02.200
Definitely concur with that all looking at it and everything,
323
00:22:02.279 --> 00:22:06.200
and that's why we're glad we have organizations as as
324
00:22:06.319 --> 00:22:10.400
as yours, the s i A and to really keep
325
00:22:10.440 --> 00:22:13.519
us in form of what's going on with policy, but
326
00:22:13.559 --> 00:22:17.200
also innovation out there and really kind of you know,
327
00:22:17.359 --> 00:22:21.319
walking walking that that that great balance of innovation but
328
00:22:21.440 --> 00:22:25.079
also also safety as well, Paul, Before I let you go,
329
00:22:25.880 --> 00:22:27.599
is there anything else out there you'd like to share
330
00:22:27.880 --> 00:22:30.319
with the audience who are watching us and who are
331
00:22:30.319 --> 00:22:31.640
listening to us.
332
00:22:34.480 --> 00:22:37.599
You know, I think that we've covered it. This is
333
00:22:37.839 --> 00:22:42.519
a you know, California being so active in AI, and
334
00:22:42.599 --> 00:22:45.839
I think it's been very helpful for further in the debate.
335
00:22:46.720 --> 00:22:49.400
But we need to take care. We need to take
336
00:22:49.440 --> 00:22:54.559
care here. I think when we're talking about UH, we're
337
00:22:54.599 --> 00:22:59.359
talking about regulating things where we don't actually have the
338
00:22:59.440 --> 00:23:03.440
science in place and the technology in place and the
339
00:23:03.480 --> 00:23:06.599
standards in place to achieve some of the goals that
340
00:23:06.640 --> 00:23:08.480
we lay out, and we have to be careful there.
341
00:23:09.200 --> 00:23:15.000
We can't just we can't just demand that industry takes
342
00:23:15.000 --> 00:23:19.279
certain steps where those steps aren't possible. And the spectrum
343
00:23:19.279 --> 00:23:23.079
of liability here is so great that this kind of
344
00:23:23.079 --> 00:23:26.039
bill is going to have a real lasting effect on
345
00:23:27.119 --> 00:23:32.000
the USAI industry. And I really think we need to
346
00:23:32.000 --> 00:23:35.319
be looking at these things in a collaborative manner and
347
00:23:35.359 --> 00:23:39.920
trying to figure out what the best solutions are. I
348
00:23:40.559 --> 00:23:44.000
you know, would credit the California legislature for taking some
349
00:23:44.039 --> 00:23:49.039
really positive steps on other AI bills, and I think
350
00:23:49.119 --> 00:23:52.680
that it's great that they are so focused on this area,
351
00:23:55.079 --> 00:23:59.240
and I think I basically set my piece Paul.
352
00:23:59.599 --> 00:24:02.359
All right, what sounds good. We'll call that a mic
353
00:24:02.440 --> 00:24:05.119
drop moment. And thank you so much, Paul.
354
00:24:05.200 --> 00:24:07.720
And if ones like to know more information about some
355
00:24:07.720 --> 00:24:11.519
of the great things that they ask I A is doing,
356
00:24:11.599 --> 00:24:12.640
how can they keep up?
357
00:24:13.279 --> 00:24:16.319
Yeah, well, follow us. We are on We are on
358
00:24:16.319 --> 00:24:22.039
Twitter now known as x si A Policy. Check out
359
00:24:22.039 --> 00:24:27.039
our website SAA dot net. We are very active in
360
00:24:27.079 --> 00:24:34.000
the space and always issuing new thoughts and hopefully constructive recommendations.
361
00:24:33.200 --> 00:24:34.440
On the way forward.
362
00:24:35.079 --> 00:24:37.880
Well, I appreciate I appreciate the time, Paul, Thank you
363
00:24:37.920 --> 00:24:39.200
again for having.
364
00:24:39.000 --> 00:24:41.720
Me back on a huge shout out to my friend
365
00:24:41.720 --> 00:24:44.440
Paul Leis. I really enjoyed chatting with him. When it
366
00:24:44.480 --> 00:24:47.119
comes to AI and what it means for business is
367
00:24:47.119 --> 00:24:51.359
just a really balance of views of what's going on
368
00:24:51.960 --> 00:24:55.200
here in California and Paul is in d C. We're
369
00:24:55.240 --> 00:24:58.039
doing a lot of great things to make sure not
370
00:24:58.119 --> 00:25:01.519
only things are safe, but also the innovation is not steiny.
371
00:25:01.640 --> 00:25:04.359
So you want to hear your thoughts. Where do you
372
00:25:04.400 --> 00:25:07.599
come down on this? Please share with me. We would
373
00:25:07.640 --> 00:25:12.880
love to hear your view, all right, folks. It's time
374
00:25:12.920 --> 00:25:15.359
for me to make like a tree and get out
375
00:25:15.359 --> 00:25:17.799
of here. But until next time, do me a favor.
376
00:25:17.920 --> 00:25:21.799
Stay healthy, stay safe, and remember I love you guys
377
00:25:21.839 --> 00:25:22.359
your life.
378
00:25:22.599 --> 00:25:25.440
CEUs is right around the corner. Can wait to see
379
00:25:25.440 --> 00:25:26.599
everyone here. Take care.
380
00:25:28.920 --> 00:25:33.119
In this world of technology, things are ever changing, rearranging.
381
00:25:34.000 --> 00:25:36.799
You need someone to help you out. I know someone
382
00:25:36.799 --> 00:25:39.319
that can be alone. You'll be with Paul.
383
00:25:39.440 --> 00:25:40.079
I'm a day
384
00:25:40.200 --> 00:25:43.440
Slaine in the tech Zone.
