Latest / a16z AI Policy Brief / Catching Up on the Current Moment in AI Policy

Transcript

Catching Up on the Current Moment in AI Policy · a16z AI Policy Brief

0:00
AI policy is moving faster than any other policy set that I've
0:02
ever seen.
0:03
SPEAKER_00: And now we have moved into more meat on the bone.
0:06
We're starting to have real conversations within committees
0:09
about what could go in
0:10
SPEAKER_00: these packages.
0:11
SPEAKER_00: People want to protect the American public,
0:13
SPEAKER_00: which is ultimately the goal of these legislative proposals.
0:16
SPEAKER_00: Now you've got
0:17
SPEAKER_00: 10 companies,
0:18
the 10 biggest labs in the country or maybe in the world that
0:21
are coming up with
0:22
SPEAKER_00: their own benchmarking ideas.
0:24
SPEAKER_00: They have to be blessed by a regulator.
0:26
SPEAKER_00: So if you have a enterprising young startup that is trying to also build,
0:31
they can basically shape the rules to such that no one else is
0:35
going to be able to
0:35
SPEAKER_00: compete.
0:37
SPEAKER_00: That is how the regulatory capture game works.
0:39
SPEAKER_00: We need people to be active on this and share ideas and concepts,
0:44
SPEAKER_00: especially from the little tech community.
0:45
SPEAKER_00: The time is now.
0:53
SPEAKER_01: Welcome back to the A16Z AI Policy Brief.
0:56
SPEAKER_01: Today,
0:56
I'm joined by my colleague Colin McKeown for a conversation
0:59
about where AI policy
1:00
SPEAKER_01: stands right now.
1:01
SPEAKER_01: We talk through what's happening in Washington,
1:03
SPEAKER_01: where real consensus may be starting to emerge,
1:05
SPEAKER_01: how states are still driving a lot of the debate,
1:07
SPEAKER_01: and what all of it means for startups trying to build in a fast-changing
1:10
SPEAKER_01: environment.
1:11
SPEAKER_01: Here's our conversation.
1:14
SPEAKER_01: Colin, welcome back to the AI Policy Brief.
1:16
SPEAKER_00: It's good to be here.
1:16
SPEAKER_01: It's been a while.
1:17
SPEAKER_01: It has been a while.
1:18
SPEAKER_01: I think we last did this in September.
1:21
And it's hard to think back to exactly where we were in AI
1:24
policy at that moment.
1:25
SPEAKER_01: So I think over the course of the summer,
1:28
SPEAKER_01: states were moving toward the tail end of their legislative sessions.
1:31
We had a couple of bills that I think had passed the
1:33
legislature in California or
1:35
SPEAKER_01: awaiting signature,
1:36
SPEAKER_01: passed the legislature in New York awaiting signature.
1:38
SPEAKER_01: So we were really focused on state stuff.
1:40
SPEAKER_01: And then we had been through one round of
1:43
trying to see if Congress could move forward with some kind of
1:46
preemption.
1:47
SPEAKER_01: And that had failed over the summer in July.
1:49
SPEAKER_01: There were a couple more ahead in the fall.
1:52
SPEAKER_01: A lot has happened since that moment in September.
1:55
SPEAKER_01: So where do you think we are now relative to where we were then?
1:59
SPEAKER_00: Yeah, it's funny.
2:00
Everyone's always complaining about the law not following as
2:03
quickly as the tech does.
2:05
SPEAKER_00: And I feel like that's probably still the case.
2:08
The zeitgeist is absolutely like,
2:09
we don't want to have enough AI policy out there.
2:12
SPEAKER_00: I mean, I would say you probably agree with this from my perspective.
2:14
I mean,
2:15
AI policy is moving faster than any other policy set that I've
2:17
ever seen.
2:18
SPEAKER_01: It's moving incredibly, incredibly quickly.
2:20
SPEAKER_00: Avalanche of bills at the state level.
2:22
SPEAKER_00: Yeah.
2:22
SPEAKER_00: Avalanche of bills at the federal level.
2:24
SPEAKER_01: Yeah.
2:24
SPEAKER_01: Well, bills in both places.
2:26
And then I think the thing that I have kind of underweighted
2:28
even in the last
2:28
SPEAKER_01: couple of weeks is we're seeing real legislation at the state level.
2:31
SPEAKER_01: So we have chatbot bills in Oregon and Washington that have been enacted.
2:34
SPEAKER_01: Tennessee.
2:35
SPEAKER_01: Right.
2:35
SPEAKER_01: Yeah, exactly.
2:36
SPEAKER_01: Yeah.
2:36
SPEAKER_01: And then like serious proposals that are really moving forward.
2:39
SPEAKER_01: So I agree, though.
2:40
SPEAKER_01: It's not just the ideation.
2:42
It's the actual policymaking that's really moving forward at a
2:44
rapid pace.
2:45
SPEAKER_00: Yeah.
2:45
SPEAKER_00: I also think it's worth,
2:47
SPEAKER_00: you know,
2:47
SPEAKER_00: just mentioning we talk about this often,
2:49
but maybe the rest of sort of the people who are building this
2:52
space don't think
2:53
SPEAKER_00: about this.
2:53
SPEAKER_00: But
2:54
We're talking thousands of bills that have been introduced at
2:57
the state level.
2:57
SPEAKER_01: Yeah.
2:58
SPEAKER_00: Hundreds of bills that have been introduced at the federal level.
3:00
SPEAKER_00: Now, not all of those will be enacted in the law, but some of them will.
3:03
SPEAKER_01: Yeah.
3:04
And it's,
3:04
you know,
3:04
look,
3:04
I think that that's generally positive,
3:07
I think,
3:07
for the space.
3:08
SPEAKER_00: But there is obviously some downsides to that.
3:11
SPEAKER_01: Positive that things are moving that quickly.
3:12
SPEAKER_00: Well, it's positive.
3:12
SPEAKER_00: Yeah.
3:13
SPEAKER_00: I think it's positive that lawmakers are thinking about this.
3:15
SPEAKER_00: And it's very front of mind.
3:16
SPEAKER_00: I mean,
3:16
I think that there are certainly real issues with any
3:18
technology that need to be
3:19
SPEAKER_00: taken seriously.
3:20
SPEAKER_00: So I think it's positive.
3:21
There's also downsides to it because I don't think that some of
3:24
these policy sets
3:25
SPEAKER_00: are well thought through.
3:26
SPEAKER_00: We also don't know where the technology is headed and where it's going,
3:29
SPEAKER_00: which obviously we'll get into today.
3:32
SPEAKER_00: Maybe just a little bit,
3:33
SPEAKER_00: sort of back to your question,
3:35
SPEAKER_00: I would say things are progressing nicely.
3:39
And I would say,
3:40
especially at the federal level,
3:42
one of the principal things that you need is
3:45
for any piece of legislation to advance is you need to have
3:48
White House buy-in in
3:49
SPEAKER_00: this day and age.
3:50
SPEAKER_00: You know,
3:51
SPEAKER_00: back,
3:52
SPEAKER_00: I think,
3:53
SPEAKER_00: 20,
3:53
SPEAKER_00: 30 years ago,
3:54
it wasn't as important to have executive buy-in before the
3:59
Congress was moving
3:59
SPEAKER_00: legislation.
4:01
SPEAKER_00: Nowadays,
4:02
SPEAKER_00: The,
4:02
SPEAKER_00: you know,
4:03
the executive sets a lot of sort of the overarching policy and
4:06
then Congress fills
4:07
SPEAKER_00: in the gaps.
4:08
SPEAKER_01: And I think it's really important.
4:09
SPEAKER_01: So can you talk about the shift there?
4:10
SPEAKER_01: Because I would say,
4:11
SPEAKER_01: yeah,
4:12
SPEAKER_01: like to most of the general public,
4:14
SPEAKER_01: including people who follow AI policy fairly closely,
4:16
I think they would say the White House has been active on AI
4:19
policy for a long
4:20
SPEAKER_01: period of time and vocal on it for a long period of time.
4:22
But I do think there's been a shift in the last couple of
4:25
months from in terms of
4:26
SPEAKER_01: what you're saying,
4:27
SPEAKER_01: in terms of like the the real.
4:29
SPEAKER_01: policymaking component of that.
4:31
SPEAKER_01: So like, can you talk about what that shift has been like?
4:33
SPEAKER_00: Yeah,
4:33
SPEAKER_00: look,
4:33
SPEAKER_00: and I would say at the front end,
4:35
SPEAKER_00: the president deserves a lot of credit for the leadership here,
4:37
SPEAKER_00: right?
4:39
SPEAKER_00: The president,
4:39
SPEAKER_00: the president's team,
4:41
SPEAKER_00: Kratios,
4:42
SPEAKER_00: David Sachs,
4:43
SPEAKER_00: Sriram,
4:44
SPEAKER_00: our former colleague,
4:45
there are a number of people that deserve a lot of credit for
4:48
advancing the ball,
4:49
SPEAKER_00: I think,
4:49
SPEAKER_00: in a very intelligent way.
4:50
SPEAKER_01: Yeah.
4:52
SPEAKER_00: And it started really with the National AI Action Plan,
4:57
which was kind of a six-month effort at the beginning of the
5:00
administration to put
5:01
SPEAKER_00: forth a whole-of-government approach on how we're going to attack AI.
5:05
SPEAKER_00: And I thought that was really, really important.
5:08
SPEAKER_00: But that was step one.
5:10
SPEAKER_00: Step two was sort of the Genesis mission,
5:12
SPEAKER_00: which I think built in more on the infrastructure side of things,
5:16
SPEAKER_00: energy side.
5:17
SPEAKER_00: data centers, et cetera.
5:18
SPEAKER_00: Research.
5:19
SPEAKER_00: Research, right?
5:20
SPEAKER_00: But it's also become,
5:21
SPEAKER_00: I think,
5:21
SPEAKER_00: very apparent that the federal government needs to act from a legislative
5:25
SPEAKER_00: perspective.
5:26
And those proposals early on did not have specifics on what
5:30
should go in a legislative proposal.
5:32
SPEAKER_00: So now version three of this
5:34
SPEAKER_00: is the president and the White House,
5:36
SPEAKER_00: along with,
5:37
SPEAKER_00: I think,
5:38
SPEAKER_00: collaboration with the executive agencies,
5:40
SPEAKER_00: coming forward with proposals,
5:42
SPEAKER_00: real granular proposals on what should go in a legislative framework,
5:45
SPEAKER_00: almost like recommendations.
5:47
SPEAKER_01: Yeah, so that was flowing out of the December executive order.
5:51
SPEAKER_01: So the December executive order said,
5:53
SPEAKER_01: in 180 days,
5:55
SPEAKER_01: the White House will issue some guidance for Congress.
5:58
SPEAKER_01: So you understand in a way that I do not how a bill becomes a law.
6:01
SPEAKER_01: So my,
6:03
SPEAKER_01: like,
6:04
SPEAKER_01: just from the theory of how it works,
6:06
SPEAKER_01: president enforces the law,
6:08
SPEAKER_01: legislative branch writes the law.
6:09
SPEAKER_01: So I would have thought coming into this, this is on Congress.
6:13
Congress's job is to come up with what the legislative
6:15
framework looks like.
6:16
SPEAKER_01: I feel like we heard in lots of meetings we did last spring,
6:19
SPEAKER_01: we would kind of consistently hear,
6:20
SPEAKER_01: like,
6:21
SPEAKER_01: we want to see the White House weigh in here.
6:23
SPEAKER_01: So can you talk about that?
6:24
SPEAKER_01: Like,
6:24
why was it important for Congress to have the White House not
6:28
just say we want to
6:29
SPEAKER_01: see AI legislation or,
6:30
SPEAKER_01: you know,
6:31
SPEAKER_01: Congress needs to play an assertive role here vis-a-vis the states.
6:34
SPEAKER_01: Why was it important that they actually come up with a framework for what
6:37
SPEAKER_01: legislation might look like?
6:39
SPEAKER_00: Yeah, well, it's still the Congress's job.
6:40
SPEAKER_00: And the Congress has, I think, been very active on this.
6:43
SPEAKER_00: Back to our point in probably the September pod that we did.
6:46
SPEAKER_00: Obviously, there's been a lot of movement in the Congress.
6:48
There has been sort of the moratorium that went through the
6:50
reconciliation bill.
6:51
There's been another conversation about some level of
6:54
preemption,
6:54
which we should talk about.
6:56
SPEAKER_00: So it's been active.
6:57
SPEAKER_00: And I think there have been quite a few...
7:00
SPEAKER_00: across a number of different committees,
7:02
SPEAKER_00: hearings,
7:03
SPEAKER_00: and there are certainly legislative proposals out there.
7:07
But it's helpful on the front end to understand where the
7:09
president of the United
7:10
SPEAKER_00: States stands on some of these things.
7:12
So A,
7:12
you don't get all the way through the process and then get a
7:14
veto once it gets to him.
7:16
SPEAKER_00: But also...
7:18
The executive and the executive agencies have to be able to
7:21
implement the legislation.
7:22
SPEAKER_00: So legislation is like a framework.
7:24
And then the agencies take it and they put even more detail in
7:27
it and they implement it.
7:28
SPEAKER_00: And that becomes really the enforceable law, as you kind of just noted.
7:32
SPEAKER_00: It's just really, really important to understand where the president is.
7:35
A,
7:36
for the reasons I just laid out,
7:37
but B,
7:37
also sort of the politics of it all,
7:40
right?
7:40
SPEAKER_00: The president is the leader of one political party in this town.
7:46
The president needs to be able to sort of guide that political
7:48
party on a number,
7:49
SPEAKER_00: on a range of topics,
7:51
SPEAKER_00: especially something as important as AI.
7:53
SPEAKER_00: So it's a really important signal that to Republicans,
7:57
SPEAKER_00: hey,
7:57
SPEAKER_00: you're heading in the right direction here,
7:59
SPEAKER_00: you're not heading in the right direction here,
8:01
SPEAKER_00: and then go in and fill in the details
8:05
SPEAKER_00: it will then come to me and we'll sign it, right?
8:07
SPEAKER_00: That's the general idea.
8:08
SPEAKER_01: So let's dig in a little bit on the proposal.
8:11
SPEAKER_01: I'd like to sort of get at three parts of it.
8:13
SPEAKER_01: Like one is just what we think it says, what it says and how we read it.
8:16
And then I'd like to sort of see if we can think about how is
8:20
the minority receiving it?
8:21
SPEAKER_01: Like how would Democrats look at this?
8:24
SPEAKER_01: What would they like in this?
8:25
SPEAKER_01: And what would they see is insufficient?
8:27
SPEAKER_01: And then how has it been received?
8:28
SPEAKER_01: And we can talk about maybe how it's been received publicly,
8:30
SPEAKER_01: but I think it's more interesting for people to hear
8:33
how you think it's been received in the political process like
8:36
how has it been
8:36
received on the hill right so just like start maybe as a
8:39
starting point like what
8:40
are the what are the elements in this that we think are
8:43
meaningful well i should be
8:44
asking you really shouldn't i why don't you start yeah look i
8:47
mean look i i think i
8:48
think the the the white house and the president have been very
8:51
careful to sort of
8:53
craft this around what they call the four c's right which i
8:56
think is actually a
8:57
SPEAKER_00: really smart framing and the four c's
9:00
are geared towards addressing the things that have arisen as
9:04
problems from the
9:07
general public and a number of the stakeholders over the last
9:10
year and a half of
9:11
SPEAKER_00: real conversation in this space.
9:13
SPEAKER_00: Four Cs are children, communities, copyright, and censorship.
9:18
SPEAKER_00: And honestly, I think each of those issues
9:25
SPEAKER_00: has been around,
9:26
SPEAKER_00: I think well before AI,
9:28
but I think AI has the ability to sort of supercharge each of
9:31
them in a way that we
9:32
SPEAKER_00: haven't seen before.
9:33
So I think it would be actually helpful to hear you kind of
9:36
like talk through a
9:38
couple of them in terms of just from a policy set and then we
9:40
can kind of get back
9:41
SPEAKER_00: into the politics.
9:41
SPEAKER_01: Yeah, so let's start by what we're seeing here.
9:45
SPEAKER_01: We can talk about like the things that we're not seeing,
9:47
SPEAKER_01: but I think as a starting point,
9:50
SPEAKER_01: the cornerstone
9:52
SPEAKER_01: of our approach on policy,
9:54
SPEAKER_01: which I wish I could take credit for because I love it,
9:56
SPEAKER_01: but it really is you.
9:57
SPEAKER_01: You deserve a lot of credit.
9:58
It's really you and what Mark and Ben and Jay and you
10:03
established before I arrived
10:04
SPEAKER_01: at the firm,
10:05
SPEAKER_01: but is anchored around focusing on regulating use,
10:08
SPEAKER_01: not development.
10:09
And that means trying to ensure that when people are harmed in
10:13
different ways,
10:14
that
10:14
SPEAKER_01: when AI is used to harm people,
10:16
SPEAKER_01: that the people who are perpetrating the harm can be held to account.
10:21
SPEAKER_01: And there are elements in here that do that.
10:23
I think that principle is very expansive because you can use it
10:27
in a consumer
10:28
SPEAKER_01: protection context,
10:28
SPEAKER_01: a civil rights context,
10:29
SPEAKER_01: an antitrust context.
10:31
SPEAKER_01: Criminal law context.
10:32
SPEAKER_01: So there's a million things that could be in here related to that idea.
10:36
SPEAKER_01: But I think we see in here some elements of it.
10:38
SPEAKER_01: So there's a focus on the use of AI for fraud, for instance.
10:42
And I think when we see that,
10:43
I get excited because so much of the policy proposals.
10:46
SPEAKER_01: It's the right framing.
10:47
SPEAKER_01: It's the right framing.
10:48
SPEAKER_01: So many of the policy proposals we see are paperwork proposals.
10:51
SPEAKER_01: Correct.
10:52
SPEAKER_01: We need you to conduct an impact assessment.
10:54
SPEAKER_01: We need you to do an audit.
10:55
SPEAKER_01: And those things are the things that we think really burden startups.
10:59
SPEAKER_01: And burden startups without, I think, a lot of benefits on the other.
11:03
SPEAKER_01: Right.
11:03
SPEAKER_01: It probably doesn't keep people safe, right?
11:06
SPEAKER_01: Is an impact assessment likely to prevent fraud?
11:09
SPEAKER_01: Maybe in some circumstances, but I think the chances are pretty small.
11:12
SPEAKER_01: And when I think of
11:14
SPEAKER_01: when I'm moving through the world and I am a potential victim,
11:18
what is the thing that can give me some comfort that either I
11:23
won't be harmed,
11:24
that companies will be engaging in preventative activity on
11:27
their own to ensure
11:28
SPEAKER_01: that they're not incurring criminal liability,
11:30
SPEAKER_01: or in the event that I am harmed,
11:32
SPEAKER_01: that there's recourse.
11:33
SPEAKER_01: And I think we primarily think of that in how we think about AI policy,
11:38
SPEAKER_01: but also just how the law thinks about it.
11:39
SPEAKER_01: We think about that in terms of
11:41
SPEAKER_01: penalizing harmful use.
11:43
So seeing those elements in here,
11:44
I think was really encouraging and exciting.
11:47
SPEAKER_01: And there aren't a lot of development-oriented proposals.
11:49
SPEAKER_01: So I think that's really critical.
11:51
I can just like tick through a couple of the other kind of top
11:54
line things that we see.
11:55
We've spent a lot of time thinking about what the right
11:57
approach is on kids or what
11:58
SPEAKER_01: a right approach is on kids.
11:59
SPEAKER_01: It's such a hard issue.
12:01
SPEAKER_01: Right.
12:01
SPEAKER_01: And it's been around for a very long time.
12:02
SPEAKER_01: It's been around for a long time.
12:03
SPEAKER_01: It's not, as you said, not specific to AI.
12:08
SPEAKER_01: And there are really challenging...
12:11
SPEAKER_01: They're really challenging dynamics there.
12:12
SPEAKER_01: It's important that kids are that we try to minimize harm to kids.
12:18
It's important that we give parents an active role in
12:21
protecting kids and making
12:22
SPEAKER_01: decisions on behalf of their own.
12:24
SPEAKER_01: minors.
12:26
SPEAKER_01: It's also important that kids learn how to use AI tools.
12:28
SPEAKER_01: And so the idea that if you're under 18,
12:31
SPEAKER_01: you're not going to have access to AI tools,
12:33
SPEAKER_01: that's not going to help kids prepare for an AI economy.
12:36
SPEAKER_01: It's going to cut them off from important information.
12:39
And there are Republican concerns about that information access
12:42
and there are
12:42
SPEAKER_01: Democrat concerns about that information access.
12:45
SPEAKER_01: Kids have privacy rights too.
12:46
SPEAKER_01: And so striking this right balance, I think is really challenging.
12:50
SPEAKER_01: And there are, I don't think there are
12:52
SPEAKER_01: ultimate solutions necessarily in the framework.
12:54
But there are ideas that move in that direction that I think
12:56
are constructive starting point.
12:57
SPEAKER_00: It's very interesting on this.
12:59
SPEAKER_00: Actually, maybe we should just talk about this a little bit more.
13:01
Some of the proposals that are out there that I think are
13:04
really challenging,
13:05
SPEAKER_00: to your point,
13:06
SPEAKER_00: are banning children's use under the age of 18.
13:12
SPEAKER_00: It's tough, right?
13:13
It's a tough conversation because I think we're now seeing
13:15
things like Alpha
13:16
SPEAKER_00: School,
13:17
SPEAKER_00: for instance,
13:18
who have kind of reinvented the way that children are learning
13:22
in the classroom.
13:23
And I think what we found is it's really important for kids to
13:26
have playtime.
13:27
SPEAKER_00: It's really important for them to build social skills.
13:30
SPEAKER_00: And it's really important actually for them to learn at their own pace.
13:33
When you have a teacher and 30 kids in a classroom and
13:36
everyone's learning arithmetic,
13:38
at the exact same pace it's very likely that some of these kids
13:42
are not going to be
13:42
able to keep up yeah um and if you now can shift that model to
13:49
something like alpha
13:49
school where you have kind of a personalized ai tutor in
13:53
conjunction with a teacher
13:54
in a classroom some kids can learn and the test scores are
13:59
basically uh supporting
14:01
this they can learn for two and a half or three hours a day and
14:04
the rest of the
14:04
SPEAKER_00: time they can be building social skills or playing or
14:07
SPEAKER_00: Being a kid,
14:08
SPEAKER_00: while other kids need to spend an hour in the morning,
14:10
SPEAKER_00: an hour in the afternoon,
14:11
SPEAKER_00: an hour in the evening learning because that's how they learn better.
14:15
SPEAKER_00: If you were to ban that,
14:16
it seems like it would stunt the growth for all the reasons
14:19
you've outlined,
14:20
SPEAKER_00: but also certainly from an educational perspective.
14:23
SPEAKER_00: How do you think about that?
14:23
SPEAKER_01: Well,
14:24
I think bans are something that you do when the overwhelming
14:28
majority of use cases
14:29
SPEAKER_01: are negative.
14:30
And I think there are a range of different views about the
14:33
value proposition behind AI.
14:35
SPEAKER_01: We obviously are incredible optimists.
14:37
We really think people's lives will be fundamentally altered in
14:40
ways that are net
14:42
SPEAKER_01: positive,
14:42
SPEAKER_01: net incredibly positive by AI.
14:46
If you have our view,
14:47
then clearly a ban means you're taking enormous benefits off
14:50
the table
14:51
SPEAKER_01: to try to address concerns and harms that we acknowledge exist.
14:55
But the consequence in terms of the benefits you're removing
14:58
are not justified by
15:00
SPEAKER_01: the harms that you're preventing.
15:02
SPEAKER_01: And I think even a less optimistic view than ours,
15:05
I think most people can recognize that AI is going to have
15:09
really positive
15:10
SPEAKER_01: transformative impacts.
15:11
And that even if you're skeptical of some of those,
15:15
we're moving to a world where
15:18
SPEAKER_01: knowing how to use AI,
15:19
being AI literate is gonna be an important part of how people
15:24
do their jobs,
15:25
SPEAKER_01: run their lives,
15:26
SPEAKER_01: interact with other people.
15:27
And so the idea that you would not start to develop those tools
15:32
until after you
15:33
SPEAKER_01: turn 18,
15:34
SPEAKER_01: I think is really problematic.
15:35
SPEAKER_01: It was encouraging over the summer,
15:37
SPEAKER_01: I think summer,
15:38
SPEAKER_01: early fall,
15:38
SPEAKER_01: probably around the time we talked last,
15:40
Governor Newsom vetoed a bill that arguably could be described
15:44
as kid safety in
15:45
SPEAKER_01: California.
15:47
And because he thought it was going to cut kids off from
15:49
important ability to
15:51
SPEAKER_01: access the tools in ways that are meaningful and positive.
15:55
And he also simultaneously signed another bill that he thought
15:57
was going to do was
15:58
a narrower way to address potential harms without getting in
16:01
the way of some of
16:02
SPEAKER_01: those benefits.
16:03
SPEAKER_01: So I think the idea of bans is really important.
16:06
SPEAKER_01: is really probably not the right way to think about cost and benefits.
16:08
SPEAKER_01: It's really challenging.
16:09
But we have worked really hard to make sure that the way that
16:13
we think about policy
16:14
SPEAKER_01: is responsive to the concerns that the courts have.
16:16
SPEAKER_01: And it's something we hear all the time on the Hill.
16:18
SPEAKER_01: There need to be real rules and restrictions.
16:20
SPEAKER_01: We've got to keep kids safe.
16:21
SPEAKER_00: There's no question about it.
16:22
The only other thing I would say on this,
16:24
and if you have additional,
16:26
we can do that.
16:28
SPEAKER_00: you know,
16:28
SPEAKER_00: the new political vector of attack for some of these really,
16:31
SPEAKER_00: really bad ideas is through kids' bills.
16:33
SPEAKER_00: They're using kids' bills almost as a shell so they can dump in a lot of
16:38
SPEAKER_00: ban-oriented or things that look very innocuous but are actually very,
16:43
very difficult and challenging not only to implement but I
16:46
think also would be bad
16:47
SPEAKER_00: and negative through the lens of kids' bills.
16:51
SPEAKER_00: Yes.
16:51
SPEAKER_01: Yeah, it's frustrating.
16:52
SPEAKER_01: I mean,
16:52
SPEAKER_01: it's frustrating to see that in part because,
16:54
SPEAKER_01: you know,
16:54
SPEAKER_01: again,
16:55
SPEAKER_01: you understand the political process.
16:56
I sort of focus on like what's the policy that we feel like we
16:59
could we could get behind.
17:01
SPEAKER_01: Like we want to we're looking for stuff we can get behind.
17:03
And there have even some of the most there are sometimes bills
17:05
that we see that are
17:06
SPEAKER_01: really like effectively equivalent to bans,
17:08
SPEAKER_01: really problematic.
17:09
And then they'll also include provisions like it should be a
17:11
violation of criminal
17:12
SPEAKER_01: law to solicit a minor for unlawful sex,
17:15
SPEAKER_01: which is.
17:17
SPEAKER_01: I think already prohibited by criminal law,
17:18
SPEAKER_01: but to the extent there's a loophole,
17:20
SPEAKER_01: to the extent it's not sufficiently covered,
17:22
SPEAKER_01: is definitely something that should be in place.
17:24
SPEAKER_01: So to your point, those are the kinds of things that we can get behind.
17:28
SPEAKER_01: It's harm-based.
17:29
SPEAKER_01: It's targeted at protecting kids.
17:31
It's making sure that when you're not using AI,
17:35
you'd be held accountable criminally.
17:37
SPEAKER_01: Then when you are using AI, you should also be held liable criminally.
17:40
SPEAKER_01: We support those concepts to include alongside them
17:44
a broad ban of kids access is problematic and makes those kinds
17:48
of legislative
17:48
SPEAKER_01: proposals something we couldn't get behind.
17:50
SPEAKER_00: Yep.
17:51
SPEAKER_00: Agreed.
17:52
SPEAKER_01: All right.
17:52
SPEAKER_01: What other parts of the framework should we touch on?
17:56
SPEAKER_00: Let's talk about the four Cs, I think.
17:57
SPEAKER_00: Okay.
17:57
SPEAKER_00: All right.
17:58
SPEAKER_01: So in the community section,
17:59
SPEAKER_01: there's one,
18:00
like one of the things that we've done some work on energy and
18:05
the president
18:05
SPEAKER_01: developed the rate payer protection pledge,
18:08
SPEAKER_01: which most of the hyperscalers have signed on to.
18:10
SPEAKER_01: And the basic idea is like,
18:12
SPEAKER_01: they're bringing their own power and water to data center projects.
18:15
SPEAKER_01: Like they're not asking other people to pay for their costs.
18:18
SPEAKER_01: They're bringing their own power and water.
18:20
We have thought that's important for little tech because if
18:23
you're a startup and
18:26
there are sort of sweetheart deals that can go to large
18:28
companies and you're left
18:29
out of those deals and you're left holding the bag from a cost
18:32
perspective or the
18:34
SPEAKER_01: hyperscalers are getting kind of
18:36
preferential access to energy to compute,
18:39
that's going to be problematic for little tech.
18:41
And so I think the president's steps there are really positive
18:44
from our standpoint.
18:45
SPEAKER_00: Incredibly positive.
18:46
SPEAKER_00: And look, I think that that should be an obvious thing.
18:49
You can't build a data center that is also an explode energy
18:51
prices and explode
18:52
SPEAKER_00: water prices for a community.
18:54
SPEAKER_00: That's not realistic.
18:54
SPEAKER_00: And I think I think the companies rightfully acknowledge that.
18:57
SPEAKER_00: So I think that that's less controversial.
18:59
SPEAKER_00: I think that that is exactly the way to solve this.
19:02
SPEAKER_00: You know, the the
19:03
SPEAKER_00: The other thing here is,
19:05
SPEAKER_00: just from a messaging and a political point of view,
19:08
SPEAKER_00: it really does seem like
19:10
SPEAKER_00: data centers,
19:11
SPEAKER_00: and I think the companies behind the data centers,
19:13
SPEAKER_00: and I look,
19:13
SPEAKER_00: I think this would apply to us,
19:15
have done a quite a poor job of actually messaging about the
19:18
benefits of these
19:18
SPEAKER_00: things,
19:19
SPEAKER_00: right?
19:20
SPEAKER_00: Certainly there are some downsides and,
19:21
SPEAKER_00: you know,
19:21
we've all seen the videos of like low humming noises and things
19:24
like that.
19:25
SPEAKER_00: Those need to be fixed and things need to be solved there.
19:28
SPEAKER_00: So I'm not suggesting that,
19:29
SPEAKER_00: but we're talking about a real tax base increase,
19:33
SPEAKER_00: unlike these communities I've ever seen.
19:35
SPEAKER_00: So now all of a sudden you have the best schools,
19:36
SPEAKER_00: you have the best sidewalks,
19:37
SPEAKER_00: you have the best public transportation.
19:40
SPEAKER_00: that's really hugely beneficial.
19:42
SPEAKER_00: I don't think it's being communicated effectively.
19:45
SPEAKER_00: You know,
19:46
that's something that I think we as an industry collectively
19:48
need to take on as a
19:49
SPEAKER_00: challenge.
19:51
SPEAKER_01: Okay,
19:51
SPEAKER_01: so some other elements in the proposal,
19:53
SPEAKER_01: the workforce proposals,
19:56
SPEAKER_01: that's been something that I've,
19:58
long been excited by generally and wish that there'd be more
20:03
legislative momentum
20:04
in that area because i think republicans and democrats talk
20:07
about that well they
20:07
they have like different constituencies in mind when they think
20:10
about someone who
20:10
might be displaced by a worker who might be displaced by ai and
20:14
i should say kind
20:15
of at the outset like our view is that ai is going to have a
20:18
significant positive
20:19
effect on the labor market not a negative effect yes that's
20:21
consistent with the
20:22
SPEAKER_01: history of
20:23
SPEAKER_01: the introduction of technologies.
20:24
SPEAKER_01: We're always wary of disruptions and there are,
20:27
SPEAKER_01: there are clearly disruptions,
20:28
SPEAKER_01: but,
20:29
SPEAKER_01: but overall,
20:30
SPEAKER_01: typically there are gains for labor,
20:32
SPEAKER_01: not,
20:33
SPEAKER_01: not costs.
20:34
SPEAKER_01: Right.
20:35
SPEAKER_01: But even if you have that positive view,
20:36
SPEAKER_01: like there,
20:37
there are workers who will be disrupted and those workers
20:39
should be supported in
20:40
SPEAKER_01: their transitions.
20:41
SPEAKER_01: They should be ideally like reskilled into jobs that are,
20:45
SPEAKER_01: jobs that will employ them in an AI economy,
20:48
SPEAKER_01: to the extent they're unable to find jobs,
20:51
SPEAKER_01: they should be supported during a period of unemployment.
20:53
SPEAKER_01: And my sense is that Republicans believe that, Democrats believe that.
20:58
SPEAKER_01: There's a lot of consensus behind it.
21:00
There are a whole bunch of different concepts from unemployment
21:05
insurance to
21:07
SPEAKER_01: reskilling programs to just things that don't require any expenditures.
21:11
SPEAKER_01: They're just
21:12
how we track disruptions in labor markets so that we can try to
21:15
see them early on,
21:17
SPEAKER_01: see those disruptions.
21:19
SPEAKER_01: So the way that the framework is structured in that section,
21:23
SPEAKER_01: I think is at least a good starting point.
21:24
SPEAKER_01: I mean,
21:24
SPEAKER_01: I don't think it's probably the end of the road,
21:26
SPEAKER_01: but a starting point for thinking about how we support workers.
21:29
SPEAKER_00: Yeah, I agree.
21:29
SPEAKER_00: I mean, look, I think the big challenge with this is
21:33
SPEAKER_00: it's been very difficult with everyone that I've talked to,
21:36
to pinpoint exactly where we're going to see massive disruption
21:40
because the
21:40
SPEAKER_00: disruption hasn't really happened yet.
21:42
SPEAKER_00: Yeah.
21:42
SPEAKER_00: There have been,
21:42
SPEAKER_00: you know,
21:43
SPEAKER_00: a couple,
21:43
SPEAKER_00: you know,
21:44
SPEAKER_00: the infamous sort of,
21:46
SPEAKER_00: you know,
21:46
SPEAKER_00: block announcement where they slash 50% of the workforce.
21:49
And,
21:49
you know,
21:50
I think that there's a number of factors that went into that.
21:52
Obviously the justification was AI,
21:53
but I think there's more to that story.
21:56
SPEAKER_00: But
21:57
SPEAKER_00: You know,
21:58
the World Economic Forum came out with an article earlier this
22:01
year that basically
22:03
suggested there's 1.3 million new jobs based on sort of AI over
22:08
the last two years.
22:09
SPEAKER_00: Yeah.
22:09
SPEAKER_00: Which is a pretty sizable explosion.
22:13
SPEAKER_00: Not to mention,
22:14
SPEAKER_00: later on in that article,
22:16
SPEAKER_00: it's basically like a three to one job gain versus loss in the AI market.
22:23
SPEAKER_00: And I...
22:24
i think it's just it's really tough to quantify i mean we've
22:27
talked about you know
22:27
mark talks about the lump of labor fallacy all the time the
22:29
idea that the pie is
22:30
SPEAKER_00: fixed and
22:31
You know,
22:32
if you remove a slice of the pie,
22:34
then there's sort of a collapse of market.
22:35
But that's not necessarily the case because the pie,
22:38
we believe,
22:38
is going to grow.
22:39
SPEAKER_00: So my comments are to suggest that it might not be a real issue.
22:44
SPEAKER_00: It's just really hard to pinpoint how we should attack this.
22:47
SPEAKER_00: That being said,
22:48
there are a lot of really smart people that I think inside and
22:51
outside government
22:52
SPEAKER_00: that care about this and are working on this.
22:54
The one area that I think we should be focusing on is
22:56
retraining because there are
22:58
SPEAKER_00: certainly going to be,
23:00
SPEAKER_00: you know,
23:01
SPEAKER_00: call center seems to be one that comes up frequently.
23:04
It seems to me like call centers are going to be an area in
23:08
which AI could easily
23:10
SPEAKER_00: disrupt and probably disrupt in the next 12 months or so.
23:13
So now we're going to have a certain number of people that are
23:17
going to be displaced from a job.
23:18
SPEAKER_00: How can we
23:19
organize them into something new and hopefully in an area where
23:22
I think AI is
23:23
SPEAKER_00: exploding and creating new jobs.
23:25
SPEAKER_00: Yeah.
23:25
I think regardless of the statistics,
23:27
now we're going to switch roles a little bit.
23:28
SPEAKER_01: Yeah.
23:29
Like I think the thing that you constantly urge me to do in my
23:32
work is like,
23:33
SPEAKER_01: it's not just what's on the page.
23:34
SPEAKER_01: It's not just what the data says.
23:36
What is it like when you're sitting across from someone who
23:38
says that they lost
23:39
SPEAKER_01: their job because of AI?
23:40
SPEAKER_01: Or what is it like when you're hearing from a lawmaker who's saying,
23:43
SPEAKER_01: I hear what you're saying in terms of statistics,
23:45
SPEAKER_01: but I'm going to town halls.
23:47
SPEAKER_01: And in the town halls, this is what I'm hearing.
23:48
SPEAKER_01: People are anxious.
23:50
SPEAKER_01: I think there's the reality of the disruption we have to deal with.
23:53
But then there's the perception of the disruption,
23:54
which I think is also a big thing.
23:56
SPEAKER_01: It may be that you don't lose your job because of AI,
23:58
SPEAKER_01: but you're nervous because of losing your job because of AI.
24:01
SPEAKER_01: Yeah.
24:03
SPEAKER_01: I think if the policy solutions were like extremely high cost,
24:07
SPEAKER_01: very disruptive,
24:08
SPEAKER_01: challenging for little tech to implement,
24:11
SPEAKER_01: challenging for government to implement,
24:12
SPEAKER_01: then maybe we would want to have some caution and concerns about them.
24:16
SPEAKER_01: But I think there's a lot of low hanging fruit.
24:17
SPEAKER_01: And so it's great to see the White House doing this.
24:21
SPEAKER_01: I thought in the action plan,
24:23
the previous guidance from the White House that was more
24:26
focused on executive
24:26
SPEAKER_01: branch activity,
24:27
SPEAKER_01: there was a ton of stuff in there that was really positive.
24:30
SPEAKER_01: I think if you had just like
24:32
cut it out of the action plan and you put it on just a blank
24:34
document and you
24:35
SPEAKER_01: showed it to Democrats,
24:36
SPEAKER_01: they would have thought,
24:36
SPEAKER_01: oh,
24:37
SPEAKER_01: these are a bunch of Democrat ideas.
24:38
SPEAKER_01: Like they were,
24:39
SPEAKER_01: they were really like,
24:40
I think sort of like little P progressive concepts for how we
24:44
support workers who
24:45
SPEAKER_01: are disrupted.
24:45
SPEAKER_01: And so there's just,
24:47
SPEAKER_01: I feel like the,
24:48
SPEAKER_01: this is not a sexy topic,
24:50
SPEAKER_01: but it's a topic where there's a lot of bipartisan consensus.
24:52
SPEAKER_01: And also I think importantly, a lot of room to really help people.
24:55
SPEAKER_00: Yeah.
24:55
SPEAKER_00: You know, the only thing I would note on that is just
24:58
We're going to have a conversation,
25:00
I think,
25:01
federal versus states at some point here.
25:03
SPEAKER_00: I think it's important to discuss that.
25:06
SPEAKER_00: But this is also an area where I think states are stepping up,
25:09
SPEAKER_00: and they're also starting to have real conversations.
25:11
So the collaboration between federal governments and state
25:13
governments to try and
25:14
SPEAKER_00: address this problem with the private sector,
25:16
SPEAKER_00: I think,
25:16
SPEAKER_00: is ultimately going to be
25:18
SPEAKER_00: the all hands on deck approach that we're going to have to take.
25:21
SPEAKER_01: Yeah.
25:21
SPEAKER_01: Okay.
25:22
Um,
25:22
so I want to like,
25:23
I'm going to take through a couple of things quickly.
25:25
SPEAKER_01: Cause then I want to like get to a different perspective on this.
25:28
SPEAKER_01: I want to get to the Democrat perspective on this,
25:29
SPEAKER_01: but like importantly,
25:31
SPEAKER_01: there's stuff in here on intellectual property rights and like making
25:35
signaling clearly the White House's view that we view as the
25:38
correct legal
25:39
determination that most of AI training is fair use and
25:43
protected by copyright law.
25:45
SPEAKER_01: And you have written about that a lot.
25:47
SPEAKER_00: Obviously, Jay has as well.
25:48
SPEAKER_01: Yep.
25:49
And then some stuff on censorship that I think actually has
25:52
more bipartisan appeal
25:53
SPEAKER_01: than you'd think on its face.
25:54
SPEAKER_01: Mm-hmm.
25:56
They're often like Republicans who are pushing the censorship
25:59
related concepts out
26:00
SPEAKER_01: of their concern of of a a bias against conservative ideas.
26:07
SPEAKER_01: I think there are principles in here that also Democrats might be might.
26:10
SPEAKER_01: There are ideas here related to job voting,
26:13
SPEAKER_01: for instance,
26:13
SPEAKER_01: and that's something that is of concern to Democrats as well.
26:15
SPEAKER_01: So I actually think this is kind of,
26:17
SPEAKER_01: you know,
26:17
SPEAKER_01: grounded in a First Amendment approach to speech issues,
26:20
SPEAKER_01: which I think is really positive.
26:23
SPEAKER_01: Then there is,
26:24
SPEAKER_01: well,
26:25
maybe we should like put a placeholder on the preemption stuff
26:27
and talk about that
26:27
SPEAKER_01: in more detail in a few minutes.
26:29
SPEAKER_01: So we've been through what's in the document,
26:32
SPEAKER_01: and I think sort of talked about the White House's perspective.
26:35
SPEAKER_01: Right.
26:35
SPEAKER_01: If you're a Democrat on the Hill, and you are...
26:40
SPEAKER_01: reading this, you're receiving this.
26:43
SPEAKER_01: Are there things in here that you like?
26:45
Are there principles in here that you think are things that are
26:50
offered to
26:50
SPEAKER_01: Democrats as a way to hopefully bring them to the negotiating table?
26:54
And what are you not seeing here that would be important at the
26:57
negotiating table
26:58
to see that gets you to feel more comfortable with the concept
27:03
of some sort of
27:04
SPEAKER_01: congressional action in AI?
27:06
SPEAKER_00: Yeah.
27:07
SPEAKER_00: One is...
27:11
SPEAKER_00: We agree with the overall framing with some of these categories,
27:15
SPEAKER_00: but there's not enough substance here for us to evaluate,
27:18
SPEAKER_00: which I think is...
27:20
You know,
27:21
a somewhat fair criticism,
27:22
but that's also Congress's job,
27:23
like we talked about.
27:24
Congress needs to fill in some of the gaps that have been in
27:26
this general framework.
27:27
SPEAKER_01: Yeah.
27:28
SPEAKER_01: Go ahead.
27:29
SPEAKER_01: Is it a fair criticism?
27:30
SPEAKER_01: I've heard that.
27:31
SPEAKER_01: It's hard for me to...
27:32
I mean,
27:33
it's not a legislative proposal,
27:35
but it wasn't intended to be a model bill.
27:38
SPEAKER_00: It wasn't.
27:39
SPEAKER_00: It wasn't at all.
27:39
SPEAKER_00: I mean, I think that, you know, I think it's...
27:43
some of the gamesmanship that you would see from any sort of
27:46
cross-party negotiation.
27:48
SPEAKER_00: But look,
27:48
SPEAKER_00: I think that there is enough meat on the bone to get started,
27:51
SPEAKER_00: and that was exactly what it was intended to be.
27:54
SPEAKER_00: But if you're asking them, okay, is this something that you agree with?
27:58
SPEAKER_00: The answer is reflexively going to be, I need more detail.
28:00
SPEAKER_00: And I think that that's effectively what we've seen.
28:04
SPEAKER_00: The second category here is
28:08
SPEAKER_00: they want to see something like an SB-53 safety framework in here,
28:12
SPEAKER_00: quote unquote safety framework,
28:14
SPEAKER_00: for all of the reasons that we talked about at the top.
28:17
SPEAKER_00: These disclosure frameworks that seem fairly innocuous,
28:22
SPEAKER_00: these safety reports that must be put out,
28:25
SPEAKER_00: these transparency standards,
28:27
SPEAKER_00: et cetera,
28:28
SPEAKER_00: they all sound really good.
28:29
SPEAKER_00: But in practice, are they actually keeping people safe?
28:32
SPEAKER_00: They, in effect,
28:34
create sort of a regulatory barrier because the people with the
28:36
most lawyers are
28:37
SPEAKER_00: able to comply,
28:38
which is why we've been critical of this from a little tech
28:40
perspective for a long
28:40
SPEAKER_00: time.
28:41
SPEAKER_00: And you can talk about this eloquently.
28:43
SPEAKER_00: And look,
28:43
SPEAKER_00: the third category,
28:44
which I think is probably the most important political
28:46
perspective is that anything
28:47
SPEAKER_00: Donald Trump does,
28:48
SPEAKER_00: they're going to dislike.
28:49
SPEAKER_01: Yeah.
28:50
SPEAKER_01: You know?
28:50
SPEAKER_01: Yeah.
28:51
SPEAKER_01: So what, so,
28:53
So I actually hear that as fairly bleak in that it wasn't going
28:59
to be a longer
28:59
SPEAKER_01: document by design.
29:00
SPEAKER_01: And I think in a helpful way, it's short.
29:02
SPEAKER_01: Because as you're saying, this isn't the president's job.
29:05
SPEAKER_01: This is Congress's job.
29:06
SPEAKER_01: So you don't want the president to be dictating what the actual text is.
29:09
SPEAKER_01: You want it to be conceptual.
29:10
The second category are things we would strongly oppose because
29:14
we think it's going
29:15
SPEAKER_01: to make it really hard for little tech to compete.
29:17
SPEAKER_01: If we get all the things in your second category, then
29:20
then we will have very concentrated AI markets where the
29:26
startups will be at a
29:27
SPEAKER_01: significant disadvantage.
29:28
SPEAKER_01: And then the third is...
29:30
SPEAKER_01: I don't know how you impossible to deal with,
29:32
SPEAKER_01: um,
29:32
SPEAKER_01: in terms of like really trying to get a framework to the table.
29:35
SPEAKER_01: And so I guess part of what I'm wondering is like,
29:37
SPEAKER_01: what are the things like if we can envision a final package,
29:40
SPEAKER_01: what are the things that Democrats like,
29:42
what are the kinds of things that you could add to this that
29:44
would get Democrats
29:45
SPEAKER_01: to?
29:45
SPEAKER_01: Yes.
29:46
SPEAKER_01: I think there's probably like a lot of stuff in the regulate,
29:49
SPEAKER_01: um,
29:49
SPEAKER_01: harmful use category that appeal to Democrats.
29:52
SPEAKER_01: Yeah, I agree.
29:52
SPEAKER_01: Like,
29:53
SPEAKER_01: um,
29:54
SPEAKER_01: enforcement of civil rights law like that,
29:56
SPEAKER_01: that actually think that's a bipartisan concept because,
29:59
SPEAKER_01: um,
30:00
SPEAKER_01: Democrats think about equity issues,
30:01
SPEAKER_01: but enforcement of civil rights law includes First Amendment violations,
30:04
SPEAKER_01: which Republicans care about.
30:06
So so making clear that when AI is used to violate a person's
30:09
civil rights,
30:10
SPEAKER_01: like that should be a thing I think that's appealing to Democrats.
30:13
Democrats have concerns about enforcement agencies not having
30:16
the resources they
30:17
SPEAKER_01: need,
30:17
SPEAKER_01: whether that's the Justice Department.
30:20
SPEAKER_01: If you're talking about civil rights specifically,
30:21
SPEAKER_01: it's the Civil Rights Division,
30:23
SPEAKER_01: the Federal Trade Commission.
30:25
The Federal Trade Commission is going to be enforcing consumer
30:27
protection law.
30:28
So I wonder if that's something that maybe gets Democrats a
30:31
little bit more excited
30:33
SPEAKER_01: about some of the ideas here.
30:35
else that you think could be on the table for Democrats that is
30:38
compelling?
30:38
SPEAKER_00: Yeah, I think all of those things are germane.
30:40
SPEAKER_00: I think the other piece, my mind goes to Kids Online Safety Act.
30:44
SPEAKER_00: Been a major bipartisan bill for a very long time.
30:46
SPEAKER_00: Obviously,
30:47
SPEAKER_00: originally geared towards more social media,
30:49
SPEAKER_00: but is,
30:49
SPEAKER_00: I think,
30:49
SPEAKER_00: also applicable in the age of AI.
30:51
SPEAKER_00: I think there are some problems with it,
30:53
but I think both Republicans and Democrats have acknowledged
30:57
the fact that this
30:58
SPEAKER_00: needs to move forward in some way,
31:00
SPEAKER_00: shape,
31:00
SPEAKER_00: or form.
31:01
SPEAKER_00: So if there is a strong form of COSA,
31:04
SPEAKER_01: Yeah.
31:04
SPEAKER_00: That moves forward and it gets attached to this bill.
31:07
I think that's also something that would get them excited about
31:09
voting for this.
31:10
SPEAKER_01: Yeah.
31:10
SPEAKER_01: And COSA is bipartisan.
31:12
SPEAKER_01: Senators Blackburn and Blumenthal.
31:13
SPEAKER_01: Correct.
31:14
SPEAKER_01: It's been on the table for a long period of time.
31:15
SPEAKER_01: Very long time.
31:16
SPEAKER_01: It's had committee support for a long period of time.
31:18
SPEAKER_01: Hasn't gotten enacted.
31:20
SPEAKER_01: And so that would be a significant win.
31:21
SPEAKER_00: Yeah.
31:22
SPEAKER_00: And look,
31:22
SPEAKER_00: I mean,
31:23
SPEAKER_00: you know,
31:23
SPEAKER_00: the AI Safety Institute,
31:24
SPEAKER_00: which has now been sort of has been changed into Casey,
31:29
SPEAKER_00: which is,
31:30
SPEAKER_00: you know,
31:30
SPEAKER_00: I think effectively sort of a very...
31:33
SPEAKER_00: And similar mission,
31:34
SPEAKER_00: but also now with sort of innovation backbone to it,
31:38
SPEAKER_00: codifying that and making sure that it's permanent,
31:40
I think is something that they've been advocating for for quite
31:43
a bit of time.
31:43
And I think that that is something that is relatively
31:47
non-controversial at this point.
31:49
SPEAKER_00: And I think something that could get done.
31:50
SPEAKER_01: And it seems like there are a lot,
31:52
SPEAKER_01: lots,
31:52
SPEAKER_01: everyone seems to like Casey,
31:54
SPEAKER_01: which I think is interesting.
31:55
SPEAKER_01: Because not everyone liked the Safety Institute.
31:57
SPEAKER_01: Everyone seems to like Casey.
32:00
SPEAKER_01: But there are lots of like,
32:01
SPEAKER_01: everyone likes it,
32:02
SPEAKER_01: but it seems like there's not a lot of content now on what it is.
32:05
And so I've heard some Democrats talk about ideas of Casey
32:09
being used for like
32:10
SPEAKER_01: cybersecurity capabilities,
32:12
SPEAKER_01: testing and benchmarking,
32:13
SPEAKER_01: which I think is an interesting idea.
32:15
We've done some thinking internally on how could you make
32:19
benchmarking resources
32:21
available to startups and then have some kind of regulatory
32:24
mechanism where
32:25
SPEAKER_01: startups are
32:27
startups and larger companies are testing their AI models
32:31
against the benchmarks
32:31
SPEAKER_01: that Casey offers,
32:33
SPEAKER_01: for instance.
32:33
SPEAKER_01: Or I think ideally, it wouldn't be one benchmark.
32:36
SPEAKER_01: We'd have concerns,
32:37
SPEAKER_01: I think,
32:37
about regulatory capture if Casey or NIST is publishing just
32:40
one benchmark that
32:41
SPEAKER_01: startups have to test against.
32:44
SPEAKER_01: But if there's a plethora of benchmarks and startups are able in a cost,
32:50
reasonable cost kind of way to test their models against a
32:55
cybersecurity benchmark
32:56
SPEAKER_01: ecosystem,
32:57
and then consumers can make determinations about what models
33:00
they think are safe
33:01
SPEAKER_01: and make decisions about what they use.
33:02
SPEAKER_01: And the government could use it for procurement.
33:05
SPEAKER_01: Casey Nist could write reports about the benchmarking industry.
33:08
They could write reports about what benchmarks are a little
33:09
better,
33:10
a little bit worse.
33:11
SPEAKER_01: I think there's something there that might be
33:13
SPEAKER_01: might be workable.
33:14
SPEAKER_00: Right.
33:14
SPEAKER_00: Now,
33:14
SPEAKER_00: and I think we should also talk about the things,
33:17
the ideas that probably would not end up in a bill that is
33:20
likely to pass but are
33:21
SPEAKER_00: still,
33:22
SPEAKER_00: I think,
33:23
SPEAKER_00: generally democratic ideas that they would want in there.
33:25
SPEAKER_00: Sure.
33:25
SPEAKER_00: Licensing regimes are one thing.
33:27
SPEAKER_00: Creating an entirely new agency that
33:31
SPEAKER_00: effectively centralizes all AI policy.
33:34
SPEAKER_00: And before a model is released,
33:36
the public must go to this agency and the agency must bless it
33:41
and then it can be
33:41
SPEAKER_00: released.
33:42
SPEAKER_00: Yes.
33:42
And we should say- We should talk about,
33:43
let's talk about the downsides of that.
33:45
SPEAKER_00: Well,
33:45
SPEAKER_01: let's talk about the downsides of it,
33:46
SPEAKER_01: but let's also talk,
33:47
SPEAKER_01: I think when people hear us talk about this,
33:49
SPEAKER_01: they think,
33:50
SPEAKER_01: I've never heard any Democrats propose a licensing regime.
33:53
SPEAKER_01: And that was proposed on the Hill-
33:56
SPEAKER_01: A couple of years ago, you had companies coming out in favor of it.
33:59
SPEAKER_01: You had administration in support of it.
34:02
SPEAKER_01: But that aside,
34:03
I think it's really important to talk in specific ways about
34:08
what we hear as
34:09
SPEAKER_01: licensing,
34:10
SPEAKER_01: even if other people in the ecosystem hear it really differently.
34:12
There are a set of policy proposals that when we hear people
34:15
talk about them with
34:16
SPEAKER_01: us,
34:16
SPEAKER_01: we hear licensing.
34:18
SPEAKER_01: One of them is a permutation of this benchmark concept.
34:20
SPEAKER_01: So I think if Casey or NIST puts out a benchmark,
34:24
And if you build at the frontier,
34:26
you are required to test against this benchmark.
34:28
And if you don't hit a certain score,
34:30
you cannot release your model into the world.
34:32
SPEAKER_01: I think we hear that as licensing.
34:34
SPEAKER_01: Yes, correct.
34:35
SPEAKER_01: Because that is permission to build,
34:38
permission to release that is based on government blessing a
34:43
particular...
34:44
SPEAKER_01: a particular metric or a particular performance.
34:46
SPEAKER_01: And I think in our view,
34:48
SPEAKER_01: right now,
34:48
SPEAKER_01: I think we have faith that if Casey came up with a metric,
34:51
SPEAKER_01: it would be a metric that would be workable.
34:53
But I think we have concerns that if you make Casey the
34:56
bottleneck for that,
34:57
SPEAKER_01: that will lead to regulatory capture over time.
34:59
It will be a small group of companies who will shape what that
35:02
benchmark looks like.
35:03
SPEAKER_01: They'll shape it in a favorable way.
35:05
It will be easier for them to test against it and easier for
35:07
them to pass that test.
35:08
SPEAKER_01: And again, I think a lot of people don't think of that as licensing.
35:11
SPEAKER_01: I think it's important to say that that's how we hear it.
35:13
SPEAKER_00: Well,
35:13
SPEAKER_00: in addition to that,
35:14
SPEAKER_00: what about,
35:15
SPEAKER_00: you know,
35:15
SPEAKER_00: almost like a self-regulatory organization,
35:17
SPEAKER_00: like an SRO model.
35:18
SPEAKER_00: Yeah, let's talk about that.
35:19
SPEAKER_00: Right,
35:20
SPEAKER_00: where you have,
35:21
SPEAKER_00: okay,
35:21
SPEAKER_00: now you've got 10 companies,
35:23
SPEAKER_00: the 10 biggest labs in the country,
35:25
SPEAKER_00: or maybe in the world,
35:26
SPEAKER_00: that are coming up with their own benchmarking ideas.
35:29
SPEAKER_00: They have to be blessed by a regulator.
35:31
Maybe it's Casey,
35:32
maybe it's NIST,
35:33
maybe it's,
35:34
you know,
35:34
pick your agent,
35:35
maybe it's Commerce.
35:36
SPEAKER_00: Right.
35:36
SPEAKER_00: That has to then bless those.
35:39
SPEAKER_00: We do this a lot, actually, in the financial services space.
35:42
then those top 10 companies are basically just deciding exactly
35:45
what they want.
35:46
SPEAKER_00: So if you have a enterprising young startup that is trying to also build,
35:51
they can basically shape the rules to such that no one else is
35:55
going to be able to
35:55
SPEAKER_00: compete.
35:57
SPEAKER_00: That is how the regulatory capture game works.
35:59
SPEAKER_01: So there's a version of this proposal that we hear repeatedly,
36:04
SPEAKER_01: which is,
36:04
SPEAKER_01: it's just going to be voluntary.
36:05
SPEAKER_01: Don't worry about it.
36:06
SPEAKER_01: It's not regulatory, just voluntary.
36:09
SPEAKER_01: And so you can choose to participate, but you don't have to.
36:13
SPEAKER_01: Any voluntary participant gets some benefit by going through this regime.
36:18
SPEAKER_01: So you have to hit a certain performance benchmark,
36:20
SPEAKER_01: or you have to get certified by a verification organization.
36:24
SPEAKER_01: Once you're certified,
36:26
SPEAKER_01: It's not that there's a penalty.
36:28
SPEAKER_01: Once you're certified, you get this benefit.
36:30
SPEAKER_01: You get some liability protection.
36:32
And there are different ways that people think about what the
36:33
liability protection might be.
36:35
SPEAKER_01: You can think of the carrot in a lot of different ways.
36:39
SPEAKER_01: And then you can think of the verification system,
36:41
SPEAKER_01: the benchmarking,
36:42
SPEAKER_01: and then a bunch of different ways.
36:43
SPEAKER_01: But I think it's important for us to be clear.
36:45
SPEAKER_01: We hear this as also sort of a form of licensing.
36:48
SPEAKER_01: That's probably not exactly the right term for it.
36:50
But we hear that as an approach that would be really bad for
36:53
startups because...
36:54
Let's say that verification process is disproportionately hard
36:57
for little tech
36:58
SPEAKER_01: relative to big tech.
36:59
SPEAKER_01: If they can't do it without some competitive disadvantage,
37:03
SPEAKER_01: then even if it's purely voluntary,
37:05
SPEAKER_01: if the carrot's a really good one,
37:07
SPEAKER_01: like a really meaningful one,
37:08
SPEAKER_01: like some amount of liability protection,
37:10
SPEAKER_01: and they're disadvantaged in
37:12
going through that process to get that carrot,
37:14
then they are at a disadvantage,
37:15
ultimately.
37:16
SPEAKER_00: Significant disadvantage in that situation.
37:18
SPEAKER_00: Yeah.
37:19
Look,
37:19
I think it's a really important point and one that's not
37:21
understood.
37:22
SPEAKER_00: I mean, I go back to this.
37:24
We've mentioned this a couple times,
37:25
but it's a really important point to drive home.
37:27
There are these bills and these shells and these frameworks out
37:30
there that look very innocuous.
37:32
SPEAKER_00: They look very common sense.
37:34
SPEAKER_00: Yeah.
37:34
SPEAKER_00: But they are...
37:36
SPEAKER_00: wolves in sheep's clothing.
37:39
And this is the way that people sort of use these as beachheads
37:42
to get things through.
37:43
SPEAKER_00: And ultimately they're very,
37:45
SPEAKER_00: very consequential,
37:45
SPEAKER_00: not only for American consumers,
37:47
SPEAKER_00: but certainly for startups,
37:48
SPEAKER_00: the smallest of the small.
37:50
SPEAKER_01: I want to get to state stuff in a minute,
37:52
SPEAKER_01: but before we do,
37:53
SPEAKER_01: let's ask like the big question,
37:55
SPEAKER_01: which is your world,
37:56
SPEAKER_01: which is,
37:57
SPEAKER_01: is this going to get done?
37:59
SPEAKER_00: I think it will get done.
38:00
SPEAKER_00: In all of legislating,
38:02
you go through the phase where people are just throwing ideas
38:05
at the wall.
38:06
SPEAKER_00: You're throwing ideas, you're throwing ideas, you're throwing ideas.
38:08
SPEAKER_00: And I think the last year and a half has effectively been that.
38:11
SPEAKER_00: There's been a couple of real attempts.
38:14
SPEAKER_00: Even the moratorium play,
38:15
SPEAKER_00: which I think ultimately was the right thing to do at the right time,
38:20
SPEAKER_00: was completely misunderstood.
38:21
SPEAKER_00: And we'll talk about that.
38:22
SPEAKER_00: But it almost got done at the end of the day.
38:25
SPEAKER_00: And I think that that would have been sort of a
38:28
temporary reprieve ultimately wouldn't have solved the entire
38:31
issue we need.
38:32
SPEAKER_00: And then,
38:32
and that's why I think this comprehensive proposal that's both
38:35
come out from the
38:36
SPEAKER_00: white house,
38:36
SPEAKER_00: something that we published,
38:37
SPEAKER_00: that's why you're seeing a move towards,
38:39
SPEAKER_00: okay,
38:39
SPEAKER_00: we need something that's bigger than this.
38:42
SPEAKER_00: Um, but those are all kinds of steps in a process, right?
38:46
SPEAKER_00: And now we have moved into more meat on the bone.
38:50
We're starting to have real conversations within committees
38:52
about what could go in
38:53
SPEAKER_00: these packages.
38:54
SPEAKER_00: Yeah.
38:55
SPEAKER_00: And, um,
38:57
SPEAKER_00: you know, above all, I think the politics are pretty good right now.
39:00
SPEAKER_00: The politics on both sides of the aisle, people want to get things done.
39:03
SPEAKER_00: People want to protect the American public,
39:05
SPEAKER_00: which is ultimately the goal of these legislative proposals.
39:09
SPEAKER_00: But I will go back to sort of my,
39:11
SPEAKER_00: maybe my crappy analogy here,
39:12
SPEAKER_00: which is legislating is like big wave surfing.
39:15
I've never big wave surfed,
39:16
but you spend all of this time putting together a team,
39:20
and you know surfing the area and you get the jet skis you're
39:24
going to raise money
39:25
and sponsors and things like that but ultimately you cannot
39:30
ride a big wave until
39:32
the swells and the weather are correct and politics is the same
39:35
politics makes good
39:36
policy policy doesn't make good politics we need a political
39:40
environment by which
39:41
we can get something going and get it over the finish line and
39:44
i think that we're
39:45
SPEAKER_00: fast approaching that and it could happen
39:47
SPEAKER_00: between now and the end of the year, certainly.
39:50
SPEAKER_01: I think if we do a documentary on how an AI bill becomes law,
39:53
SPEAKER_01: it will be just as popular as a 100-foot wave.
39:56
SPEAKER_01: It will be just as popular.
39:57
SPEAKER_01: Yeah, I'm sure it will.
39:58
SPEAKER_00: But it's all progress, right?
39:59
SPEAKER_00: I mean, it takes a long time.
40:01
SPEAKER_00: We think about legislating here in 10-year trunks.
40:05
SPEAKER_00: And that is mind-blowing, especially in the tech space.
40:09
SPEAKER_00: But especially if you're working at a startup, oh my gosh, 10 years?
40:12
SPEAKER_00: Takes that long?
40:13
SPEAKER_00: But that is reality.
40:14
SPEAKER_00: And there are plenty of arguments,
40:17
SPEAKER_00: both from a philosophical perspective,
40:19
but also from people who are just big history buffs on whether
40:22
or not it was
40:23
SPEAKER_00: designed to be that way.
40:24
SPEAKER_00: And I think it is.
40:24
If you read what sort of the founding fathers were pushing for,
40:27
it is supposed to be difficult.
40:29
SPEAKER_00: It's supposed to take a long time.
40:31
SPEAKER_00: And politics are afoot.
40:32
SPEAKER_00: We're also in an election year.
40:34
SPEAKER_00: Things get really difficult.
40:35
SPEAKER_00: It feels like the Spider-Man meme.
40:37
Everyone's kind of pointing at each other and saying,
40:39
you didn't do this,
40:40
you didn't do this.
40:41
SPEAKER_00: But the quiet thing that's happening behind the scenes is a lot of very,
40:46
SPEAKER_00: very productive policy conversations.
40:48
SPEAKER_00: So look, I remain bullish.
40:51
SPEAKER_00: It is...
40:53
You could tell me that we're going to get something done
40:55
between now and the end of
40:57
SPEAKER_00: the year,
40:57
or you could say that it's going to take another two years and
40:59
I would not be
41:00
SPEAKER_00: surprised either way.
41:01
SPEAKER_00: But the point is,
41:02
is that there's positive momentum and everyone knows that we
41:04
need to get something
41:05
SPEAKER_00: done.
41:05
SPEAKER_01: Low pressure system in the North Atlantic starting
41:12
analogies yeah all right so we've talked a lot about dc yeah
41:15
but i think it's
41:17
arguable and it's hard to have a counter-argument that when you
41:21
think about ai
41:22
policy what's really happening in ai policy now it's
41:25
principally about state
41:26
activity um we're talking about the possibility of a bill
41:30
becoming a law in dc um
41:32
at in the states it's really happening over a thousand bills i
41:35
think i think the
41:36
SPEAKER_01: i've
41:37
SPEAKER_01: 1,500 roughly is like the count now at the state level in this session.
41:40
SPEAKER_00: Yeah, back-to-back sessions, back-to-back years.
41:42
SPEAKER_01: Yeah, like 1,000 plus last year, 1,500 plus this year.
41:47
SPEAKER_01: And that's not just bills that are being introduced.
41:50
I think last year there were somewhere around the 200 range
41:52
that were actually enacted into law.
41:54
SPEAKER_01: This year we've seen significant chatbot bills in Oregon and Washington.
41:59
SPEAKER_01: We've seen legislation in Utah.
42:01
SPEAKER_01: Stuff's moving in Tennessee.
42:03
California and New York are always typically a little bit later
42:06
in terms of what we
42:06
SPEAKER_01: see actually get enacted.
42:08
SPEAKER_01: And I'm sure there will be activity there as well.
42:11
SPEAKER_01: So how do you size up the dynamic now between what's happening in D.C.
42:15
SPEAKER_01: and what's happening in state capitals?
42:17
SPEAKER_00: Yeah, the way I think about it, it's two sides of the coin, right?
42:21
SPEAKER_00: In one sense, it's really productive.
42:22
SPEAKER_00: I mean, I would say that
42:24
First and foremost,
42:26
in all 50 states,
42:27
there will be an AI bill that's passed this year.
42:29
SPEAKER_00: And some of them can be relatively low level.
42:33
SPEAKER_00: I think probably low impact to startups or good governance things.
42:37
SPEAKER_00: To the extent of what we were talking about earlier,
42:39
SPEAKER_00: which is banding under 18s from using chatbots.
42:46
SPEAKER_00: or very,
42:47
very robust safety frameworks like we have seen in both
42:50
California and New York
42:51
SPEAKER_00: that were passed last year.
42:52
SPEAKER_01: Yeah.
42:53
SPEAKER_01: Can I talk about the Tennessee bill that was introduced?
42:55
SPEAKER_00: Yeah, yeah, please.
42:55
SPEAKER_00: Because it's not, it actually has gotten much better.
42:57
SPEAKER_01: This one's important.
42:58
SPEAKER_01: But I think it's actually like,
42:59
just as a starting point to understand where some state
43:03
lawmakers are starting,
43:05
the initial draft of this bill that we saw would make it a
43:08
class A felony in
43:10
SPEAKER_01: Tennessee.
43:11
SPEAKER_01: Right.
43:11
SPEAKER_01: That means murder.
43:12
right class a felony that it means like first degree murder not
43:16
like negligent
43:17
manslaughter you know it's first degree murder um to use an ai
43:23
system uh for an
43:25
activity that you would require a licensing a license for like
43:29
practice law or
43:30
medicine right the same that maybe there should be penalties
43:33
for that i mean i
43:34
think we generally think of course you shouldn't be able to use
43:36
ai in a way that
43:37
You couldn't,
43:38
if you were not using AI,
43:40
the prohibition should be at parity.
43:43
But the idea that it would be first degree murder to
43:47
essentially misrepresent the
43:49
SPEAKER_01: status of a license feels like an extreme starting point.
43:54
SPEAKER_00: It seems like a very extreme starting point.
43:57
And I would say that there's probably 25 states with something
44:01
that is almost
44:02
SPEAKER_00: similar to that,
44:03
SPEAKER_00: right?
44:03
SPEAKER_00: Yeah.
44:04
SPEAKER_00: So the conversation is...
44:07
Because I think there's a lot of really good work being done by
44:11
the states.
44:12
SPEAKER_00: But there's also some bills out there that are incredibly extreme,
44:15
SPEAKER_00: be crippling to the industry,
44:17
SPEAKER_00: you know?
44:18
SPEAKER_00: And,
44:19
SPEAKER_00: you know,
44:19
SPEAKER_00: I think in addition to that,
44:21
SPEAKER_00: we now have both federal and state data center moratorium.
44:25
SPEAKER_00: And make no mistake about it.
44:27
a moratorium or a ban on building data centers in the United
44:30
States is effectively a ban on AI.
44:34
SPEAKER_00: On the development.
44:35
SPEAKER_00: They can't exist without compute power.
44:38
SPEAKER_01: Well,
44:38
and probably particularly the ban component of it will skew
44:43
toward a harsher
44:44
SPEAKER_01: penalty on little tech.
44:45
SPEAKER_01: The hyperscalers will have
44:48
SPEAKER_01: they will be constrained,
44:48
SPEAKER_01: it will be problematic for them,
44:49
but they will likely in a world of constrained supply will
44:52
probably do relatively
44:54
SPEAKER_01: well.
44:54
SPEAKER_00: Right, exactly.
44:55
And we can talk about,
44:56
we talked about this on the last podcast,
44:58
but we can talk about it again.
44:59
SPEAKER_00: I mean, like where this,
45:02
SPEAKER_00: political advocacy in this push is coming from,
45:04
SPEAKER_00: you know,
45:05
SPEAKER_00: I,
45:05
SPEAKER_00: I think,
45:06
SPEAKER_00: I think what gets reported is much different than reality.
45:09
SPEAKER_00: I think in reports,
45:10
SPEAKER_00: it's,
45:10
SPEAKER_00: you know,
45:11
it's like a small community of people who really care about
45:13
their community or
45:14
SPEAKER_00: pushing back on this.
45:15
SPEAKER_00: And it's a very coordinated national campaign,
45:19
SPEAKER_00: um,
45:20
SPEAKER_00: uh,
45:20
SPEAKER_00: that has,
45:21
SPEAKER_00: you know,
45:21
SPEAKER_00: has been crippling in a lot of situations,
45:23
SPEAKER_00: you know,
45:24
SPEAKER_00: they've been,
45:24
SPEAKER_00: they've been quite effective.
45:26
SPEAKER_00: Um,
45:27
SPEAKER_00: But yeah, look, that's my general feeling.
45:29
And I think this also comes up in the context of the preemption
45:33
conversation that
45:33
SPEAKER_00: we're having,
45:34
SPEAKER_00: which we should discuss because I think it's important.
45:36
SPEAKER_00: There is a lot of really good work going on in the States,
45:38
SPEAKER_00: and there's a lot of really,
45:39
really unfortunate things that I think are happening that would
45:42
have really major
45:42
SPEAKER_00: impacts.
45:44
SPEAKER_00: So why don't we talk about preemption?
45:46
SPEAKER_00: Yeah.
45:46
SPEAKER_00: Let's talk about sort of,
45:47
SPEAKER_00: you know,
45:48
SPEAKER_00: I,
45:48
SPEAKER_00: you know,
45:49
my perspective is this conversation has been completely lost on
45:52
what is actually
45:52
SPEAKER_00: reality,
45:53
SPEAKER_00: but do you want to talk about this?
45:54
SPEAKER_01: And actually I'll, I'll, I'll, I think like we should own it.
45:57
SPEAKER_01: Cause I think this is a thing we really have believed in.
45:59
SPEAKER_01: And I feel like we, we have lost it.
46:00
Like we we've,
46:01
I feel like when we,
46:03
I was excited about the way that we approached it.
46:05
SPEAKER_01: And I feel like the general perception has moved against us,
46:08
SPEAKER_01: not in the direction that we've been pushing in.
46:10
I think just,
46:10
you know,
46:10
there are a lot of people have skepticism about the technology.
46:13
SPEAKER_01: The term moratorium is not one that people are responsive to.
46:16
So,
46:16
you know,
46:17
I get that there are things about it that have made our
46:19
arguments hard to make.
46:21
SPEAKER_01: We,
46:21
SPEAKER_01: you know,
46:21
SPEAKER_01: that there have been headwinds for us,
46:24
SPEAKER_01: but I feel like the core concepts here has,
46:28
SPEAKER_01: I don't know,
46:29
SPEAKER_01: hesitant to use the term unobjectionable,
46:30
SPEAKER_01: but I feel like the general thrust should be,
46:34
SPEAKER_01: I would hope would be one that would be between the 40 yard lines,
46:37
SPEAKER_01: which is,
46:38
SPEAKER_01: we should have an AI policy,
46:40
SPEAKER_01: what the constitution says we should have in lawmaking generally,
46:43
SPEAKER_01: which is there's Congress,
46:45
SPEAKER_01: there are state legislatures,
46:46
SPEAKER_01: they each have an important role,
46:48
SPEAKER_01: but they're not equivalent.
46:49
And Congress has primary responsibility for regulating
46:52
interstate commerce.
46:54
States have the ability to police harmful conduct within their
46:59
jurisdictions.
46:59
SPEAKER_01: It's very clear states have police powers.
47:01
So we're not talking about a world where it's just Congress or
47:04
just states.
47:06
SPEAKER_01: Often the rhetoric,
47:07
SPEAKER_01: that we hear is Congress hasn't acted.
47:09
So states need to act and states should act,
47:11
but consistent with their constitutional authority.
47:14
We've also heard some state lawmakers celebrate when they pass
47:16
a bill and they say
47:17
we're setting a national standard for AI development that I
47:21
think to some extent
47:23
SPEAKER_01: will run into legal headwinds.
47:24
In many cases,
47:25
that viewpoint I think will be held to be unconstitutional
47:28
because states cannot
47:30
interfere in interstate commerce in a way that creates a burden
47:34
that is clearly
47:35
SPEAKER_01: excessive relative to its in-state benefits.
47:38
SPEAKER_01: And so, again, the specifics here are complicated.
47:43
SPEAKER_01: They'll play out over a long period of time.
47:45
But I think it is very important that there be this distinction
47:50
between state
47:50
SPEAKER_01: authority and federal authority.
47:51
SPEAKER_00: Absolutely.
47:52
SPEAKER_00: And we should be very clear.
47:54
SPEAKER_00: The states have a role to play.
47:56
When we're talking about preemption,
47:57
we are not talking about preempting all state law.
48:00
SPEAKER_00: We are talking about
48:02
ensuring that there is both a federal framework that governs
48:05
the space and certain
48:07
SPEAKER_00: state frameworks that govern in this space.
48:11
SPEAKER_00: It is the conversation,
48:12
SPEAKER_00: I think,
48:12
SPEAKER_00: around the moratorium,
48:14
the meme that was created around it was what Congress was
48:17
attempting to do is
48:18
basically make it effectively illegal for any states to come
48:22
forth with any sort of
48:23
SPEAKER_00: AI legislation or regulation.
48:25
SPEAKER_00: And that's just simply not true.
48:27
SPEAKER_00: The state's
48:28
If the states did not pass a single additional bill from this
48:31
moment forward,
48:33
SPEAKER_00: they would have all kinds of authorities to regulate this space.
48:36
SPEAKER_00: And they are actively regulating this space.
48:37
SPEAKER_01: Well,
48:38
and so there was an exemption in the moratorium for generally
48:40
applicable law,
48:40
SPEAKER_01: which would have meant,
48:41
SPEAKER_01: like,
48:42
SPEAKER_01: you want to punish...
48:43
SPEAKER_01: fraudulent activity using AI.
48:45
SPEAKER_01: Right.
48:45
SPEAKER_01: You want to update your criminal law to make sure it can account for AI.
48:49
SPEAKER_01: It would be fine as long as the law is generally applicable.
48:51
And typically in public policy,
48:53
it's a best practice to not make something tech specific.
48:56
SPEAKER_01: Like, why would you want your new fraud statute to just single out AI?
49:00
SPEAKER_01: In theory, you should make it about harmful fraudulent conduct generally.
49:04
SPEAKER_01: Exactly.
49:05
SPEAKER_01: And then cover AI.
49:06
And so even the moratorium,
49:08
I think,
49:08
would have left a lot of room for states to act.
49:12
SPEAKER_01: Our view has been states could go after AI related harm specifically.
49:17
SPEAKER_01: So our view is not exactly in line with the moratorium.
49:20
I think if a state wanted to have not just a generally
49:22
applicable fraud statute,
49:23
SPEAKER_01: but a fraud statute related to AI specifically in our view,
49:27
SPEAKER_01: that's workable.
49:28
SPEAKER_01: That's not a problem for little tech.
49:31
But as you're saying,
49:32
the perception of the moratorium was really different.
49:34
I think the White House in the federal framework also took some
49:39
steps to really
49:40
SPEAKER_01: clarify what this would look like in a way that was positive.
49:44
So the preemption idea they have is Congress should preempt
49:47
state AI laws that
49:48
SPEAKER_01: impose undue burdens to ensure a minimally burdensome
49:53
SPEAKER_01: national standard.
49:55
It was focused more on preemption related to development than
49:59
harmful use.
50:00
SPEAKER_01: And I think that's a really positive thing.
50:02
And then there's another bullet that says this national
50:04
standard should respect key
50:06
SPEAKER_01: principles of federalism and not preempt,
50:08
SPEAKER_01: and then goes into a few different things.
50:10
SPEAKER_01: Traditional police powers,
50:12
SPEAKER_01: as we've discussed,
50:12
SPEAKER_01: state zoning laws,
50:14
SPEAKER_01: I think,
50:14
SPEAKER_01: obviously seem like that's an area where states should continue to act.
50:18
SPEAKER_01: So
50:20
SPEAKER_01: That seems like a valuable,
50:21
SPEAKER_01: you know,
50:22
SPEAKER_01: starting point,
50:23
SPEAKER_01: I think,
50:23
for these congressional discussions and making clear that when
50:26
Congress acts,
50:26
SPEAKER_01: it needs to do it consistent with its constitutional authority.
50:29
SPEAKER_01: It can't interfere with state police powers.
50:30
SPEAKER_00: Yeah, I agree with you completely.
50:32
SPEAKER_00: And I would also just say,
50:33
SPEAKER_00: I mean,
50:33
SPEAKER_00: just to put it more simply,
50:35
SPEAKER_00: I think what we're asking for and I think what maybe maybe,
50:37
SPEAKER_00: you know,
50:39
SPEAKER_00: I'll speak for the rest of the industry here.
50:41
we're asking for is a very similar setup to what every other
50:44
major industry minus
50:45
SPEAKER_00: the insurance industry has in the United States,
50:47
which is we have to have a federal framework and we have to
50:50
have a state framework.
50:51
SPEAKER_00: That is what happens in financial services and energy, et cetera.
50:55
SPEAKER_00: You go all the way down the list.
50:57
SPEAKER_00: I think it would be
50:58
SPEAKER_00: a really,
50:59
really strange thing to have the state of California or the
51:04
state of Utah govern
51:06
SPEAKER_00: all financial services in the United States.
51:08
SPEAKER_00: That's just not how we're set up.
51:10
Would we be in favor of abolishing the SEC and the CFTC and
51:13
their jurisdiction or
51:14
SPEAKER_00: the FDIC or the OCC and saying,
51:18
SPEAKER_00: oh,
51:18
SPEAKER_00: no,
51:18
we're just going to let California sort of set the standards
51:21
for everyone in the
51:22
SPEAKER_00: United States?
51:23
SPEAKER_00: I don't think that's good government.
51:24
SPEAKER_01: I think that's right.
51:25
SPEAKER_00: And I think that's a bipartisan issue, by the way.
51:27
SPEAKER_00: I do think that more and more that we're talking to people,
51:30
I think the more and more they understand sort of the
51:32
intricacies of this,
51:32
especially when we're talking about a framework that includes
51:35
more than just
51:36
SPEAKER_00: preemption.
51:37
SPEAKER_01: Sure.
51:37
Well,
51:37
I also think Democrats like it when it's New York and
51:40
California who are acting.
51:42
SPEAKER_01: But if you said,
51:42
SPEAKER_01: look,
51:43
this theory of authority would mean Texas and Florida can try
51:46
to set a national
51:47
SPEAKER_01: standard in issues that would advance Republican priorities.
51:50
SPEAKER_01: I think Democrats become much more anxious about that.
51:52
SPEAKER_01: Yeah.
51:53
SPEAKER_01: So I think...
51:55
our goal here is to kind of establish a principle that,
51:58
again,
51:59
it's not our principle.
52:00
It's a constitutional principle to delineate between state and
52:03
federal authority
52:04
SPEAKER_01: and do it in a way that has staying power,
52:05
SPEAKER_01: that would work for Republican states,
52:07
SPEAKER_01: for Democrat states,
52:08
SPEAKER_01: would work for Republican Congress,
52:10
SPEAKER_01: Democrat Congress,
52:11
SPEAKER_01: Republican administration,
52:12
SPEAKER_01: Democratic administration.
52:13
SPEAKER_01: And I think if we do that well, that will be good for Little Tech.
52:16
SPEAKER_00: Yeah,
52:16
SPEAKER_00: and I also think that we're kind of framing this in terms of fairness,
52:20
SPEAKER_00: and maybe we should.
52:21
SPEAKER_00: I mean, if we do not have a federal standard,
52:25
SPEAKER_00: you know, my firm belief is we will lose this AI race to China.
52:31
We will not have the capability of creating a national AI
52:34
market in the way that we need to.
52:36
And then ultimately we will lose to our competitor,
52:38
our number one foreign adversary.
52:41
SPEAKER_01: Yes.
52:42
SPEAKER_01: And I have wondered about
52:46
SPEAKER_01: why components of this concept are as controversial as they are.
52:49
So there are people who are skeptical of preemption,
52:52
which I think is fine.
52:53
SPEAKER_01: But sometimes I've tried to start those conversations by saying,
52:56
SPEAKER_01: I get you don't like the solution or particular solutions,
52:58
SPEAKER_01: but are you saying there's not a problem?
53:00
SPEAKER_01: Like, are you saying that for startups looking at
53:04
SPEAKER_01: We have resources to devote to tracking state policy.
53:08
SPEAKER_01: For the two of us, it's still overwhelming.
53:10
SPEAKER_01: It's still really hard.
53:11
SPEAKER_01: That's right.
53:11
We have more resources to vote to it than most of the startups
53:15
in our portfolio.
53:17
They're in a position where it's much harder to figure out how
53:19
do you comply with this?
53:20
SPEAKER_01: How do you track this bill that might affect you?
53:22
We brought to one of our portfolios this thing that would have
53:25
essentially put them
53:25
SPEAKER_01: out of business and they weren't even tracking it.
53:27
SPEAKER_01: And we were like,
53:28
SPEAKER_01: there's this thing moving that you need to take account of.
53:32
SPEAKER_01: And they weren't even following it.
53:34
And it would have really crippled their business in a state
53:35
that's a significant
53:36
SPEAKER_01: state that would have been impactful for their business.
53:38
SPEAKER_01: And so it seems to me like there's a problem.
53:42
SPEAKER_01: I have wondered even,
53:43
SPEAKER_01: again,
53:44
SPEAKER_01: I have found there to be just good guidance in looking to existing law.
53:47
One of the principles that we've looked to is a dormant
53:51
commerce clause principle.
53:52
SPEAKER_01: So the idea that it's not just that
53:55
Congress needs to take steps to preempt state action in order
53:59
to limit state authority.
54:01
State authority is limited even if Congress doesn't act because
54:04
of the interstate
54:05
SPEAKER_01: commerce clause.
54:06
And there's a principle within the dormant commerce clause
54:08
concept known as the
54:10
SPEAKER_01: excessive burden principle,
54:12
SPEAKER_01: which is the idea that from this test called Pike,
54:16
SPEAKER_01: that is Pike balancing.
54:17
SPEAKER_01: So out of state costs, the cost that a law would impose on people.
54:22
SPEAKER_01: the 49 states that didn't pass the law and on companies trying to operate
54:27
SPEAKER_01: throughout the country.
54:30
SPEAKER_01: Are those clearly excessive relative to the local in-state benefits?
54:36
And we recently wrote a post articulating that that sets up
54:40
this balancing test
54:41
SPEAKER_01: with costs on one side and benefits on the other,
54:44
SPEAKER_01: local benefits,
54:45
SPEAKER_01: costs.
54:45
SPEAKER_01: And typically,
54:46
SPEAKER_01: and there are legal scholars who have written about this,
54:48
SPEAKER_01: judges aren't well positioned to do this balancing.
54:52
SPEAKER_01: And so what we were calling for was
54:54
SPEAKER_01: more of giving judges tools to do this cost benefit analysis.
54:58
So just giving them data so that they can help to understand
55:02
the costs and the
55:03
benefits and giving them tools to do this weighing because
55:05
they're going to be
55:05
SPEAKER_01: doing it.
55:07
SPEAKER_01: And it's better for them not to just do it on vibes.
55:09
SPEAKER_01: There's a lot of like judicial delineation that is just kind of like,
55:13
SPEAKER_01: we know it when we see it,
55:14
SPEAKER_01: we have a general feeling about it.
55:16
SPEAKER_01: Shouldn't this process be more informed by data?
55:19
SPEAKER_01: More data might result in some laws getting struck down,
55:21
but it might actually provide an evidentiary record that helps
55:23
some laws get
55:24
SPEAKER_01: upheld.
55:25
SPEAKER_01: So to me,
55:25
SPEAKER_01: it feels totally non-controversial,
55:27
SPEAKER_01: got pushback from certain components of the law and AI community.
55:32
SPEAKER_01: And so some of that just doesn't quite make sense to me.
55:34
SPEAKER_01: I mean,
55:35
I think you know it better than I do because you're not
55:37
surprised to see the
55:37
SPEAKER_01: politics play out in all these different ways.
55:39
But it feels to me like there is kind of a approach here that's
55:43
constitutional,
55:44
SPEAKER_01: that gives judges real tools to make empirical assessments that will be
55:49
good for states be good for congress good for companies that um
55:53
we care about
55:54
smaller companies with smaller legal teams so i hope that's the
55:56
direction we go in
55:57
no i i think that and it's and it's incredibly helpful we're
55:59
going to leave
56:00
everyone with a little bit of homework which is go and research
56:03
pike and figure out
56:03
what pike is yeah um yeah you know i look i i would also maybe
56:08
just to just to
56:09
conclude here i i do feel very very bullish in the direction
56:13
that we're heading and
56:14
SPEAKER_00: i do think that
56:15
the right people are now starting to get to the table and push
56:19
in the right direction.
56:21
SPEAKER_00: The maybe one call to action here is,
56:25
SPEAKER_00: it's an unfortunate reality,
56:26
SPEAKER_00: but especially in tech,
56:28
SPEAKER_00: most people don't take action until something really,
56:30
SPEAKER_00: really negative has happened.
56:32
And we're,
56:33
I think probably within a year of a state bill passing that
56:37
would be
56:38
SPEAKER_00: pretty awful for the industry collectively.
56:41
SPEAKER_00: And I also think that probably pretty awful for the industry,
56:44
SPEAKER_00: but also cloaked in a way that's probably good for the American public,
56:49
SPEAKER_00: but probably not very good for the American public.
56:51
SPEAKER_00: Now, I don't have a crystal ball.
56:52
SPEAKER_00: I can't predict exactly what that is, but something is going to happen.
56:57
SPEAKER_00: We need people to continue to come to the table.
57:00
SPEAKER_00: We need people to be active on this and share ideas and concepts,
57:05
SPEAKER_00: especially from the little tech community.
57:06
And that's what we're,
57:07
you know,
57:07
principally working on on a regular basis.
57:10
SPEAKER_00: But it is, you know, the time is now.
57:12
SPEAKER_00: The time is now to act and be there because as we say all the time,
57:16
SPEAKER_00: you know,
57:16
SPEAKER_00: if you're not at the table,
57:17
SPEAKER_00: you're on the menu,
57:19
SPEAKER_00: you know.
57:20
SPEAKER_01: Colin, this is super fun.
57:21
SPEAKER_00: Super fun.
57:22
SPEAKER_00: Thanks, Matt.
57:23
SPEAKER_00: Talk soon.