- 6 hours ago
Category
🗞
NewsTranscript
00:00You have been one of the most consequential lawmakers in the state when it comes to AI legislation.
00:05You proposed and passed SB53, a Transparency in AI Act, among other things, protect whistleblowers,
00:16disclose safety incidents for AI models, requiring companies to publish their safety protocols.
00:24As you look back on SB53, is it mission accomplished for AI regulation, or is it the best we could
00:33do at the time?
00:35I think it's neither. It's definitely not mission accomplished.
00:39And I think we addressed one specific issue, ensuring transparency around safety protocols.
00:46We know that AI has so much potential to make the world a better place and brings significant risks.
00:52And all of the big labs have all gone all over the place and said,
00:59we have safety protocols, we're testing our models, we're doing everything we can, but trust us.
01:06And we want to make sure that there's transparency.
01:10And then the whistleblower protections and the reporting of critical incidents.
01:13So it was a really important step.
01:14The strongest such regulation in the country and other states are now basically cutting and pasting our work,
01:22with permission.
01:23We encourage it.
01:24And Illinois went a little bit beyond where we went.
01:27We need to catch up.
01:29So it addresses an issue, but there are other issues that we need to address.
01:33So this is going to be a work in progress.
01:37Speaking of the AI companies asking us to trust them,
01:41and that was something that a previous speaker today,
01:43Yashua Bengio, one of the godfathers of AI, professor at the University of Montreal,
01:48he said that we should not be blindly trusting AI companies.
01:52How much leeway should they have?
01:56I agree with Professor Bengio, and we work closely with him over our two-year journey.
02:07And, of course, we want to make sure that these labs can succeed
02:12and that they are able to create products, models that ultimately make the world a better place
02:20and that serve humanity and not vice versa.
02:25And so that means we should not simply say, let's trust them.
02:30They'll do the right thing, and we can just focus on other things.
02:35We need to make sure that we are verifying and that we have guardrails in place.
02:40Again, this is the basic thing, that AI, we need to ensure that AI is serving humanity
02:45and making humanity better and more successful and not vice versa.
02:52So SB53 succeeded one of your other proposed bills, SB1047,
02:57and the governor vetoed that, and there were two elements to that bill.
03:01One was a kill switch for models that perhaps get out of control,
03:07and then liability insurance for AI companies that do wrong.
03:10What was unpalatable, maybe to the industry or to the governor, about those two ideas?
03:17Yeah, so those were probably the two most contentious pieces,
03:20and we did a lot of work on that bill.
03:22We had a lot of changes around taking into account differences between open source
03:26and not open source, and other, we had a lot of feedback,
03:32and we made a lot of good changes to the bill.
03:34Is feedback a euphemism?
03:35Yes, feedback and attacks, and it was the whole spectrum.
03:40It was a lot of constructive feedback, to be fair.
03:43But the core of that bill was that if, because you can already be sued.
03:49If you create a model, like any product, that burns someone's house down
03:54or causes some harm under just regular tort law, you can be sued.
03:59But what we created was requiring testing of large models before they're deployed.
04:07And if you do not test, if you violate the law,
04:12and then it leads to some catastrophic harm, huge harm,
04:17then you can be sued only by the attorney general, not by anyone,
04:21just by the attorney general, and requiring a kill switch.
04:25That bill, I mean, it was very healthy, because I think we really changed the conversation
04:31around AI safety and the need for guardrails.
04:34And as the governor said in his veto message, the bill created its own weather system,
04:39and that is absolutely true.
04:42But, you know, at the time, I was, and I have very thick skin.
04:45I come out of San Francisco politics, which we refer to as a knife fight in a phone booth.
04:49So I have very, very thick skin.
04:53But, you know, I was called a D-cell, a doomer, that I should be a fantasy writer,
04:58that these risks were all completely made up.
05:00They were just regulating math, all the things that were said, and the discourse.
05:07And ultimately, two years later, we were right.
05:10And we saw with the release of Mythos and the testing that was done,
05:14that, you know, these risks are not fake.
05:20They're not made up.
05:21They're real.
05:22And we have the ability to place the right guardrails.
05:26And we should not do what we did with data privacy and social media,
05:31which is just to do nothing.
05:33It's 2026, and there's no federal data privacy law.
05:36There's basically nothing on social media other than protecting them from lawsuits.
05:40And we've seen the problems, and this technology is significantly more powerful and bigger
05:50and going to have way more impacts, and we should not repeat that mistake.
05:55And it's not about shutting it down.
05:57It's about making sure we understand the risks and we get ahead of the risks.
06:00And, again, making sure that AI serves humanity, not vice versa.
06:04Well, do you think that the public sentiment around AI and the legislative consensus
06:09around what these bills should include are changing?
06:12For example, earlier today, Emily Chang mentioned the wave of anti-AI sentiment
06:18in polling and the booing of graduation speeches.
06:21She called it a hate wave.
06:22Do you think that some of these ideas that were rejected two years ago
06:26are maybe now more in the mainstream?
06:28Absolutely.
06:29And you look at, there's even, like, potentially movement in Congress.
06:34Now, we want to make sure it's not movement with a limited regulation
06:37and a broad preemption.
06:39Like, we have to guard against that.
06:42But I think the politics have completely shifted here.
06:45And so two years ago, when I was being called a decel and a fantasy writer,
06:49I don't think that would be the case today.
06:51People might agree or disagree on different things.
06:54And even on SB53, we didn't have, other than Anthropic,
06:58we had no support from the larger tech companies.
07:01There was one in particular that was sort of quietly, really not loving it.
07:06I'm going to guess that's OpenAI, but you can be discreet.
07:09And I will leave that for your speculation.
07:13And now we had the Illinois version of SB53 with an audit requirement.
07:19And if I think OpenAI and Anthropic both supported it.
07:23So that, and so there's been a real shift.
07:29And I will say, you know, it's not to bring politics into it,
07:33but running for Congress now, this year,
07:36I get to hear what's on people's minds.
07:39And when you're a candidate, when you're elected official,
07:41you hear what's on people's minds.
07:42When you're a candidate, you really hear what's on people's minds,
07:44which is great, it's very healthy for democracy.
07:47And AI is probably the most common question that I get asked about.
07:55I like it.
07:56Every house party, every event, on the street.
08:00And people are asking, not in a confrontational,
08:04this is what you must do.
08:05It's a very, it's like, which is often the case.
08:08Like, I think this, and I want to make sure that you agree with me.
08:10It's very much open-ended questions.
08:14Like, there's this issue, a lot of jobs might go away.
08:17There are safety risks.
08:18We have deep fakes.
08:19We have chatbot issues with kids.
08:21What should we do?
08:23And people, honestly, they know there's an issue.
08:25They know government needs to do something significant.
08:28And they often just don't know what it is.
08:31And so they're asking me.
08:32And that, I think, shows how the politics have shifted.
08:36And then the last thing I will say is, when SB 1047 was pending in 2023,
08:40with all of the opposition to it, there were multiple statewide polls in California,
08:46including one that was conducted collaboratively by supporters and opponents.
08:50They all showed support pushing 80%.
08:54So in the public, this is not controversial.
08:57Well, and because it is on people's minds,
08:59there are a lot of ideas out there right now for how to regulate AI.
09:03And I want to get your take on a few of them.
09:04So if you are elected in the fall, Senator Bernie Sanders becomes one of your colleagues.
09:10He has proposed two ideas.
09:12One is a data center moratorium.
09:15And I know some states like New York are also considering that.
09:18Good idea or bad idea?
09:19So I don't think a moratorium is the way to go.
09:23And we also, but we need real regulation to make sure that the public is not being harmed by data
09:30centers.
09:32Sometimes the data center conversation happens exclusively in the context of AI.
09:37But as I say to people, if you want to go to whichever search engine or AI tool you use,
09:43go and ask to list all the industries that rely on data centers.
09:47It's a massive list, frequently having nothing to do with AI.
09:50And so I don't think a moratorium is the way to go.
09:53With that said, we do need good land use rules on it.
09:57You shouldn't be able to put data centers anywhere.
10:00I'm not a fan of like surrounding a town with data centers like we've seen in some parts of the
10:05East Coast.
10:07And data centers should absolutely not be putting strain on the electric grid.
10:12And so whether that means generating their own energy or helping upgrade the grid,
10:19they should be paying for their own usage and not be jacking up people's costs
10:28because they're putting that strain on the grid.
10:31And same with water.
10:32We should find a way to, whether it's recycled water or whatever the case may be,
10:37we can put good rules in place for data centers without putting a moratorium in place.
10:43Another idea he's proposed is the U.S. government taking a 50% ownership of AI companies.
10:49My understanding is to fund a sovereign wealth fund,
10:52perhaps in part to ease some of the dislocations that might come from AI.
10:5750%, there might be some people in this room that think that's high, but is it a good idea?
11:01Yeah.
11:02So a sovereign wealth fund, I think, is a reasonable idea.
11:0550% seems pretty high.
11:07But I think the bigger picture is when you look at polling in Europe and the U.S.
11:14around AI anxiety, it's much higher in the U.S. than in Europe.
11:17And that's a very easy explanation because in Europe, they actually have a real safety net.
11:22If you lose your job, whether it's because of AI or for any other reason,
11:25you're still going to have health care tomorrow.
11:27Your kid's still going to have health care for their asthma.
11:30You're still going to have income support.
11:32You're still going to have child care.
11:34You're not going to become destitute.
11:37And that is an indictment of the U.S., frankly, in terms of the lack of a safety net,
11:42which Donald Trump has made profoundly worse in his attack on anyone who's not super wealthy.
11:48And so we need to shore up that safety net.
11:51That's why I support Medicare for All.
11:52And we need to have true universal income, universal health care.
11:58Your health care should not be connected to your job.
12:01It's a dumb system that we have in this country.
12:04It harms people.
12:05And we should have true universal health care and universal child care.
12:09And we should have more robust unemployment.
12:13We did that on unemployment insurance.
12:15We did that during the pandemic.
12:17And it worked out well, longer and more.
12:20And you give people, when you have that safety net with income and health care, et cetera,
12:26if their job does disappear, they will then have the space to be able to figure out what's next.
12:33And maybe that means going back to school or taking classes or trying to start a business,
12:39whatever the case may be, they will have the space to be able to figure out what their next step
12:43is
12:43instead of being out on the sidewalk the next day.
12:46And so whether that is through, you know, I've been public.
12:51I support using tax policy, not just on the AI left.
12:55If you're an industry that is now benefiting tremendously from AI because you are reducing your head count,
13:00there should be taxation to compensate for that, to help us with the safety net.
13:06Whether you do it that way or through a sovereign wealth fund,
13:09I think there are different ways to do it.
13:11But we have to do something.
13:12Well, is it more important to shore up the safety net or protect workers from being dislocated from AI in
13:17the first place?
13:18Well, I think shoring up the safety net is incredibly important beyond AI.
13:22In terms of protecting workers, I think there are things you can do to,
13:26you're not going to stop technology,
13:27but there are things you can do to, for example, prolong it.
13:31Like we have a bill that I voted for twice.
13:34The governor vetoed twice to say for autonomous big rig trucks that you have to have a safety monitor.
13:39We have millions of people in this country who drive trucks for a living.
13:43I don't want to see them all have their jobs evaporated overnight.
13:46So should there be a period of time when there has to be a safety monitor?
13:50I think most people in the public probably would want a safety monitor, at least for a while,
13:54for big rig trucks that are going to be on our roads.
13:56The industry doesn't like it, and I understand why they don't like it.
13:59But to me, that was a reasonable approach.
14:02And so I think we need to look for various ways not to stop the technology,
14:07because that's not going to work,
14:09but to sort of extend and cushion the blow
14:16and give people space to be able to figure out what is next for me and my family.
14:22Okay. Well, somehow we're already short on time, and I wanted to get to a few other things.
14:26If you get to Washington, do you plan to extend your leadership on tech issues to the Congress?
14:33And also, do you think that the government should take a look again at competition laws?
14:39It's been very quiet for the last two years,
14:41particularly as they allow the tech giants to extend their dominance into this next wave of innovation.
14:47Yes, and yes. I will absolutely focus on tech policy.
14:52I think it is incredibly important, and that will be, among other things, a focus for me.
14:58And I also believe that there are more and more particularly,
15:03not only there are long-term members who are fantastic on tech policy,
15:07but newer members who are coming in and who really understand the need for significant change.
15:16And I think there will be opportunities, and so that will be a focus.
15:20On competition, yes. And I actually had a bill this year.
15:23It was SB 1074. It was an unfortunate number, because it was just 1047,
15:28but flipping the seven and the four.
15:29So it probably triggered some people.
15:31But it would have banned self-preferencing by the big five.
15:37Some of the practices that Google, Meta, Apple, Amazon, and Microsoft engage in
15:43are pretty outrageous in terms of screwing over mid-size companies and startups,
15:49and it ends up harming consumers and innovation.
15:54And so we worked with a big coalition of startups led by Y Combinator,
15:59along with consumer advocacy organizations like the Economic Security Project,
16:04to advance this bill.
16:05We got it out of one committee.
16:07Big tech is really powerful, and they killed it in the second committee.
16:11I know there's been efforts in Congress, bipartisan, to advance that kind of law,
16:15and I hope that happens again.
16:17Last question.
16:18How much AI are you bringing into your day-to-day life as a legislator,
16:22as a candidate, and in your personal life?
16:24I use it to try to figure things out for myself,
16:28and I increasingly do the whole back and forth,
16:30because sometimes the first answer is clearly garbage.
16:33And if you keep pushing it and go back and forth,
16:35maybe it gets a little tighter and a little better.
16:39In terms of in the legislative process,
16:43I don't think it's being used that much.
16:45And I've thought about taking text from a bill and feeding it in and saying,
16:49tell me all the weaknesses and loopholes in this.
16:52I think we probably should be doing that.
16:54I'm sure there are advocates that do that.
16:57It helps them figure out what things need to be tightened.
17:00But I think in government in general,
17:02we're probably a little bit behind the private sector on this.
Comments