- 6 minutes ago
Category
🗞
NewsTranscript
00:00Co-chairs the AI caucus in the House of Representatives.
00:03Beyond that, pursuing a master's degree in computer science with a specialization in AI.
00:09And we've had this kind of parlor game, this recurring segment on the show,
00:12where you ask a member of Congress if Congress is equipped to regulate AI.
00:17Do they understand the technology well enough?
00:19I think that Don Beyer, congressman from the 8th District in Virginia,
00:22stands a better chance than most at answering in the affirmative on that.
00:24You know I'm going to ask him anyway.
00:25You're going to ask him anyways. You've got to do it.
00:26Again, it's a parlor game. It's part of the routine here.
00:28Congressman Beyer, great to have you with us.
00:30And let's start with AI.
00:31We were talking about Taiwan, all the other issues that were discussed,
00:34presumably between President Trump and President Xi.
00:37I'm curious sort of where you think this conversation has evolved over the course of the last week.
00:41You've listened to the speeches that were delivered at the UN,
00:43the hope from many world leaders that there be some agreement,
00:46broad multinational agreement on an AI framework.
00:48You heard President Trump trying to rebrand this as superintelligence,
00:53not artificial intelligence,
00:54and indicated that he thinks things are just fine and should stay the way that they are.
00:59Your take on how this conversation has evolved and indeed where it needs to go next from a policy perspective.
01:04David, thank you.
01:05At breakneck speed, we're in a very different place than we were just two months ago.
01:09I mean, two months ago, we had, you know,
01:12we had the Casey Institute at National Institute for Standards and Technology
01:16that supposedly was doing the overall,
01:18I think it was trying to do overall safety for AI.
01:21Otherwise, you had an administration that wanted no guardrails whatsoever,
01:25that wanted nothing to interfere with innovation,
01:27that kept saying, but China, when you try to do anything responsible.
01:32Then the mythos came out from Anthropic,
01:35which could have destroyed cybersecurity and every financial institution in the world.
01:39And the White House said, oh, my goodness, something's going on.
01:42And then the open AI's bots started invading other databases at Hugging Face,
01:49and then yet another computer scientist resigned.
01:52And all of a sudden, my mailbox and my phone is ringing off the hook from constituents saying,
01:58what's happening to our future?
02:01Will there be a future for my children?
02:04So, happily, I mean, I know it's very confusing,
02:07but happily, we have people from all over the world finally converging,
02:11saying, yes, there are many, many wonderful things coming from artificial intelligence,
02:15but there are genuine existential risks also,
02:19and we need to be responsible enough to address them.
02:22Even if the majority of people agree that something needs to be done,
02:26given the different opinions, even within, I look at, you know,
02:29some folks in the Republican Party have been very out front about saying,
02:32yes, we need to do something about putting a pause on data centers,
02:35and then you have the president saying, you know, full speed ahead,
02:38we're going to call it superintelligence, there's divisions in the Democrats, too.
02:41Even if people agree something needs to be done, do people, here's my usual question,
02:45understand it well enough to know what that needs to be?
02:48And realistically, can there be any kind of uniform action,
02:52given the divisions across all political lines right now?
02:56Christina, I think you asked the most interesting question,
02:58because the best computer scientists in the world,
03:02people like Demis Asavas at DeepMind or Dario Amadei at Anthropic,
03:08they don't understand why it works.
03:10They're very clear that, you know, they put this big thing together
03:12with all the huge chips and lots and lots of parameters,
03:17but no one really understands why this apparent,
03:23it's not really conscious, or maybe it's not conscious,
03:25but its ability to solve problems that mankind's not been able to do
03:29for tens of thousands of years.
03:31All these math Olympiad problems, for example.
03:34That's why, even though we don't understand it,
03:36we still have a responsibility to try to regulate it or to keep it safe.
03:41And I can talk more about that if you want.
03:43Well, let me get into that.
03:45I mean, you live and represent a district very close to Reagan National Airport.
03:49I know that you've come out in recent days
03:50about the strategic management of airspace routing trajectory system,
03:54this AI system that's being used or going to be implemented at that airport.
03:58And so there is a very real-world application of these technologies.
04:02And correct me if I'm wrong,
04:04it seems like you feel like it's not adequately tested yet
04:07to understand how effective that might be
04:09or what kind of knock-on effects it might have.
04:12David, on the airport, AI,
04:16I have the privilege of representing National Airport,
04:19which has the busiest runway in America,
04:21and tragically had the last major commercial airline disaster
04:25a little less than two years ago
04:27when the American airplane hit the military helicopter
04:30and 67 people died.
04:32So when Secretary Duffy, who is a friend,
04:35rolled it out and said,
04:36we're going to try this new AI system
04:38at Reagan National Airport in Dulles,
04:41my constituents went a little crazy.
04:44And I pointed out that to do that,
04:47we need to explain it very clearly
04:49that the air traffic controllers still get to make the final decision,
04:52that we weren't going to rely on AI we don't understand
04:56to decide whether a plane can take off or land safely.
04:59I think we had some good conversations.
05:01The Secretary has guaranteed to me and my staff
05:06that no decision will be made
05:10after the airplane has left the runway,
05:12which I think is what our constituents needed to hear.
05:14When you look at the AI landscape as it is right now,
05:17can you think of one tangible, doable step
05:20that you think Congress could take in the short term
05:23to make AI safer
05:24and to make the general public feel better about AI?
05:28Yes, I think the very most important thing,
05:30Christina, is to set up a structure.
05:32There are a number of different ideas out there.
05:34My favor of the week is to use something like the Federal Reserve,
05:38where there is a federal component
05:40with people appointed by president,
05:43confirmed by Congress, overseeing it.
05:45But you still let the labs,
05:47Anthropic, OpenAI, Google, Gemini,
05:50hire some of the very best people to safety test.
05:53The parallel is we don't, you don't take any drugs,
05:57I don't, that the FDA hasn't insisted be safety tested.
06:01We don't drive cars and let the safety of that,
06:04they have been tested.
06:05So we need to do the same thing with the lab models.
06:09You know, most AI in the world is not large language models.
06:14It's, you know, much more basic things.
06:17When do you need your next dental appointment?
06:19What we need to be concerned about
06:20are the regenerative AI that's coming up with ideas
06:26seemingly out of nowhere
06:27that we often trust because they're breakthrough,
06:30but sometimes they're not exactly right.
06:32They could be hallucinations.
06:34You know, a story that really knocked me back this week
06:37was the one out of Australia,
06:39that you had an open AI agent
06:41hacking into an Australian government website.
06:44And that was many months ago.
06:46It took months for the company,
06:47I presume, to identify that that had happened
06:50and then time for them to bring that
06:52to the Australian government.
06:53So maybe picking up on Christina's question,
06:55what more can the government do to,
06:57if not encourage, mandate these companies
07:00to have to be more forthright
07:01when there are incidents like the ones
07:03that we've been talking about here,
07:05be that in Australia or in the private sector?
07:07And I note Axios this morning out with a report
07:09that open AI, anthropic, and security researchers
07:11are investigating tens of thousands of incidents
07:14in which their frontier models took steps
07:16that outside evaluators would consider problematic
07:18according to sources.
07:19And those are the ones they know about.
07:19Yeah.
07:21And David, Christina,
07:22that's actually a low-hanging fruit.
07:25We've been down this road before
07:26with cybersecurity threats.
07:28So some time ago, maybe 10 years ago,
07:30we instituted mandated
07:32automatic cybersecurity threat notices.
07:35So there's a big national structure
07:38that's open to everybody.
07:39Every time there's a cybersecurity threat,
07:41it is mandated that it be reported.
07:42And then the good news about that
07:44is lots of other people can jump in
07:46to figure out how to fix it.
07:48And so we need to do the exact same thing
07:49with these AI deals.
07:51And, you know, it was upsetting
07:53when I went into Hugging Face
07:54and got into databases that were,
07:56you know, belonged in the private sector.
07:58But much more so when you discover
08:00that they're going into government databases.
08:02Remember how worried, at least these Democrats,
08:05when we feared that those people
08:07were turning over social security records
08:09to maybe private companies
08:11or to Elon's large language models,
08:14was just think of any of these lab models
08:17going into social security, IRS,
08:20our HIPAA databases on our medical records, etc.
08:23All of a sudden, there literally is
08:24no privacy in the world.
08:26Congressman, let me ask you last
08:27about something unrelated here,
08:29and that is what President Trump
08:30calls his triumphal arch.
08:32And as I read through what you've written
08:34and said in recent weeks,
08:35you've got some different names for it.
08:37The Vanity Arch, I think, is one that you've used.
08:39The other, the Illegal Gridlock Increasing
08:42Narcissistic Arch Project.
08:45We learned...
08:46Are you going to use enough adjectives there, David?
08:47Enough, yeah.
08:48David likes adjectives, that's okay.
08:50String of modifiers.
08:51But we've heard some new information
08:53from the president about what he wants it to be.
08:55There could be snipers on the arch.
08:57It could be some kind of arsenal
08:58or garrison for weaponry.
09:02Again, you live very close
09:04to where this would be built.
09:05You have relatives, close relatives,
09:07who are buried in Arlington National Cemetery
09:08right across the river
09:09from where this would be constructed.
09:11What would you like to see happen here
09:13in terms of process,
09:15barring him not building this completely?
09:17What would you like to see happen here
09:19in the weeks and months ahead?
09:21Well, I'd like to see it not built completely.
09:23We, problematically,
09:25the public sector boards
09:28that will approve it or not approve it
09:30have all been appointed
09:30by the president this year.
09:32So it's going to get approved in that sense.
09:35Our local Arlington County, Virginia,
09:37is fighting it.
09:38The veterans groups,
09:39you know, related to all the folks
09:40buried in the cemetery,
09:41including my parents,
09:43my grandparents,
09:44are fighting it.
09:45There's no citizen support for it.
09:47I hope we'll go back to recognize
09:49that the president's called this
09:51a monument to himself.
09:53I'm not making that up.
09:54He really says,
09:55this is to honor me
09:56and all that I've done.
09:58Every other thing we've had like that
09:59have been presidents long dead.
10:01The Washington Monument,
10:02the Lincoln Memorial,
10:03the Jefferson Memorial.
10:04Kennedy Center.
10:05Kennedy Center, yeah.
10:07And when we have a war,
10:08we build the Vietnam War Memorial
10:10or the World War II Memorial.
10:14There's just no reason for this, Arch.
10:16People are particularly concerned
10:18because of the sacrilege
10:20to the 80,000 men and women
10:22who served our country
10:23who were buried right there
10:24at the cemetery.
10:25But then there are all the issues
10:26of the sniper and the airplanes
10:27and the traffic,
10:28and it will be a mess.
10:31All right.
10:32Don Beyer, thank you so much
10:33for joining us.
10:34Always good to speak to you.
10:34We appreciate it.
10:35And good luck with your degree.
10:37Have you gotten the degree yet?
10:38We wanted to check in on that.
10:39Are we on track for graduation?
10:41No, I'm just working on
10:41my latest course this morning
10:42before coming on with you.
10:43I've got maybe,
10:45I can only fit in one course
10:47a semester with this other job
10:48I have.
10:48Understood.
10:49So two more years.
10:50It would not look good
10:50if you were able to finish it quickly
10:52in terms of the lawmaker
10:52that we know you have to do.
10:53One credit at a time.
10:54Many a working man and woman
10:56has done the same thing.
10:56I'll take this.
10:56You
Comments