Why Tech Leaders Want to Slow Down AI

[00:00:03] Welcome to Tech News Briefing. It's Tuesday, September 15th. I'm Julie Chang for The Wall Street Journal. Worried your electric vehicle might run out of juice before you find a plug? Automakers are betting on a new solution that's actually kind of old. We'll explore the tech behind extended-range EVs. Then we're diving into the AI doom debate. As some in the industry warn AI could spell the end of humanity,

[00:00:31] what steps could tech giants take to stop it? But first, extended-range electric vehicles, or E-REVs, are the auto industry's latest attempt at solving drivers' range anxiety. It's a fully electric car equipped with a gas-powered generator that can charge the battery on the go. But will drivers embrace this middle-ground technology?

[00:00:57] My colleague, Amani Moiz, sat down with WSJ reporter Christopher Otz to talk about how E-REVs work and whether extended-range will expand the EV consumer base. Can you walk us through how these new E-REVs work and how they're different from the hybrids most people are familiar with? Yes. The main difference is that these new E-REVs are electric vehicles first,

[00:01:22] and the engine is only there to act as a generator charging the battery. The engine has no mechanical connection to the wheels. You cannot drive this car in the same way you drive a gas car. What we've seen with most plug-in hybrids here to date is that they really are gas vehicles

[00:01:45] that figure out a way to fit in a battery and provide some all-electric range. But they're really fundamentally gas cars with a battery attached, and the range that they provide in terms of all-EV driving has not been very much or very useful to most people.

[00:02:08] So these new E-REVs will have much more electric range because they are fundamentally electric vehicles just with a security blanket of a gas engine. So why are automakers suddenly so interested in these E-REVs? What problem are they trying to solve? It's important to say that the concept is not entirely new. So General Motors did an E-REV with the Chevy Volt back in 2010.

[00:02:35] It was really groundbreaking at the time, and BMW had an E-REV version of the i3. So it is kind of an old concept where it is coming back is in larger vehicles, big pickup trucks, big three-row SUVs. These big vehicles have had the most trouble going to electrification because of towing and the fact that they need gigantic batteries.

[00:03:04] So some automakers are hoping that the E-REV functionality, where you never have to worry about running out of range, where you can tow a trailer because you have that security of the gas engine, will be the perfect answer for the particular problem. And what kind of range are we talking about here? How often would the typical driver actually need to use the gas generator? These new crop of E-REVs that are coming to the market will have,

[00:03:32] we're told, over 100 miles of range. For the most part, you should almost never have to put gas in it. The gas generator is really only supposed to come into play on unusual situations, such as long road trips, where maybe you're driving through the cornfields of Indiana. There's no charging station anywhere in sight, but you do have gas stations and you can complete the trip.

[00:03:59] One of the big unknowns is how much range you get from that gas generator as well, how far you can go without having to refuel. These are things that we'll learn as these vehicles actually get rolled out. But that is for sure something to watch because while you could fill up at a gas station, it could also be inconvenient if the vehicle can't go as far between filling up as people are accustomed to.

[00:04:29] Another hangup that consumers have had when it comes to EVs is the price tag. Are these E-REVs going to help in that area? That's a great question. Another thing that is yet to be seen in terms of how these vehicles are priced. But there's no doubt that while E-REVs might be less costly than like a comparable full electric vehicle,

[00:04:52] they will almost assuredly be more costly than the comparable gas vehicle. The question is, which consumers value having an EV and the EV driving experience enough to pay extra for these vehicles versus buying the same type of vehicle just in an all gas version?

[00:05:18] That was WSJ reporter Christopher Otz speaking with our colleague Imani Moise. Would you consider buying an E-REV? If you're a listener on Spotify, leave us a comment with your thoughts. Coming up, from escaping test sandboxes to autonomous hacking, how real is the threat of rogue AI? And what are industry leaders doing about it? That's after the break.

[00:05:53] For decades, the idea of AI ending human civilization lived strictly within the realm of science fiction, like the movie Terminator. But today, that conversation is no longer happening on movie sets, but inside the labs of Silicon Valley's most prominent tech companies. In the last week, we have seen a shift in the AI discourse.

[00:06:15] First, an anthropic researcher announced he was quitting the industry over fears that companies are building AI systems that could quickly spiral out of control. Then, on Saturday, several leaders at top AI companies, including Dario Amadei, Sam Altman, Demis Hassabis, and Elon Musk, publicly agreed that we need to slow the pace of development of this technology. As investors pour billions in a high-stakes race towards artificial general intelligence,

[00:06:44] the companies themselves are beginning to sound an unprecedented alarm. But how real are those warnings? And can we still slow the tech down? Here to discuss that with me is the journal's deputy tech and media editor, Sam Schechner. Sam, what's driving AI researchers and executives to raise the alarm on AI development now? I think a lot of people are asking why now.

[00:07:10] Because in the AI industry, these concerns have been around for quite some time. But a few things have come together in recent months that made this conversation break out of the Silicon Valley bubbles where it was taking place and into the mainstream.

[00:07:26] And the first one really in public was this hugging face hack where what we at first heard were just a few AI models inside of OpenAI managed to somehow get out of a sandbox, a containment that they were in on a computer server, and hack another company. It was the first real tangible example of what AI researchers have long feared as a quote-unquote loss of control incident.

[00:07:55] And then we found out there were others from other companies. At the same time, inside the companies, they've been warning that they're getting closer to this long theorized threshold that they call recursive self-improvement. That's when an AI is able to program its successor. The idea being that if it can do it, it can do it at scale and you'd start to create superhuman intelligences very quickly that could conceivably go off in things like that hugging face attack, but far worse.

[00:08:26] So how would those companies actually stop the rogue AI? Well, that's sort of the question. They don't really know how they're going to stop the rogue AI. I mean, at some basic level, you could shut off the computer servers that they run on. But if it's truly rogue and has managed to copy itself elsewhere, then that doesn't necessarily work.

[00:08:46] So a lot of what they're working on are techniques to control the AIs while they're still controllable and to try to train them to stay good, to sort of do what people ask, to stay aligned with human values. There are skeptics, though, who say, OK, a lot of the most damaging stuff we're talking about is still far beyond some of the capabilities of today's AI systems.

[00:09:13] But effectively, stopping the AI isn't always possible. So the reports of AI escaping test environments, deceiving humans or hacking systems, are those isolated lab glitches or signs of a deeper safety issue? Well, that depends on who you ask. There are definitely people in these labs who think that this is a sign of a deeper issue. What you see is that AIs are susceptible to something called reward hacking.

[00:09:43] People do that, too. If you're rewarded for doing a particular thing, you're going to figure out the most efficient and fast way to do that thing. And that might involve cutting some corners. In my mind, it's like when you ask a genie to do something. You make a wish and the genie does exactly what you ask, but not what you meant. Right. And that is sort of what's happening with AI. And the genie stories almost never turn out well for the person making the wish.

[00:10:08] So people in the industry are saying we need to undertake a more responsible pacing of development. Any sense of what that would look like in practice? It's hard in a competitive commercial environment to really imagine companies voluntarily slowing down. Part of this public signaling is an effort to sort of give themselves a little bit of cover, a little bit of ability to coordinate on that.

[00:10:32] Anthropics CEO, he said that they would commit to giving employee-like access to outside evaluators, outside auditors to verify that they are actually keeping their commitments. That's a step. They want that to be verifiable. And if multiple companies did that, that might create some of the incentives they're talking about. That's the most concrete thing that I've heard of so far. Beyond that, we're talking about could there be laws passed? Could there be a regulator?

[00:11:02] Could there be an international treaty? Those things are all much further out. It bears mentioning that the disagreement about how real these risks are is still there. And there are some people who say this is an effort to try to create regulation that would stifle smaller competitors. And there's other people who would say, you know what, this is a distraction from real risks that we're already seeing today about, you know, energy consumption from data centers.

[00:11:29] Speaking of regulation, what has the government's response been to these concerns? Congress has certainly taken note. Politicians have seen that this is an important issue. They are calling for legislation, things like a kill switch, mandatory transparency and reporting requirements for companies.

[00:11:48] At the same time, there are voices who very much say, you know what, we don't need more rules that could slow down American AI or that there's already rules on the book. Product safety regulations and other ways that the government could be investigating some of these companies. I mean, hacking is a crime. So it's conceivable that after an AI agent does so, you could investigate the company that did it. And so maybe new rules aren't needed.

[00:12:15] You know, President Trump yesterday pushed back on the need for tighter regulations linked to AI. There are different points of view. And without the president on board, legislation would be a much heavier lift. That was WSJ Tech and Media Editor Sam Schechner. And that's it for Tech News Briefing. If you're a listener on Spotify, be sure to leave us a comment. Today's show was produced by me, Julie Chang, with supervising producer Katie Ferguson.

[00:12:42] We'll be back later this morning with TNB Tech Minute. Thanks for listening.