AYESHA RASCOE, HOST:
Anthropic CEO Dario Amodei says AI companies are moving dangerously fast. In an essay posted online yesterday, he says, quote, "AI brings risks, and because it is such a powerful technology, these risks are serious."
This is just the latest alarming headline about artificial intelligence. On Thursday, Anthropic published a report in which it says it intervened in research being conducted using its products, fearing users may have been trying to develop bioweapons. On Tuesday, a safety researcher announced he had resigned from Anthropic because he feels it and rival OpenAI are more focused on competition than ensuring their products remain under human control.
Thomas Dekeyser teaches at the University of Southampton in the U.K. and is author of "Techno-Negative: A Long History Of Refusing The Machine." He joins us now. Welcome to the program.
THOMAS DEKEYSER: Hi. Great to be here.
RASCOE: What do you make of Jacob Coxon's resignation from Anthropic and his claim that Anthropic and OpenAI are, quote, "racing straight to self-improving superintelligence and gambling with our lives" and could threaten humanity within a decade?
DEKEYSER: So when I read a statement of someone resigning, I'm always interested in, OK, but how did we get to that place in the first place? Right? These AI agents were produced by someone. The conditions under which an AI agent goes rogue were produced by the company itself. So there's something interesting about this narrative, this sort of focus on the rogue AI rather than on the company itself.
RASCOE: I have to point out that Anthropic is a financial supporter of NPR, but obviously, we don't shy away from covering them. Both companies - Anthropic and OpenAI - had models break out of testing environments and gain access to systems out in the world this summer. So it sounds like, from what you're saying - is that the concern is less that AI is getting stronger, but that the companies are making it stronger?
DEKEYSER: There's a little bit of a danger of talking about AI in terms of the speculative future potential risk rather than maybe what it is already doing right now - what kind of harms it might be having, performing in other kinds of contexts.
RASCOE: Well, what do you think the harms are now?
DEKEYSER: I think there's environmental issues. I think there's issues with the rolling out of data centers in places that leads to people not having access to clean water or people's bills going up, and so on. There is something about the narrative of existential risk, specifically if it comes from within, let's say, Anthropic or OpenAI - is that it sort of further underscores this idea that AI is kind of this ultrapowerful thing that will revolutionize everything. And I'm hesitant because it's a narrative that actually is beneficial for companies that are looking for immense investment, right? So I think this is why I kind of always sort of hesitate in the face of these conversations.
RASCOE: When you hear a story about Anthropic stopping these customers because they thought they were working on bioweapons, is there enough protection from bad actors who could use them to do harm?
DEKEYSER: I don't know if we currently we're at a moment where AI could be used in incredibly destructive ways, let's say, for purposes of terrorism or something. I don't know if AI is currently at that point. What I do know is that, for the companies themselves, they are more interested in sort of self-regulating rather than getting national or international regulation sort of looking over their shoulders.
RASCOE: AI is not the first technological innovation that has changed how humanity operates. What kind of advancements do you see as similar to AI that changed the way society works?
DEKEYSER: Maybe electricity, sort of early 20th century - how that sort of led to all kinds of various changes on so many different levels of society. And then I would say the second example would be the internet - internet infrastructures and what all that did.
RASCOE: Well, does AI, then, represent a paradigm shift when it comes to humans and technology because AI can create art and perform jobs that humans are supposed to do, which is not like the internet?
DEKEYSER: I think, technically, it's obviously able to do things that were not possible before. But the promise is that it will radically change everything. So I think we need to be careful about overstating what it can already do and what it will be able to do in the future, again because I think there is a benefit for those specific companies hoping to kind of push these technologies. There's a benefit in the kind of general population fully believing that.
RASCOE: That's Thomas Dekeyser of the University of Southampton. Thank you so much for joining us.
DEKEYSER: Thanks very much. Transcript provided by NPR, Copyright NPR.
NPR transcripts are created on a rush deadline by an NPR contractor. This text may not be in its final form and may be updated or revised in the future. Accuracy and availability may vary. The authoritative record of NPR’s programming is the audio record.
