Earlier this month, a researcher from Anthropic publicly resigned and said the company’s AI model could pose a threat to humanity.
Leaders of AI companies chimed in.
Anthropic CEO Dario Amodei published an essay calling for international cooperation to slow down AI development. OpenAI CEO Sam Altman joined in on social media saying he agreed with Amodei. OpenAI is now delaying plans for its public listing.
What do artificial intelligence researchers make of those warnings?
WUWM education reporter Katherine Kokal called up UW-Milwaukee professor Linnea Laestadius, who specializes in how AI impacts mental health and communities.
This conversation has been edited for length and clarity.
Katherine Kokal: We're coming off a month where AI leaders made all of these warnings. I want to know how those sounded to you as a public health researcher specializing in AI.
Linnea Laestadius: I wasn't super surprised by the arguments or the narrative. So there's been a large strain within the AI development community for a long time that people will sometimes casually refer to as "doomers" or "doomerism."
There has for a long time been this concern about catastrophic harm that AI systems could create. And arguably as systems become more advanced, the risk of that may increase. Some of that is the possibility of people using AI tools to create harm, right? So for bioweapons, I think the Houthi rebels maybe were using AI for developing things. That's one risk. The other falls under kind of the category of misalignment. And that's really getting at, are these AI systems doing what we want them to do in the way we want them to be doing them, and are we on the same page with the system about what are the appropriate values?
So a lot of the things we've been seeing where it hacked into a system and nobody asked it to do that — they gave it a very benign task to do and then it didn't have sufficient guardrails around that. Then the AI system took what was the most expedient way to solve the problem, which was not something the company had anticipated.
I'm not super surprised by this because it isn't a new narrative at all. The question is how much of this is real and how much of this is hype? And I think that's an ongoing question.
I think there was a pretty strong backlash saying, 'Wait, you're the one developing this technology. What do you mean it can kill us all? You stand to benefit from this.' What do you think about those responses?
The CEOs of these companies really balance this idea that their products may have tremendous benefit for society and humans, but also run these potential for pretty catastrophic risks. But certainly there are people saying, "Well, why do you keep developing these? And maybe you just were sloppy and irresponsible in the work you were doing." There are people both inside and outside AI communities who have made these arguments.
It's very hard as an outsider to judge any of this. I think it's also important to focus on the more immediate risks that we know are posing concern. So risks to the environment, surveillance, the impact on education, exploitation of labor in less developed countries in terms of training models. So there's a lot of risks that they don't want to talk about, maybe in favor of these catastrophic risks.
WUWM is a service of UW-Milwaukee.
