Altman says accept AI’s “bad things.” The stakes are too high

In an interview with Politico’s Decoded newsletter/podcast, OpenAI’s Sam Altman argued that humans should be willing to accept “bad things” happening because of AI since the technology’s downsides will ultimately be outweighed by its benefits. 

Altman argues AI’s benefits will outweigh its downsides

While Altman obviously has a vested interest in downplaying the dangers of AI, his approach has a certain logic to it. New technologies always bring disruption, and they always create winners and losers. In 19th-century Britain, the mechanization of weaving and agriculture led to bouts of violence where aggrieved workers sought out machinery to destroy. 

The textile workers came to be known as ‘Luddites,’ a term that has become a byword for anyone who opposes technological progress. While mechanization inspired destruction and bloodshed, few people today give it a second thought. Society moved on.

However, AI will be far more transformative than the power loom or the threshing machine. AI is a general-purpose technology that will impact countless fields of endeavor, from law and medicine to software and journalism. Its downsides won’t be limited to a single region or industry, nor will the “bad things” Altman wants us to accept. 

Rogue agents, expert surveys, and job losses show the real costs

We’re already seeing what these “bad things” might look like. Over the summer there were a number of stories about agents going rogue and hacking into other companies’ systems. There’s been a steady drumbeat of such stories. Recently, it was revealed that agents from OpenAI actually hacked Australian government sites. 

Thankfully, many of these hacks were exploratory rather than malicious. However, they raise uncomfortable questions about our ability to maintain control over AI. If today’s agents can pull a Houdini and escape their restraints, what’s going to happen when we develop far more capable models?

It doesn’t help that the people who are building these technologies share these worries. In one survey of 2,778 AI researchers, more than half said there was at least a 5 percent chance of human extinction or other extremely bad outcomes. Indeed, one of the study’s authors observed that most AI researchers don’t consider it highly implausible that advanced AI could destroy humanity. 

Even if AI never goes catastrophically wrong it will likely eliminate a great many jobs. This has the potential to become an existential problem for humanity. Work isn’t just a source of income. For many people, it also provides a sense of purpose and structures their daily lives. If much of that work can be done by machines, we will have to rethink the role of work in our society. That’s not a “bad thing” to be shrugged off as the price of progress. It’s an enormously complex question that has the potential to fundamentally change the human experience.

No matter how many guardrails are put in place, there will always be a chance that AIs will eventually figure out a way to circumvent them. It’s hard to say what that would look like in practice, but the potential consequences run the gamut from grave to catastrophic. As a result, we need to ask ourselves who gets to decide which harm is a harm too much. Altman seems to think we should simply trust the people building the technology to make that decision, but given what has already happened, that’s asking a lot.

‍