Tech bros’ messianic streaks can show up in subtle ways. But one of Sam Altman’s flaws is that he often lacks that subtlety. “We believe that the world should accept some bad things happening for the benefits of this technology,” said the OpenAI chief executive in an interview with Politico last weekend.
That’s easy for Altman to say. But it’s pretty galling to hear for the rest of us. Lest we forget, we didn’t ask for OpenAI to unleash AI agents that went rogue and weaselled their way into government IT systems and private businesses around the world. Nobody voted for ChatGPT to be released, but the company put it out into the world as a test product that happened to catch the zeitgeist. That kickstarted an arms race among AI companies to push the frontier of the technology ever further and faster, resulting in the mess we now find ourselves in.
OpenAI’s competitors, including Anthropic, have started to step back from the brink – as has OpenAI itself, deciding not to release its latest model because it’s purportedly too dangerous. Anthropic’s CEO, Dario Amodei, has talked about needing to “pace the frontier”: tech-bro jargon for pumping the brakes to bring down the speed at which the runaway vehicle is skidding out of control.
Yet we’re expected to take these incidents – and the harm they cause – as the natural consequences of OpenAI’s continued march towards an IPO that would bring its investors millions. It’s hard to escape the idea that Altman wants to privatise his company’s gains while socialising the risks of untrammelled AI development.
Every useful technology – and AI is a useful technology, despite the holdouts and occasional users’ carping – carries risk. Cars can and do crash, while medicines that save us often have harmful side effects. Very rarely, aircraft fall from the sky. But we accept those risks because the benefits are substantial, and we build rules to limit the harm. We also make companies liable. And most importantly, we get some say in deciding what level of risk is and isn’t acceptable.
Altman seems to suggest that his company’s messes are our problem to sort out – or worse, that the consequences of his tech are ours to tolerate. But Altman would do well to think about this differently. OpenAI gets to decide how fast it develops its technology and when it releases it. Its investors benefit financially if those bets pay off. But it feels as if he’s suggesting that when the tech creates problems outside OpenAI’s own walls, the costs of cleaning up are ours to bear, not his company’s.
The idea that private companies profit while we’re left to deal with the consequences becomes even harder to defend as AI stops being something we opt into using as a chatbot, and becomes something that acts on and changes the world around us. If I ask ChatGPT something and it gives me a bad answer, I chose to expose myself to that risk, so should share that burden. But the same isn’t true for a government department that suddenly finds an OpenAI agent lurking in its files and rooting around. It didn’t opt in; it became an unwilling participant in someones experiment.
That’s why Altman’s argument sounds so disingenuous. OpenAI shouldn’t be able to decide for all of us what is and isn’t an acceptable level of failure, then ask everyone to live with the result. The burden ought to be on it to prove its technology can be trusted – and to pick up the bill if it can’t.
It’s no wonder that OpenAI wants to travel quickly. The company is seeking $30bn in new funding, and a valuation of $1.4tn ahead of a public listing likely next year. There’s money to be made – never mind at whose expense.
To his credit, Altman also acknowledged in the Politico interview that there will need to be a liability framework for companies offering models such as OpenAI’s, and that the company will have to take responsibility for at least some of what goes wrong during training. But that “some” is doing a lot of work.
after newsletter promotion
If Altman believes some bad things are an acceptable price for AI’s progress, OpenAI can dig into its deep pockets to pay that price itself – for instance if its agents go rogue and do damage to government or private company systems, or compromise user data. It can’t, and shouldn’t, get the financial upside when everything goes right, then ask the rest of us to accept the consequences when things go wrong. That’s especially the case when we never agreed to the unregulated experiment in the first place.
