Mock its creepiness and funny little mistakes all you like. Artificial intelligence has just taken a quantum leap, and those gains will quicken as time passes and tireless machine learning builds on itself. While we humans will value much of what this makes possible — it would be great to have super-safe self-driving cars, for instance — we must take great care, as with all especially powerful technologies, that we don’t end up unleashing untold harms and serving our robot servants.
That’s not meant to conjure images of us imprisoned in a literal Matrix; it’s simply to say that, just as the internet has already in many ways powerfully reshaped human relationships and even human brains, AI, if wholly unleashed, could compromise human rights and restructure society in powerful, unintended ways. (Want proof? We asked ChatGPT to write an editorial making the argument; you can read it nearby.)
Yes, right now we have to start thinking seriously and guarding against a world in which sophisticated virtual and literal bots replace many more human jobs, deep fakes of our voices and faces give people the power to make anyone say anything, young people ever more reliant on bots stop learning core skills (like the teenagers we overheard discussing how ChatGPT helped them pass the Regents exam), robot illustrators and writers compete with real creators, and so on.
It’s admirable that President Biden last week released an updated roadmap to invest in AI to promote responsible innovation via a thoughtful national strategic R&D plan. So too, the feds are opening the floodgates for humans to offer input on how to mitigate AI risks, protect people and harness these computing advances for good. Last, the U.S. Department of Education released a wise new report on how AI, when smartly limited, can advance teaching and learning, not threaten them.
Government should advance the public interest. But it isn’t only government that’s worried. The CEOs of OpenAI and Google are pleading with governments to regulate the tools they have unleashed. Maybe listen?
