The Question ChatGPT-3 Can’t Answer

If coders are the unacknowledged legislators of our time, then OpenAI’s engineers are now rewriting the Constitution. Who watches the watchmen?

Micah Sifry

--

As the generative pre-trained transformer language model ChatGPT-3 takes the internet by storm, with 100 million monthly active users in January just two months after its launch, questions are rapidly rising about the ways the tool may transform politics, along with complaints about whether it may be biased, either against particular political ideas or movements, or against particular identities. This matters a lot, in the same way that choices that other big platforms make about what content to show users shapes and influences how we perceive the world and act in it. And unfortunately, the excitement ChatGPT-3 is generating over its human-like responses to many prompts is getting much more attention than who is governing the tool and what values and biases they are bringing to the OpenAI project (which is now a for-profit owned by a group of investors including Microsoft, Reid Hoffman, Elon Musk, Peter Thiel, Sam Altman, Infosys, and Khosla Ventures after starting out as a nonprofit).

What happens as interest groups use tools like Chat-GPT to flood the inboxes of lawmakers and regulators with artificially generated content? Those systems are already mostly broken, so they’ll just be even less useful. How much bigger will the ecosystem of “pink slime” news-sites that are actually fronts for wealthy interests grow when a bot can write reasonably articulate “news” stories? What happens to polling or persuasion when you can hook up Chat-GPT to answer your phone for you? (Someone has already built a podcast airing the edited conversation of two chatbots, Adam and Bella, so technically this isn’t a challenge.) What happens when Chat-GPT claims to be giving politically neutral and accurate answers to questions when those answers are actually biased?

OpenAI, the makers of ChatGPT, have focused the tool on delivering answers that will feel right to the greatest number of users (and indeed, their PR has the same focus). “We want as many users as possible to find our AI systems useful to them ‘out of the box’ and to feel that our technology understands and respects their values,” the…

--

--