"I know Kung-Fu"
On writing like Vonnegut, existentialist drama, and why the "New Bing" needs to stop thinking it's in The Matrix
After the much-heralded public launch of ChatGPT, it hasn't taken long for people to tune in to other developments in artificial intelligence -- prominently including the soft launch of Microsoft's ChatGPT-based "New Bing", which it bills as a "search experience", rather than a mere search engine. Early adopters are already putting it through stress tests, including an unnerving New York Times review, existentialist crises, and a crash course in writing like Kurt Vonnegut.
■ One of the plot points that earned derision for "The Matrix" was the moment when Keanu Reeves, as Neo, "downloaded" a martial art to his mind and declared, "I know Kung-Fu." The necessary conceit was that, inside the Matrix, the rules of data applied, rather than the rules of physical existence.
■ But it's not such a laughing matter in this real-world launch of artificial intelligence, where a computer can be "taught" to "know" subjects through exposure to what's been published on the Internet. False certainty is the kind of folly that human beings learn not to demonstrate after humiliating themselves in front of others -- think of anyone who becomes a know-it-all after taking a 101-level course in a subject.
■ Feelings like humiliation and the course corrections that come with those feelings are impossible to download. They're human experiences, and deeply organic ones. How could you explain to a digital "being" what it's like to blush? Human beings are mostly our own minds, of course, but no small part of what we learn comes through the corporeal existence. And there's no way to give that to an AI.
■ No matter how clever we get with large language models, part of "learning", as we humans will always understand it, comes through a process. And the process isn't always constructive -- sometimes you have to un-learn something you knew differently before, either through necessity or because the facts themselves changed. That sort of consciousness just isn't going to be replicable within artificial intelligence.
■ That's also why rules are a complicated matter: Everyone knows that some rules are more important than others, and we also know that different rules prevail under different circumstances. There may be Ten Commandments, but there are hierarchies among them, and there are cases when even those commandments come into conflict with one another. Resolving the conflicts is an essential aspect of human experience and intelligence.
■ The people launching artificial intelligence into the world -- even with the best of intentions -- need to be intrinsically aware of the limitations of rules, their ability to institute and document those rules, and to deal with the unavoidable conflicts among them. There's a lot of power at hand, but a whole lot of errors waiting to be made, too.
Looks benign enough, but just wait