Magic Tricks and The Imitation Game
We’ve been inundated with “miracles” lately, as people discovered first Midjourney and DALL-E and then ChatGPT. There’s been a steady stream of amazement and wonder as people post examples of output generated from prompting these AI systems with “creative” suggestions.
Don’t get me wrong, I am impressed, too. In fact, I think there can be no doubt that all of these “machines” have passed the Turing test: meaning that we cannot, as humans, tell the difference between human and machine outputs.
That said, I was glad to read, among all the breathless exuberance, a more balanced reflection that gave equal and thoughtful time to both wonder and worry. Filmmaker Frank Pavich’s opinion piece in the New York Times discusses his complicated reaction to the “collaboration” between Canadian director Johnnie Darrell and Midjourney. Darrell prompted the program to create artwork from a version of Tron produced by legendary Chilean auteur Alejandro Jodorowsky, famous for a richly imagined version of Frank Herbert’s Dune that was never made. The joke is that Jodorowsky never made a version of Tron either.
Pavich helps us appreciate some of the truly impressive subtlety at work in these “fakes,” which are so good that we want them to be real. What is, of course, all too real is the effectiveness of these imitations in convincing us that they could be the product of human creativity.
Musician and songwriter Nick Cave seems less impressed. Writing on his Red Hand Files blog, Cave thinks that the lyrics that ChatGPT spits out “in the style of Nick Cave” just suck. “ChatGPT’s melancholy role is that it is destined to imitate and can never have an authentic human experience,” Cave says.
Cave believes that songwriting emerges from a kind of human suffering that AI cannot, by definition, ever experience. “ChatGPT has no inner being, it has been nowhere, it has endured nothing, it has not had the audacity to reach beyond its limitations, and hence it doesn’t have the capacity for a shared transcendent experience, as it has no limitations from which to transcend.”
Adrian Ho has been playing with GPT-3 and I’ve appreciated his reflections, too. Ho wonders if we are entering an era of “post-competence” in which we happily accept that AI can and does do better at many things that humans do with greater effort and less reliability. His more optimistic take? “I don’t think it’s the rejection of information, accuracy or being right, it’s recognizing that those aren’t jobs for us, they’re jobs for AI.” Adrian thinks that AI is giving us new things to think about and new ways to think about them. I think that’s right.
My friend Jake thinks all of this augurs apocalypse, but he’s not filled with dread about it. To the contrary, Jake is glad. Not because he sees this as the end of everything, but as the end of human supremacy. Jake calls his near ecstatic take on this “happy nihilism.”
I want to share Jake’s lightheartedness, and sometimes I do. But I am also worried about the “magical” effect of all these new technologies, namely, misdirection. When we become over attentive to the whiz bang effects of these new systems, we lose sight of the massive human effort to not only deliver what current systems do, but the ultimate goal of the so-called “strong program” in AI: conscious machines.
Last summer I re-watched two documentary films about humans and AI. Game Over, which documents the epic contest between IBM’s Deep Blue system and chess Grandmaster Gary Kasparov, and AlphaGo in which the AI Deep Mind is pitted against 9 dan Go champion, Lee Sedol. Both films elide narratives of triumph and tragedy. The other thing the films have in common is that they present different versions of the same magic trick. We might call the trick, after Turing, The Imitation Game. As we are invited into these high stakes contests between a human and a machine, we fail to notice the deception at work, so effective is the misdirection of the drama.
Deep Blue and Deep Mind were not simply machines, but human/machine teaming systems. What we miss is the scope and scale of the effort required to beat single human opponents at the very limited contests of chess and Go. Not that these are not impressive achievements, but to equate them with the dawning of intelligent machines is to rely too much on Turing’s eponymous test as the gold standard by which we should measure what we call artificial intelligence.
Are Midjourney, ChatGPT and DALL-E impressive? To be sure. Are they intelligent? Just as surely not. Is “intelligence” actually the goal of these efforts? Without a doubt these systems, and the legions that are coming in their wake, are in some meaningful sense “superhuman.” The question I want us to spend more time on is this: “What if these machines are not here to replace us, but to help us imagine better worlds”?