I just learned that the de-aging process used to produce scenes with a “younger” Harrison Ford in the most recent Indiana Jones movie was produced using an AI process that matched the Ford’s face in each newly filmed moment with a perfectly matching facial expression from archival stock of the actor.
The movie was fine, even if the third act was a little undercooked. I just want to point out two things. First, this is a natural extension of the LLM model of artificial intelligence: pulling from library of information provided (i.e. the internet or footage of Indiana Jones punching Nazis) and then reassembling that information to produce a new product. When we debate whether or not these pieces of software constitute actual “intelligence” what we are really arguing about is whether or not the substrate is sufficiently simple, sufficiently inert for the act of assembly to constitute an act of intelligence.
Case in point, nobody is arguing that de-aging Harrison Ford betrays true intelligence on the part of the software. The material being acted upon is already coherent, it’s just being reordered to convey a new message ( i.e. scenes in a different movie). Similar things could be said about ChatGPT. It’s just searching through pre-existing text and ideas, sifting for relevance, and reordering to optimally assemble into an updated product.
Operating within this metaphor, what would constitute intelligence is the that can sift through a primordial substrate of inorganic, individually incoherent components, and assemble them into original pieces of coherent information; sparks of cognition. An example of this would be to take the ambient, ineffable sentiments implied by the collective set of questions being asked by ChatGPT users (hopes, fears, feelings, etc) and produce not just answers to questions not yet answered by humans, but to answer questions not yet asked. To sift through information, break it into into its smallest possible molecules of cognition, and contribute to the broader collective body of knowledge by assembling new thoughts.
This sort of process has been occasionally posited in the form of a “matter compiler” in science-fiction, which is essentially just a 3-D printer at the molecular level. That remains pretty far off, as best I know. I suspect the same will be true of true artificial generalized intelligence. We know an awful lot about how molecules are assembled, the problem with producing a matter compiler is largely one of cost. We know comparably less about how neurons firing are assembled into acts of generative intelligence, of creativity. We will know no doubt get there, but getting there usually happens well before we cross the chasm of cost, of material feasibility.
But yeah, the new Indiana Jones movie was okay.
Economist Writing Every Day 