A panel discussion, “AI’s Role in Unlocking Human Potential,” followed tech pioneer and prophet Ray Kurzweil’s remarks at the COSM Technology Summit. The panelists raised a number of doubts about the Singularity and the transformations Kurzweil envisioned.
Oren Etzioni, CEO of the Allen Institute for Artificial Intelligence (brainchild of Paul G. Allen, Microsoft co-founder), cautioned against hype about superhuman AI: “Exponentials are very important. If we extrapolate exponentials, we can be exponentially wrong.” He made a distinction, for example, between retrieval of data and understanding it: “To take one tidbit from his talk, he talked about the ability of computers to read 100 million sentences. But they actually only retrieve words.”
Etzioni also reflected on the strictly limited focus of artificial intelligence achievements. “It taught itself to play chess in four hours? But that was in 2018. It was structured to play chess. What has it done for us lately?”
George Montañez, Assistant Professor of Computer Science at Harvey Mudd College, took issue with Kurzweil’s claim that AlphaGoZero needed no instructions to beat humans at the game of Go: “For a system like this to work, a human must define the incentive structure, also encoding the assumptions.” The sheer power of a computing system does not cause it to do anything at all. Programmers must define both the goal and the outcome. Afterward, flush with success, he noted, the programmers may believe themselves when they say, “Oh, I had no hand in it.” But the goal is really their baby, not the machine’s baby.
Robert J. Marks, director of the Walter Bradley Center for Natural and Artificial Intelligence, offered a historical take on Kurweil’s claims. One of the goals of alchemy was the creation of a “homunculus,” the manufactured human that can think for us. As we began to understand nature better, such pursuits discredited alchemy. “I believe that the Singularity will live in history as like the homunculus,” he predicted.
On a practical note, moderator Matt McIlwain, Managing Director of the Madrona Venture Group, asked, “What are the constraints to more, faster progress for the AlphaGos of the world?”
Montañez responded, “What AI and machine learning systems are doing is really revolutionary. But our systems are only as good as the data that goes into them. We don’t necessarily know how biased our data systems are.” He cited Amazon’s recent widely-publicized misfortunes with an AI hiring system which turned out to be biased against women. It had to be sidelined following an uproar.
Google could tell a similar sad story about entirely unintended racism in its images program that could be traced to problems with machine vision. The machines don’t think; they simply reflect the mass of data scarfed into them, from a variety of sources — much of which a thoughtful human might have omitted. Or, as Montañez put it, “It’s almost as if we had a central nervous system without a brain.”
Etzioni warned, in any event, “Don’t mistake a clear view for a short distance.” What if a media release went out, saying, “We are going to release an artificial scientist!” Most people would sense a problem with the claim, he said. There is more to what science personnel do than a machine can handle, for example, in case of serious illness: “Do you really want the diagnosis delivered by a machine?” Indeed. One recent effort along those lines created disastrous publicity for a California hospital, quite apart from the grief it caused to the persons concerned.
As Robert J. Marks put it, Non-algorithmic things (things that cannot be calculated), “cannot be uploaded.” Human consciousness, little as we understand it, appears to be one of those non-algorithmic things.
The panel was held at COSM, A National Technology Summit: AI, Blockchain, Crypto, and Life After Google October 23–25, 2019 was sponsored by the Walter Bradley Center for Natural and Artificial Intelligence, hosted by technology futurist George Gilder. It featured Matt McIlwain (Moderator) — Managing Director, Madrona Venture Group; Dr. Robert Marks — Director, Bradley Center for Natural and Artificial Intelligence; Oren Etzioni — Ceo, Allen Institute for Artificial Intelligence; and Dr. George Montañez — Assistant Professor of Computer Science, Harvey Mudd College