Artificial vs. Biological Intelligence In the AI field for 55 years, Kurzweil began at age 14 with MIT’s Marvin Minsky and followed up visiting with Frank Rosenblatt, the inventor of the “Perceptron,” the first artificial neural net. Like many figures in his field, Kurzweil has come to believe that artificial neurons are better than real ones, that silicon “intelligence” is better than carbon brains, and that a “singularity” is near in which computational general intelligence will reign in the universe. We will become dependent on AI prostheses in our minds and nervous systems. I believe that Kurzweil is making a fundamental mistake of categories. I have been studying the science of connectomes. Launched by Olaf Sporns and his team at Indiana University and popularized in the definitive Connectome by neuroscientist Sebastian Seung of MIT, connectomes originated in brain science. But computer and communications engineers will find this model familiar since it repeats in biology the connectivity schematics for both computers and networks of all sizes. In simple terms, a connectome is a detailed map of all the connections in a system. What I found is that to map the connectome of a single human brain requires a zettabyte or so (10 to the 21 bytes). Similarly, you can map the entire internet with a few zettabytes. But the brain functions on 12 to 14 watts of power, while the internet consumes gigawatts. After Shock climaxes with a series of fascinating master classes from the eponymous John M. Smart and his cohorts. He believes that for insights into the future “we should look inward, not outward.” Look to mind not matter. Look forward to a “destiny of density and dematerialization.” As Smart writes, “Each human connectome is the most advanced computational nanotech (physical inner space) and infotech (virtual inner space) that presently exists on Earth.” Smart doesn’t say it, but I conclude that human technology will prevail only when it amplifies this real intelligence. The effort to create computer minds that displace humans is both futile and perverse. Grab that wire of singularity and you will be shocked into submission. Your company will eventually fail. That was Peter Thiel’s conclusion in his masterpiece Zero to One. He is not in After Shock, but his exposition of the limits of computation provides a useful corrective to some of the assumptions of AI. After Shock presents an alternative to existing computers. From the queasy quagmires of quantum computing weaves in IBM Fellow and VP Jay Gambetta, the creator of Qiskit, who has some 90 publications in the field and over 10 thousand citations. While ordinary Turing machines can compute “n” states with “n” bits, quantum computers replace the binary bit with the essentially infinite “qubit.” “With a quantum computer of N qubits, one can now explore an internal quantum space of N^2… Once we are able to simulate molecules with absolute fidelity, we no longer have to make guesses or assumptions about how they will behave…” We’ll be able to model matter at will and invent new materials to spec. “Quantum computers are still in the early stages of development and far from totaling 100 ideal qubits.” Outside the pages of After Shock, Gambetta dismissed Google’s recent shocking claims of “quantum supremacy.” But Trump Administration Chief Technology Officer Mike Kratsios believes it signifies a critical edge against China. I say quantum computing is just another form of analog computation. It relegates the burden to input-output. But back to the future. The Prime Challenge of Technology In After Shock, David Brin writes: “The greatest trick of the enlightenment… has… been the deliberate dispersal of power into units small enough that they might compete and hold each other accountable, while preventing any cabal of cheaters to gain obligate power.” Apart from this almost exact evocation of blockchains, which is a futuristic technology somehow unknown to the futurists here, Brin is expounding the real rules of a prosperous and humane future. The more a technology empowers human minds the more important it is. Human minds are not aggregated in clouds or concentrated in data centers. They are dispersed around the globe. Contrary to Kurzweil, they are not fixed in capability. They are the source of all the capabilities of machines. Enabling them is the prime challenge of technology and the prime rule of successful investment. Regards, George Gilder Editor, Gilder's Daily Prophecy |
No comments:
Post a Comment