Path: utzoo!utgpu!watmath!clyde!att!osu-cis!tut.cis.ohio-state.edu!rutgers!mit-eddie!uw-beaver!cornell!rochester!pt.cs.cmu.edu!andrew.cmu.edu!ap1i+ From: ap1i+@andrew.cmu.edu (Andrew C. Plotkin) Newsgroups: comp.ai Subject: Re: Artificial Intelligence and Intelligence Message-ID:Date: 7 Dec 88 18:00:32 GMT References: <484@soleil.UUCP> <1654@hp-sdd.HP.COM> <1908@crete.cs.glasgow.ac.uk> <1791@cadre.dsl.PITTSBURGH.EDU> <1918@crete.cs.glasgow.ac.uk> <44150@yale-celray.yale.UUCP> <281@esosun.UUCP> <283@esosun.UUCP> / > ... if you had a computer console in front of you with a button / > marked "pain" (any human simulator had better have some sort of sensory / > input), would you consider it okay to push it? / Yes, if I was testing the computer's pain circuits. When a computer / is in pain (i.e. a circuit board is burning out, or a cable is being / cut), I want to be sure that it can sense its distress and accurately / report its state of well-being. Ah, but I'm not talking about a system that senses damage to the computer. I'm talking about something that applies stimuli to the simulated pain inputs of the simulated human. You brought up "computers being able to simulate humans," and I'm using that concept. To clarify it, let me describe it as a program running on a computer; with input routines that feed data to the same thought-mechanisms that human sensory nerves feed to in the human mind; with output routines that take data from the appropriate thought-mechanisms and display it in suitable form. Given any input, it will produce output as a typical human would. (Passing the Turing test, therefore.) (The "easiest" way to this is to create a trillion CPU's, each capable of simulating one neuron, and hooking them together. Sensory input could then be pushed into the "sensory neurons" directly. However, the exact mechanism is not relevant here.) Now, there's a big difference between damage to the computer and simulated pain. One degrades the performance of the simulation; the other makes the simulation yell "ouch!" (assuming it's a good simulation.) Obvious example: if a brain surgeon is working on a conscious patient, the patient feels no pain (assuming the cut scalp has been numbed.) The surgeon can poke around, feed minute electrical currents in, and so forth; the patient will see strange flashes, have odd bits of memory pop up, and so forth. If the surgeon drops his scalpel in, the patient will stop thinking or suffer functional loss, but no pain is involved, unless sensory centers are hit. / Similarly, if I put the machine in / emotional pain (by giving it a program that runs forever and does / no useful work), I hope the machine can diagnose the problem and / gracefully aprise me of my error. Keep thinking human simulation. The machine would simulate reactions like "Damn, this is boring." Or, more likely, "Why should I do this idiot work? Program it into a computer!" (Of course, if it was a simulation of a reasonably open-minded human, you could easily convince it that it was really a computer. That its optical inputs are coming from cameras would be a giveaway. But I doubt it was settle down and execute C for the rest of its existence. Assume it was a simulation of you -- would you?) --Z