You have to be careful about pushing evolutionary metaphors. There's no such thing as a "fitness landscape" when you don't have species. In a world where technological devices are all one-offs, there are no species, hence no competition between species.

More precisely, there may be lots of copies of useful designs, but where any actual species exist and evolution goes on is solely inside the Edisonian AIs. And each new instance of a device can evolve from scratch to take into account its entire milieu. So evolution is potentially restarted again and again. This is not Darwinian selection; it's not even Lamarckian. But although it may require massive computing resources, it does not require the sort of AI that Kurzweil is talking about; and there is no reason to think that his kind of AI could compete with it (because, as Wolfram has shown, increases in the complexity of a system do not correlate with increases in the complexity of its output).

If evolution is only occurring within and regarding single instances, not between them, then there is no evolutionary arms race. So yes, a technological arms race can end.

Most products don't exist in a vacuum; they must interact with other products. The shape and size of cellphones has affected the design of everything from backpacks to the interior of cars. Do you think that all of these things will stabilize one day? As new products are invented, as new needs and desires arise, existing products will have to be redesigned. What was once an optimal (or, more likely, near-optimal) design will be seen to have obvious shortcomings. The fitness landscape will have changed, and a new design will have to be generated. This in turn will affect the fitness landscapes of other products.

• Re: Edisonian AI; or, the Technological Maximum by Karl on Friday May 12, @10:32AM

  Not sure in what way we are disagreeing. Why does it matter that a particular item isn't at a global maximum? Is the concept of a global maximum even relevant? When I say "optimal" I mean something like this: "How about this one?" "Naw. Don't like it." "This one?" "Better." "Okay, what about this one?" "Cool! I'll take it!" You seem to be talking about a future in which an optimal or perfect design for a generic fork is created and then after that all forks look the same. I'm talking about a future in which all forks are subtly different, though the differences may not always be visible, because each and every one was designed on the basis of a constellation of requirements including available resources, time etc. In this context, "redesign" is a meaningless term because every single object is a redesign. When I say that an object is optimal, I mean it is a local maximum for right here, right now. I'm not talking about nor interested in global optima because they don't exist in such a post-mass production environment.

  [ Reply to this ]

• Re: Edisonian AI; or, the Technological Maximum by Ted on Friday May 12, @04:19PM

  Okay, that's clearer; I misunderstood your original post. Just to make sure I'm understanding you now: you're describing a scenario in which each and every one of us has, in essence, a personal genie that can whip up whatever we desire? (Within the limits of conservation laws; a request for a personal jetpack that never runs out of fuel/reaction mass would presumably go unfilled.) And you believe we're "about ten years away from this point in our design and manufacturing capabilities"? And you believe this will be a stable situation, with no runaway arms races ("Genie, make mine better than his"; "Oh yeah? Genie, make mine better than his")?

  [ Reply to this ]

• Re: Edisonian AI; or, the Technological Maximum by Karl on Friday May 12, @05:31PM

  Well... no. I'm not talking about specifically what we will have in our pockets and driveways in ten years, twenty, or a hundred; basically I'm saying that we're entering a period that will be homogeneous in its overall conditions, so that we can speak of, say, the hunter-gatherer era of human history, the agricultural era, the industrial, and the technological maximum. Once we're at the maximum, one date looks pretty much like another in the same sense that a hunter-gatherer plucked from 12,000 B.C. could recognize an H/G society from 600 B.C. even though he didn't know the particular language or culture he'd been dropped into. We're already using evolutionary algorithms to create new designs, so we're entering that era now. There's no firm dividing line, never will be. This is not the singularity. Personal genie? No. Different paradigm of design/manufacturing? Yes. Look at it this way; the whole idea of a "technology" is so wrapped up with our preconceptions about how an industrial society is put together that we're not even aware of the assumptions we make when we invoke the word. So "technological singularity" is a concept that's coherent only from within a certain framework. From the standpoint of that framework, yeah sure you inevitably get runaway arms races. There are enabling technologies being developed now (and I use the T-word in this case deliberately) that are potential enablers of a different process model that is not industrial anymore, and in which conditions like runaway arms races aren't important. You can have them; they're just not change drivers for the overall system anymore. If you imagine that the world at the technological maximum works the way ours does, then what I'm saying will make no sense to you. But at least consider the idea that objections that come out of the current paradigm might sound like a 10th-century peasant having New York described to him and saying, "It'd never work. Where do they keep their sheep?" Clearly a novel is in order to properly describe this stuff.

  [ Reply to this ]

• Re: Edisonian AI; or, the Technological Maximum by Ted Chiang on Saturday May 13, @10:01PM

  I can readily believe that there will be a different paradigm of design and manufacturing in the relatively near future. I just have a hard time reconciling that with your statement "At this point, it becomes nonsense to talk about 'higher' technologies." You asked, "What's higher than an optimal solution?" You've indicated that your optimal solution falls short of a personal genie, so obviously there is something higher. Until personal genies become available, it is possible to speak coherently about higher technologies. (Because you mentioned Clarke's law, I'm using "genie" as shorthand for some technological marvel.) And it's hard to say what a personal genie would have to be able to do. In the same way that our current world has created desires in people that 10th-century peasants could not understand, I can imagine that someone in the future could want things that -- had I a personal genie at my beck and call right now -- it might never occur to me to request. The specifications for a personal genie will always increase. (Not that I think personal genies are in our future.)

  [ Reply to this ]

• Re: Edisonian AI; or, the Technological Maximum by Karl on Monday May 15, @07:43AM

  I think a lot of the confusion is coming from a conflation of two ideas that I've been guilty of here: design and fabrication. It will take a very long time for fabrication processes to catch up to current physical theory (unless we evolve those to...). Meanwhile, design can leapfrog far ahead of what we're capable of actually building, potentially into a realm where increases in processing power do not add any value. What I'm saying is that your personal genie will not be using design principles, or have deep knowledge, that's very different from early-stage Edisonian AIs. In its design skills it'll be more advanced than early systems in the same way that modern internal-combustion engines are more advanced than a model-T's--however much its fabrication skills may have increased. Unless new physical principles are discovered, it's not unreasonable to imagine that an AI two hundred years from now might not be able to design a given device any better than an AI twenty years from now. At some point we reach a stage where the limitation is not in the intelligence of the AI but in the physical possibilities of what can be realized. I am suggesting that barring new science, this point can be nailed down fairly concretely. What this whole discussion reveals is that there is a subtext to people's understanding of technology--i.e. a superstitious belief that it has no limitations. If it has no limitations then it is a supernatural power. If intelligence (design) has no limits then it too is a supernatural power. This strikes me as unlikely; and if there are limitations to what is technologically possible, what would such limits look like? It is not too early in history to ask this question. A technological maximum doesn't preclude any of the products of technology that are so dear to SF writers and readers, or fans of the singularity and posthumanism--except for the sort of Matrioshka brain world where everything's made of smart matter that spends all its time thinking real hard about something or other. And I disagree that the specifications for a personal genie will always increase. Specifications for required items may always change, but that's totally in line with what I've been saying. "Always increase" sounds suspiciously like another supernaturally limitless capability. I can imagine that someone in the future could want things that they can never have.

  [ Reply to this ]

• Re: Edisonian AI; or, the Technological Maximum by Ted Chiang on Monday May 15, @04:14PM

  Fair enough. I agree that there are limitations to technology. However, I doubt that there are limitations to human desire. There will always be people asking for more, whether or not they can get it.

  [ Reply to this ]

• Re: Edisonian AI; or, the Technological Maximum by Karl on Tuesday May 16, @07:13AM

  Well, yes. As I said in the original posting, "The critical question for the future will not be 'what's possible', but 'what do you want?' and do you want it more than others competing to use the same resources?"

  [ Reply to this ]