Foresight Nanotech Institute Logo
Image of nano

Singularity, part 4

This the fourth essay in a series exploring if, when, and how the Singularity will happen, why (or why not) we should care, and what, if anything, we should do about it.

Part IV: When

So when is all this going to happen? To quote Mark Twain, I’m gratified to be able to answer that question immediately:

It depends.

In nanotech, I think that if there were a major, well-funded effort focused on getting to nanomachinery by trying lots of different pathways simultaneously — a full-court press — we might see some early limited lab prototypes in a decade; but there won’t be such an effort. Thus a better estimate would be 2030 or even 2040. (Counting from 1960, how soon would there have been a manned moon landing if there had been no Apollo project?)

In AI, the situation is perhaps brighter. The resources available to AI are likely to increase in the 20-teens as robotics and smart systems become more generally useful. As AIs start to cross the human range of capability, they will attract more investment and progress will accelerate. One key element in the takeoff is that as computers keep getting cheaper, a lot more people will be able to experiment.

On the other hand, AI has a disadvantage to nanotech. We know to some extent what the goal is in nanotech, and what an autogenous manufacturing system looks like. We’re working toward something we understand. But in AI, we don’t really know the corresponding key trick: how do you get a mind to extend itself? There are lots of ideas, including mine, but in a real sense the goal of AI is less well understood than that of nanotechnology.

Even so, what the goal would look like from the outside — an intelligent human — is perhaps more clearly seen than the nanotech one, and has been understood for 50 years. This has perhaps been at least as much a curse as a blessing. AI had its dark ages in the 80s, its renaissance in the 90s with robotics and machine learning theory, and is making pretty good progress.

I would guess — and this is blatantly a speculation, albeit a fairly well informed one, that the “secret trick” of AI will fall in the next decade. That means that the 20s will see robots not just as good as humans at specific, well-defined tasks, but able to learn new tasks the way humans do.

Please remember that AIs won’t necessarily be autonomous robots — most of them will be like having a secretary built into your computer, a phone answering system that acts like an intelligent receptionist, a self-driving car (although I imagine that in the 20s having a robot butler will be a status symbol for a while). Things (including all the software you interact with) will get smarter.

If my guesses are right, by 2030 we would be beginning to see some significant economic pressure from the AI sector. And the 30s will be interesting times.

One of the more interesting aspects might be nanotechnology. Having AI online might shorten the time from laboratory achievement of nanomachines to major real-world applications significantly. Imagine, for example, the amount of engineering necessary to make a Drexler space suit work properly and safely. AI engineers (and AI-enhanced human ones) would make a huge difference to the development time.

11 Responses to “Singularity, part 4”

  1. Says:

    If you have taught a task to one AI robot, how would you go about teaching a second AI robot?
    a. upload the learned skill from the first robot to the second
    Efficient, but may not be true AI.
    b. repeat the same teaching process with the second robot
    Should give consistent results, but would different AIs learn differently?
    c. get the first robot to teach the second
    This would be a true test of whether the first robot actually understood the task, or was just mimicking it teacher.

  2. Says:

    a robot smart enough to learn a human-type task might well be sufficiently complex to be quite non deterministic, so that each learning session would produce a different result, even on the same machine which raises a problem with (b). Also in your list, i think method (a) is a subcase of (c) anyway, so overall I think (c) is a nice idea !

  3. Says:

    In the post, Dr. Hall says:

    “In nanotech, I think that if there were a major, well-funded effort focused on getting to nanomachinery by trying lots of different pathways simultaneously — a full-court press — we might see some early limited lab prototypes in a decade; but there won’t be such an effort. Thus a better estimate would be 2030 or even 2040. (Counting from 1960, how soon would there have been a manned moon landing if there had been no Apollo project?)”

    I must disagree with him on this point. Government funded multi-year projects are generally
    required to achieve goals that are beyond the reach of private individuals or organizations.

    Things like the Manhattan Project, the Apollo Moon Missionr, and, more recently, ITER, the new fusion reactor are perfect examples of this class of activity. In each case, this goal is beyond the reach of even the largest multinational corporations due to the sheer expense. Only governments, more specifically, The United States, Europe via the EU and Japan (and possibly China) have the
    resources to throw at these kind of projects.

    Molecular Nanotechnology is quite different. Companies in this field can be started with a relatively
    small amount of capital. Zyvex, the company started by James von Ehr is a perfect example of this.

    I foresee true molecular manipulation being developed organically via a multitude of different
    companies and technologies all coming together at the right time, rather than needing a
    “Manhattan Project” style effort.

  4. JamesG Says:

    I disagree with the timeframe. The nanofactory collaboration (Freitas, Merkle, Zyvex, etc.) plans to have DMS by 2012. Now, think about that, because I really doubt it is going to take 18 – 28 years to get to nanofactories/nanorobotics from DMS. Probably more like 2 or so, imo. I think people are afraid of telling this, because of all the ridicule they get from ignorant people, but that’s how it is.

  5. Instapundit » Blog Archive » STILL MORE ON THE SINGULARITY, from J. Storrs Hall…. Says:

    [...] STILL MORE ON THE SINGULARITY, from J. Storrs Hall. [...]

  6. Says:

    The “trick” of AI was fully describd by Jeff Hawkins in “On Intelligence” like a decade ago. He’s making it a reality right now at Numenta.

  7. Says:

    This whole singularity thing is starting to take on the tone of discussions of the “End Times” and the “Second Coming.”

  8. JamesG Says:

    ‘This whole singularity thing is starting to take on the tone of discussions of the “End Times” and the “Second Coming.” ‘

    So? It’s just a coincidence that many things described in religion, will finally be possible with nanotechnology and AI. That does not mean nanotech and AI are religion, only an ignoramus would suggest that.

  9. Accelerating Future » How Long Before Superintelligence? Says:

    [...] Recently on Nanodot, Foresight Institute President J. Storrs Hall said: [...]

  10. Says:

    Understanding how AI would progress i guess would largedly depend on how do we understand how our I’s work. If a spacious frontier in our intelligence is still unknown, how could we succeed in AI’s in the real sense of the term?
    Let AI alone without I… is only A. What is not natural is not lasting and destructive!

  11. Says:

    ‘a. upload the learned skill from the first robot to the second
    Efficient, but may not be true AI.’
    I’m of the firm opinion that early AI will be industry/company owned and run. Since this is the most efficient way, it will be the way. Its also what we all (secretly or not so secretly) wish we were able to do.

    ‘If my guesses are right, by 2030 we would be beginning to see some significant economic pressure from the AI sector. And the 30s will be interesting times.’ This is what I want to read more of- also, the smart application of off the shelf tech that exists right now could put a preasure on workers the way Marshall Brain talks about it. Kiosks, kiosks everywhere. VOIP like the one Peter Voss is pushing… You don’t need anything approaching human level AI to gut most of the work force.

Leave a Reply