Hacker News Reader

Comment by thesash | Hacker News Reader

Every time I hear about the inevitability of the singularity, all I can think about is what a precariously balanced collection of hacks actual computer programs are at their core, and how far off that is from true intelligence. Am I just pessimistic or are we a complete paradigm shift or two away of even getting close to true AI?

View on HN

I share your pessimism. I think attempts to make a mind in code are a modern attempt at alchemy, and just as doomed.

My sticking point is consciousness. I don't think it's an illusion; I think it's both real and essential to intelligence. And we have no idea what it's made out of. We can break it. We can characterize the hardware conditions under which it works. But build it? No, that's completely beyond us, and I think it will be for many centuries. We don't even know where to properly start.

I had a pet rat, once. I think it was conscious, in a very small way. It visibly related itself to the world. When they can convincingly simulate that level of alertness and presence, when they can build a machine that appears at least as conscious as my pet rat did -- when they produce the tiniest amount of genuine gold -- I'll concede the singularity may be on the horizon. Until then, I'm a skeptic.

My thoughts exactly-- if we don't understand how organic consciousness works, how can we hope to replicate it artificially?

I think we are close to having the software engineering ability to create the infrastructure needed to support a self-improving optimization process. The feats of software engineering in the last decade or so seem like strong evidence that we can develop the tools to support mathematical models and methods at a large scale, when we really understand the problem.

What I don't think we are close to understanding is how to actually make practical self-improving optimization processes and especially ones that are actually beneficial, rather than detrimental, for humanity. I feel like we are just getting started on this. If progress is going to be made at the rate that Kurzweil has predicted it is going to take more than just faster processors and better supporting software infrastructure.

Thats my point though-- it seems like there are too many unknown unknowns here to make any sort of predictions. We're still several massive leaps removed from where we would need to be in order to really get started.

I would argue that evolution overall often looks like a "precariously balanced collection of hacks". So, it would be irrational to expect evolution's solution to the problem of intelligence to be any different.

That might be true if computer programs were subject to the survival of the fittest, which, in my experience, they are not.

Evolution may seem like a "precariously balanced collection of hacks" but it takes a long time. As an analogy, the singularity is so far off because our current computing power is essentially invertebrate multi-celled organisms. Computers still need a backbone, get out of the oceans, and breathe air - and this is long before they can store a human existence.

Kurzweil seems to think it's close out of hope. I think he's just insanely afraid of dying.

Reply on news.ycombinator.com