The original singularity paper | Hacker News Reader

The original singularity paper

mindstalk.net

35 pointsanirudh17 comments

Computer/human interfaces may become so intimate that users may reasonably be considered superhumanly intelligent.

In a limited way, this has already happened. Imagine talking to a human of average intelligence with the best education we could provide now sans all knowledge of and reference to the computers and the internet. S/he will be amazed by your seeming omniscience, even if the only site you had access to was wikipedia.

From the human point of view this change will be a throwing away of all the previous rules, perhaps in the blink of an eye, an exponential runaway beyond any hope of control. Developments that before were thought might only happen in "a million years" (if ever) will likely happen in the next century.

We are already seeing this happen because of the use of computers. Computers allow us to violate the economic laws and collapse the financial system. Computers have been the cause of more than one financial collapse in the past.

I have read words like, "AIG didn't factor in the probability of these things happening, because they were so remote." Isn't that the same as "happen in 'a million years'?"

Computers are allowing us to do things today, that we thought only decades ago would only be possible in a million years -- or never. Many thought we'd never get to the moon. We probably wouldn't have without computers.

"We tend to overestimate the effect of a technology in the short run and underestimate the effect in the long run." - Roy Amara

Computers allow us to violate the economic laws and collapse the financial system.

computers are fast complex calculators. they magnify stupidity, they dont create it.

The first three possibilities depend in large part on improvements in computer hardware. Progress in computer hardware has followed an amazingly steady curve in the last few decades [17]. Based largely on this trend, I believe that the creation of greater than human intelligence will occur during the next thirty years.

Yes, well, single-core clock speeds are a dead end now, and it's not at all clear how to take advantage of multiple cores in the same way. Why are we still excited about this?

Google would like a word with you.

The human mind is highly parallel. Anything that approaches it in intelligence will also be highly parallel. Even if clock speeds were a million times faster, parallel would be the only way to manage the complexity. The clock speed of a neuron is only 5ms and we do pretty well.

The "clock speed" of a neuron is actually closer to 50ms-100ms (only about 10-20Hz for meaningful purposes) -- but you have about 100 billion of them, and they're all working in parallel. The brain's building blocks are really slow relative to silicon; action potentials move at speeds measured in m/s.

That's parallelism at the "micro" level. Compare it to the parallelism in superscalar or EPIC processors (incidentally, EPIC is also hard to take advantage of). This argument doesn't show that many-core architecture is necessarily going to help.

If you can put enough processors and memory close enough, it seems as though you should be able to model a few hundred or thousand neurons on each processor, to assume the worst case. We've already moved past the need for neuron-level parallelism, so from here on the question is what order of magnitude of neurons can be modeled on a single processor.

(Again, assuming the worst case that we have to model neurons to get intelligent systems before we can figure out exactly how to produce intelligence in a less simulation-oriented manner).

"The human mind is highly parallel. Anything that approaches it in intelligence will also be highly parallel."

Birds fly by flapping their wings. Any man-made object that approaches the flying ability of a bird will also flap its wings.

That analogy fails, because man-made objects have not yet approached the flying ability of birds (or insects). Man-made flying objects are highly energy inefficient and slow and clumsy as a result. We cannot make something that flies like a bird or something that flies like an insect.

What are you suggesting - that Intel's Pentium D, Core Duo, Core 2 Duo, Core 2 Quadro and Core i7 are all exactly the same computing power as a similarly clocked Pentium 4, just with more cores?

Clock speeds in current aircooled personal CPUs are sitting where they were several years ago, but branch prediction, pipelining, caching, size reduction, dedicated MIMO/SSIMD units, integrated memory management units, and I don't know what else have all resulted in the ability to compute more in the same time (i.e useful speedup).

Why are we still excited about this?

Why not? It's exciting.

What if we are culturally programmed to refuse singularity?

The key in the article is that constructing a higher intelligence will enable us to get to the next level faster - by using our new found capacities. But all we see around us makes it pretty obvious this is not the case. Whatever means we have of improving ourselves, they're woefully underused. We actually have ways to make intellectual work better. Not science fiction, but simply a trip to the pharmacy. They work (there have been several articles on HN about provigil or ritalin), and they can be made safe - they mostly are, all we need are some proper guidelines. But what do we do? We talk about outlawing them. And they are definitely not publicly acknowledged.

And before you say that this is not the same: how exactly is it not? Can you imagine a way to improve a human being which would be socially acceptable? Not curing handicaps, but making better for the sake of making better. Just imagine the public reaction if a company offered implantable headphones. Cutting a human for no medical purpose is social suicide.

Now of course this can not stop progress forever. But it can make it a good deal slower.

Now I see what Steve Yegge was talking about in his sci-fi story...

http://steve-yegge.blogspot.com/2009/05/programmers-view-of-...

The original anti-singularity book: http://en.wikipedia.org/wiki/G%C3%B6del,_Escher,_Bach

Actually this book is not anti-singularity. It defends the idea that intelligence and self-consciousness can be attained by a machine, but says that we're still quite far from it (more than 1993+30 as the paper predicts).

Reply on news.ycombinator.com