Comment by bdr | Hacker News Reader

Comment by bdr | Hacker News Reader

When "The Singularity Is Near" was first published, I asked Bill Gates what he thought of the book. He said "I don't know how near the singularity is, but I haven't heard any convincing argument why it won't happen". That seems like a more sensible position than the one offered by Allen in this piece. There is indeed a lot of uncertainty in the rate of progress. But Allen contorts that fact to say that the singularity isn't near. He offers three basic critiques:

First, he points out that Kurzweil is extrapolating, and extrapolations can be wrong. This seems obvious. And it doesn't do much towards Paul's "isn't".

Second, he says that software and hardware will have to keep improving, and that might not happen. Again, this seems intellectually equivalent to saying "you might be wrong". No evidence is provided that progress will slow.

Third, he says that the singularity will require either a bottom-up biologically inspired model, or a non-biologically inspired "AI" system.

The complexity of the former will take a long time to overcome. Since Allen concedes that sufficient computational power is already here, he seems to be arguing that it will take over a hundred years for us to have a detailed model of the human brain. Looking at the progress of science, this seems terribly conservative, yet little justification is offered. Allen's posited "complexity brake" seems positioned tendentiously — why are we going to start hitting it now, instead of fifty years ago?

As for the AI route, he writes: "But when we step back, we can see that overall AI-based capabilities haven’t been exponentially increasing either, at least when measured against the creation of a fully general human intelligence." What does it even mean to be exponential relative to a binary variable? He is right to say that current methods are limited, and don't achieve generalized intelligence, but offers no reason why that won't happen. His argument seems equivalent to simply saying it hasn't happened yet.

This piece offers little intellectual contribution, and the only reason we're reading it piece is because of the author. It reads like he started from the conclusion he wanted, and threw some paragraphs in that direction. His argument, essentially, is that he's betting against continued progress in science, software, and hardware. That just seems crazy.

"The complexity of [a bottom-up, biologically-inspired model of the brain] will take a long time to overcome. Since Allen concedes that sufficient computational power is already here, he seems to be arguing that it will take over a hundred years for us to have a detailed model of the human brain."

Given that Paul Allen donated $100 million in 2003 (and pledged an additional $300 million in 2012) to the Allen Brain Atlas[1] and Allen Institute for Brain Science[2], he probably knows more than most people about this. He certainly isn't betting against continued progress.

[1] - http://en.wikipedia.org/wiki/Allen_Brain_Atlas

[2] - http://en.wikipedia.org/wiki/Allen_Institute_for_Brain_Scien...

Yet throwing money at sth does not make him an expert. Neuroscience is growing by leaps and bounds, ask any neuroscientist you know. It is indeed peculiar that he has a pessimistic view of it, but i would bet the best neuroscientists do not share that view.

You might want to check in with some neuroscientists. If you take offense at personal attacks at Kurzweil's sanity and intelligence you might want to ... prepare yourself. Imagine if a retired doctor started doing speaking tours about the future of the internet and computing and completely disagreed with the general consensus on HN about what could plausibly happen. Imagine what the comments about him on HN would be.

They would view Allen's response as "not delusional" rather than pessimistic. I'm often surprised by how widespread singularity fans think belief in it is.

Outside the tech industry, where people have spend their entire lives in a field that has seen no increasing exponential returns (other than problems solved purely by computation, which they consider a wonderful thing but often not a fundamental advancement to their field) there is very very little belief that "the singularity is near" or that kurzweils utopian interpretation of the singularity is correct.

Neuroscience, as a science, might be growing by leaps and bounds but if you ask the average neuroscientist how much the field currently knows about how the brain works I think the answer will be a great deal more modest. It's very early days still I think. Not to mention the fact that nobody understands how and why consciousness works, which seems to me a key feature in eclipsing human intelligence with AI.

Job prospects for neuroscientists are completely bleak. Which is a weird dynamic of "need" vs "supplied" (or, will pay/fund). It's rather sad.

( I don't believe Allen ever actually conceded the point you seem to wish to attribute to him, even after multiple readings. )

There's every reason to believe that things aren't going to get faster or better--CPU development seems to be petering out compared to the gains of yesteryear. Worse, the products that drive those sales, and the markets that drive those products, seem utterly uninterested in anything beyond simple apps and hardware-as-appliance.

The thing a lot of people seem to forget is that the market is what drives this stuff, and that in the absence of market forces not very much happens. This is the world we live in.

CPU development seems to be petering out compared to the gains of yesteryear

This is arguably true if your primary metric is single-threaded performance on branchy-integer-type code. Admittedly this is what normal users see most of the time, which is why consumer desktops and laptops haven't been very exciting the last few years. But the cost to build a petaflop supercomputer keeps dropping.

Sure, but the cost to solve a petaflop problem is not. We hit a point where the bottleneck is no longer flops but being able to organize systems that can use them effectively.

Doesn't matter how fast you think we will overcome this problem, it illustrates to many people that the intersection between the world that is not increasing exponentially and chip speeds starts to matter a great deal when you want to actually solve a problem by moving floating point numbers about.

I think the biggest test of these sorts of processing predictions will come with our transition away from traditional transistor tech. Whether we switch to circuits on graphene, replace transistors with memristors, or adopt quantum computing doesn't matter, according to singularity proponents. Our ability to process information should continue to exponentially improve. I believe, as you said, that we are due for such a switch the 5-10 year timescale because current CPU development is reaching the end of the line. Whether or not we can seamlessly switch to a new way of computing will make or break many of these singularity predictions.

Just curious, in what context did you ask Bill Gates that?

Allen's mistake regarding AI is a failure to understand that it moves exponentially by leaps, and by gradual progression that enables the big leaps.

As an example, what comes out of Siri & similar AI programs over the next ten years will ultimately be viewed as a leap, driven by mass consumer demand. But Siri only exists due to thousands of small progressions that might have seemed trivial by themselves.

Reply on news.ycombinator.com