Wednesday, June 29, 2011

Morning Singularity Watch


A few notes on the evolving progress toward a singularity of some kind. I've seen automated driving as a next step in that direction which will play out over the next decade or two.  A few items this morning tracking the trend.

First up, Volkswagen has a "production-like" research prototype for a "Temporary Auto Pilot" that can take over driving at the convenience of the driver (eg in stop and go traffic, while distracted etc).  This was done as part of a major EU research initiative spending 28 million euros on developing automated driving.

Secondly, The NYT Wheels blog reports that the state of Nevada assembly has passed a bill authorizing the transportation department to draft regulations governing automated driving on the road.

So the automated driving thing is coming along about as one might expect, and the research-industrial complex is driving hard towards this goal, so to speak

On another front, Megan McArdle posted this fascinating graph, which shows the number of new pharmaceutical molecules successfully bought to market per billion of R&D spend:


As you can see, productivity in this area has been dropping very fast - it's getting harder and harder to come up with worthwhile new drugs.  It looks to me like the spend per molecule increases by a factor of ten about every thirty years - about 8% per year.  So that's much faster than just salary growth - most of it is dropping productivity.

This makes a similar point to what I made in Moore's Law vs the Flynn effect.  Apologists for proceeding as rapidly as possible to a singularity like to claim that there's nothing to worry about because we'll use all this fantastic AI to integrate with and augment human intelligence and make being human more and more fun and fantastic.  But whenever you look at actual trends on making humans better/healthier/smarter etc, you see very modest progress and/or diminishing returns, while the progress of the machines is much faster.  To me, that suggests the main symptom of the approach to the singularity will be to render a larger and larger fraction of the human population unemployable.  And that's been going on for a few decades now:


Finally, yesterday, Jamais Cascio had a very weak argument for why there's nothing to worry about:
Our technologies are not going to rob us (or relieve us) of our humanity. Our technologies are part of what makes us human, and are the clear expression of our uniquely human minds. They both manifest and enable human culture; we co-evolve with them, and have done so for hundreds of thousands of years. The technologies of the future will make us neither inhuman nor posthuman, no matter how much they change our sense of place and identity.
and
Technology is part of who we are. What both critics and cheerleaders of technological evolution miss is something both subtle and important: our technologies will, as they always have, make us who we are—make us human. The definition of Human is no more fixed by our ancestors’ first use of tools, than it is by using a mouse to control a computer. What it means to be Human is flexible, and we change it every day by changing our technology. And it is this, more than the demands for abandonment or the invocations of a secular nirvana, that will give us enormous challenges in the years to come.
Essentially, his argument comes down to saying that since nothing really terrible has happened due to our use of technology yet, nothing really terrible can happen in the future either.  The Sumerians might beg to disagree, if only their civilization hadn't collapsed from salting their fields with new-fangled irrigation technology.  More importantly, creating intelligence that can duplicate more and more of human's mental capabilities is fundamentally different than all other prior technological progress.  Why?  Because, once that is accomplished, we have nothing left to offer the economy in the way of productivity.  It won't need us.  I have no idea what will happen as a result, but I wonder if we'll just go crazy before we get there.

32 comments:

Emil said...

I recommend Kurzweil's book The Singularity, which I think is very interesting even if very bold.

Personally, I see a struggle between Peak Oil or the Singularity.

Either way, both events are going to change the world in profound ways, and the days of 10 % unemployment(real unemployment, not government creative accounting) will seem like a fantasy.

Longterm, we will either live like we did in an earlier age or we will merge with machines. We've already started(pacemakers, different machines in our ears etc). Nanotechnology is also a huge potential.

I think in the future biological viruses may become obsolete. What's taking down your PC today may take you down tomorrow.

Stuart Staniford said...

Emil: I think you mean "The Singularity is Near" and yes, I read it a few years back. Most depressing book I ever read by a large margin. I had not thought through those issues and the book forced me to, and I still hate the conclusions I came to.

The Caretaker said...

Great post Stuart. In the same vein of thought, I see the combination of peak oil and automation greatly increasing inequality across society. The stats show this has already happened over the last twenty years, as the benefits of productivity increases have gone to the elite while wages stagnate. The future is heading towards an economic system where the globalized elite reap larger and larger productivity gains from automation, while also being the only class able to afford oil and the easy transportation lifestyle that is seen as normal today. They use the funds to buy up "distressed properties", as they are already doing, and increasingly extract their wealth from the rents of the masses. Meanwhile, the middle class will be competing with machines that can do our job faster and cheaper, and spending more on food and energy, leaving little else in the way of discretionary spending. The gap in market power increasingly leads to elite consolidation of the political system....heck, we are basically already there. The gap will grow until....well something will have to happen and it won't be pretty. Will the middle class institute redistributive economic regimes, or be coerced into fighting the middle class of the neighboring country for the scraps? time will tell.

Stuart Staniford said...

Caretaker: Yes, that's exactly my concern.

Combine that with the increasing ability of the automated infrastructure to keep tabs on anyone, and the serious eroding of legal frameworks for managing it (think of the explosion of drone killing campaigns across the planet). At some point things will come to a head, though it's probably not imminent.

So the task for now is: what kind of institutions and cultural shifts would allow us to manage this in a positive way, rather than a disastrous way. Ie, what should be the agenda of a rational reform movement, to counter the inevitable revolutionary movements?

neil craig said...

Emil if its a race with peak oil I think the recent discoveries of how to get hydrocarbons from shale, which more than double reserves works to suggest we will make it.

However one thing about a ohysics singularity is that you can never recognise it while you are inside it. I do not think we could voluntarily go back even to the 1950s when millions died of polio.

I grant that progress is going to make employment more difficult to maintain but much of that is people being able to choose not to work.

"In the swaet of thy brow shalt though eat bread" isn't that good an offer.

Stuart Staniford said...

Neil:

Wrt "much of that is people being able to choose not to work", this doesn't seem to be true in a very positive sense. Most of the loss in employment population ratio is amongst people with less education, and a lot of them have gone on disability to get by. I don't think US culture is going to adapt well to larger and larger numbers of uneducated people on disability (or any other form of welfare). See, for e

neil craig said...

Indeed it will have to adapt to it and adaption isn't easy. But better to adapt to anybody being able to have a better standrd of living, wvwn without working, than most workers had in the 1950s than us having to adapt to working hard for the 1950s standard.

Every action has positive and negative effects. If the positive ones are better you adapt to the negative ones, or better yet take further actions which reduce or eliminate the bad effects.

Johannes said...

As so often, excellent post Stuart - thank you - and excellent follow-up from The Caretaker. As for enabling the indubitable benefits of technological advances to become more widespread, I believe an increase in international cooperation is of utmost and urgent importance: what we need to aim for is much more of (democratically elected) global government.

While this currently sounds like pie-in-the-sky, with the increasingly uneven distribution of the rewards of work - be those based on merits or not - there must be redistributive economic mechanisms for the economic survival both of individuals and of consumer capitalism (billionaires don't consume enough as a percentage of wealth). Such mechanisms have, however, to a large extent become impossible to maintain with the increased international mobility of capital and workers. How keep collecting income taxes, inheritance taxes, etc from those benefiting the most in today's economy when small sunny jurisdictions can offer tax free havens? How maintain free-at-point-of-use universities if the graduates move to a different state/nation, with lower taxes and/or better earnings possibilities?

An international taxation authority might do just that, taking say 20% of global incomes above a certain yearly sum (perhaps 10,000 USD on purchasing-power-parity basis) and handing the sums out as a supplementary income to local governments all over the world, under firm central - democratically controlled - audit, and awarded on a per capita basis (again, with purchasing-power-parity adjustments).

Well - given today's animosities, international and national - it is difficult to convince people about any international cooperation, but I do think we need to realise that the immense problems we are facing can't be dealt with within the current governance framework.

Burk Braun said...

I am not sure what point you were making with the FDA section.. humans are not yet being directly supplemented with implanted chips, etc., so while education may or may not be helping us, this has little to do with the supposed singularity, which has an entirely different mechanism, supposedly/prospectively.

Anyhow, the NME issue is one of biology- there are just so many practical targets to make drugs against, and the easy way of mining natural products for drugs and drug ideas is pretty much played out- there are only so many medicinal plants out there.

The NME issue thus has nothing to do with human intelligence/ knowledge, which in this field continues to grow at a rapid clip, or with moore's law, etc. It might be likened more to the geological constraints of peak oil!

Stuart Staniford said...

Neil:

I think you are making very light of things that have been absolute cultural imperatives in the western world for thousands of years. The idea that hard work is virtuous is drummed into most of us at a very young age. The protestant work ethic is almost as old as protestantism, and Aesop's fable of the ant and the grasshopper illustrates that teaching the same has been a need for a millenia, not centuries. More recently, we had the welfare reforms of the 1990s which were all about making sure people work if at all possible (even single mothers) to get welfare. Right now, US politicians are busy searching for ways to cut spending to reduce the deficit. The chances of a US politician getting elected on a platform of giving away tax money to less-educated people who can't find a job, well, let's just say it's zero to six decimal places and will be for a long time to come.

So, if the government won't fund welfare for such folks, then what?

Stuart Staniford said...

Burk:

My point is that Moore's law is not hypothetical. It's been going on for decades, there's no sign of it slowing down whatsoever, and we can point to large numbers of occupations that no longer exist on account of it (and the associated developments in software), and we can readily foresee more occupations (like professional drivers) in the immediate path of the steamroller. And there in the employment statistics, we can also see that less educated folks are increasingly not employed. It's all happening right in front of us, and seems very likely to continue to play out in coming decades.

By contrast, the supposed benefits of human augmentation are much more hypothetical (outside of a few specialized niches like cochlear). No useful cognitive implants are available to the general human at present, and so we have no real idea how difficult it will be to get such a thing to work and bring it to market, or how fast it will then progress.

But what we *do* see when we look at every statistic of human progress (life expectancy, IQ, beneficial drugs, etc) is that progress is very slow, and usually decelerating, not accelerating. We are spending ever more on healthcare to gain less and less marginal benefit.

So, to think that the singularity is going to be great fun for the average human, you have to assume that all those decelerating health/human progress indicators are going to turn around and catch up to the computer ones (Moore's law) based on no current data whatsoever. It's a pure exercise in wishful thinking.

WwoofBum said...

Seems to me that Cascio's argument is little more than a tautology: we are human, and, since we are doing it, whatever we do, it is human because we are doing it. Holds equally well for killing ourselves off by making the planet uninhabitable for humans. The only difference being that, having killed ourselves off, there would be no humans around to say how human it was to do so.

Stuart Staniford said...

WwoofBum: ROTFL

The Caretaker said...

Some excellent discussion here.

Stuart: "Ie, what should be the agenda of a rational reform movement, to counter the inevitable revolutionary movements?"

Man, I wish I knew the answer. I work for local govt and am tinkering with how to catalyze this reform movement. I think we have to start with some basics to wake people up: my favorite is a show a chart that oil production has plateau-ed since 2005. That way you aren't speculating about the future, you are merely describing recent trends. Then a foray into "ecological economics" perhaps? I have become a big Soddy fan.

Johannes: "Such mechanisms have, however, to a large extent become impossible to maintain with the increased international mobility of capital and workers."

Exactly, and this is the other big problem beyond peak oil, and automation: the boundaries of the corporation have globalized while governments have not. That gives the corporation a lot of leverage.

This is another one I don't know how to solve. There is no way in hell a one-world democracy would be able to function effectively, yet that is what is needed. Maybe its just a series of agreements that we will tax our rich people if you will tax yours, and everyone promises not to cheat.

Michael R said...

"It's been going on for decades, there's no sign of it slowing down whatsoever"

Very much beg to differ, Stuart.

"Moore's Law" can be interpreted one of two ways. First, the literal doubling of circuit components every two years; second the real-world performance gains realised due to such doubling.

The latter, the real-world gains, have very much slowed down, as thermal issues have brought clock-frequency gains to a dead halt for several years now. Processors now incorporate more parallel computation units with their expanding transistor budget, but here Moore's Law slams right up against Amdahl's Law. Further progress in processing power is now measured in computations per milliwatt-hour.

For the former, the doubling of circuit density, the end is approaching rapidly. EUV photolithography has been stalled for years. Progress beyond the 11nm node is very much in doubt. We can reasonably expect the end of Moore's Law in this decade, and perhaps as soon as 2015.

buck smith said...

That is interesting chart but isn't what is really important if the total number of effective uses for the discovered compounds. Ulcers were treated not cured by expensive drugs until one guy proved you could cure them with them with already existing antibiotics. The number of drugs remained the same, efficacy went up costs went way down.

Valdemar said...

FYI: http://www.antipope.org/charlie/blog-static/2011/06/reality-check-1.html

Stuart Staniford said...

Michael R:

From my perspective as a guy who designs fast statistical algorithms for a living, all the clock speed limitation meant was that I had to get better at designing parallel code, since now Moore's law gives me more cores, instead of faster cores. It's turned out not to be *that* big a deal.

And in terms of delivered performance of servers, there really is no slowing.

Michael R said...

Granted, for certain economically interesting classes of computational problem, "slamming up against Amdahl's Law" is more of a gentle easing into Amdahl's law.

But in the end, Moore's "Law" is only an industrial heuristic, whereas Amdahl's Law is a mathematical equation.

I expect we will see, and relatively soon, that in retrospect the loosely-interpreted Moore's Law (i.e. inclusive of algorithmic improvements and balance-of-system improvements) looks rather more like this:

http://en.wikipedia.org/wiki/Logistic_function

Greg said...

@Valdemar: Stuart's singularity is an economic one, as outlined by The Caretaker above, not the LSD-trip artificial-intelligence-cum-mind-uploading that Charlie rightly dismisses.

@Michael R: You're probably right about Moore's law as originally stated. Once components get to be only a few atoms across in their smallest dimension, it's hard to see where further improvement will come from. And you may well be right about a system-wide slowdown and eventual halt.

But it seems to me that arguments about Moore's law vs. Amdahl's law are academic at this point.

Various historians have pointed out that it takes a long time, several decades, to properly exploit a new fundamental industrial technology. During this time the new technology's cost drops steeply, and its thermodynamic efficiency improves markedly.

VLSI silicon technology is good enough NOW for the economic singularity to occur. It's just a matter of developing new industrial processes, and changing mindsets and laws, to take advantage of it. And while that goes on, the case for automation versus labour is going to get more and more compelling.

So even if Moore's heuristic fails tomorrow, the trajectory of events in the real world won't be affected. On that topic, Jacques Attali's predictions in "A Brief History of the Future", which I originally rejected as outré, are starting to look more and more plausible.

dr2chase said...

@MichaelR - I see an end to Moore's law, but I'm not quite sure when it will appear. There is some hope that hardware designers will pull yet another fab-tech rabbit out of a hat (graphene? diamond substrates? quantum computing?). Other people are playing with special-purpose hardware for interesting problems. Others are hard at work trying to figure out how to make it easier to express parallel solutions to problems -- besides the performance issues, there are programming issues. Perhaps we decide that instead of slowly computed certain answers, we will be satisfied with rapidly computed probable answers; certainly in theoretical terms, RP kicks the ass of P (RP = problems "solvable" with randomization in polynomial time -- either with an answer that is probably right, or with an answer that is certainly right, probably delivered quickly, depending on detail).

KLR said...

Wish the automated driving thing would arrive faster - bought an extra bright/annoying tailight for my bike yesterday, in the hopes of distracting the proverbial giggling teenager reading a text while driving. Wonder what the stats are like for accidents due to IM?

sharon said...

"So, if the government won't fund welfare for such folks, then what?"

There's always Jonathan Swift's solution.

So there are really people who believe that hundreds of thousands of Americans who had paying jobs just a few years ago, all up and decided how more fun it would be to not work and just collect government benefits? Why am I not convinced?

Tim Auld said...

"if its a race with peak oil I think the recent discoveries of how to get hydrocarbons from shale, which more than double reserves works to suggest we will make it. "

I'm not sure if you are pro singularity, but this comment boggles the mind. If we ever do reach the singularity, the first thing the AI is bound to say is "You traded your aquifer for me? Idiots!"

neil craig said...

Why would an inteligent computer consider leaving small bits of Earth's geology thousands of feet underground unaltered more important than its own existence?

Moore's law is speeding up and while the solutions dr2chase gives may not all work it is clear that the unforeaeable ways it will conitnue are less unforeseeable than the were in 1950. Of course nothing can continue for ever but the fact that the rate is increasing fits better with us being at the foot than the knee of the S curve.

Tim Auld said...

I didn't suggest it was important to the computer's existence, but to the humans who created it. We're still going to need to produce food, and we don't need a super intelligent computer to tell us that permanently polluting earth and water is super stupid.

You can fantasise about all the solutions you can't imagine, but it boils down to one big gamble with unprecedented stakes.

neil craig said...

How does the presence or otherwise of bubbles of previously inaccessible shale gas thousands of feet underground impinge on the ability of plants across the world's surface to grow?

Well apart from the fact that if we burn the gas and add CO2 to the air the crops will grow faster, which does not impinge negatively on food supply.

Tim Auld said...

To release the methane they pump a cocktail of toxic and carcinogenic chemicals into the well. The process also releases heavy metals and radioactive material. This can be distributed through the aquifer and taken up by wells used for irrigation and livestock water or migrate through the ground and into above ground water systems. 20-40% of the huge amounts of water pumped into each well comes out contaminated and it is dumped or left open to evaporate, polluting the air or leaking. The condensation tanks release further pollutants which precipitate into the soil or are inhaled by people and livestock. The methane bubbling through the ground is flammable so wells and faucets can explode. Significant methane in the soil negatively impacts plant health. Leaked methane is also identified as a greenhouse gas 21 times more potent than carbon dioxide which may contribute to climate change unfavourable for farming. Watching 'Gasland' for a laugh.

The sad thing is you can produce methane using biological sources (plant, animal and human waste). The Chinese have been doing it for a long time. We're threatening an unlimited, clean supply by tapping a limited, polluting supply. It makes no sense.

neil craig said...

Perhaps you could produce evidence of the amount and toxicity of these "carcinogenic" compound causing cancer thousands of feet underground.

Technically almost anything can increase cancer, at least in labs, if used in sufficient quantities. Hence to say something is "carcinogenic" without sopecifying is simply to wave a scary flag.

How manyb people thousnads of feet underground have developed cancers from this cause. Please produce evidence. If it is zero you are clearly peddling a false scare, in the normal manner of the "environmental" movement.

As regards "radioactivity and heavy metals" - If you believe this to be a genuine threat you will have posted thousnads of times more often complaining about the mining of coal (heavy metals are obviously more easily moved around in lumps of coal than in gas). I'm sure everybody here would be interested in reading links to a few of those times.

You still haven't answered the question of how fracking gas thousnads of feet underground will prevent plants growing. I await evidence on that assertion too.

Tim Auld said...

I've got much better things to do with my time than provide evidence you will ignore. Have fun waiting for the singularity.

Stuart Staniford said...

Guys - please keep the discussion civil - thanks.

dr2chase said...

Tim A, regarding fracking and gas. Ignoring the issues of pollution from obtaining it, if we replaced our coal consumption with natural gas consumption, we'd be better off. There's all manner of pollution associated with the direct burning of coal, never mind the larger amount of CO2 per unit of energy. Burning coal gets you mercury, radioactive elements, and sulfur emissions (all of these things are somewhat controlled now, except that they end up in "sludge"). All that junk comes out of the ground with the coal, and is burned with the coal. It's intrinsic to the process. Gas is pretty clean by comparison, and easier to process to clean it -- if we can get it out of the ground cleanly.

I've got friends who've worked in the oil and gas industry, and I've asked some of them about this, and the problems from fracking, sound like problems caused by corners being cut (not unlike the Macondo blowout). If we can't stop those corners from being cut, then we probably should not frack, but there's no technical reason we can't control leaks, vapors, and emissions. If the well is correctly constructed, what goes thousands of feet down, stays thousands of feet down. Rock is just not that permeable -- if it were, we wouldn't need to frack it.