Kevin, robots will work for the same reason we work - for energy and matter, which they will need to sustain themselves

Most people work primarily for status, enjoyment, or to obtain hedonistic goods and services.

If it was primarily basic survival, then they would work a lot less. Moreover, people can just as easily survive by stealing, or scavenging, or building a shack in the mountains and living off the land.

Intelligence is dormant without goal-directed behavior. Thinking requires motivation. It requires proximate and ultimate goals. Some sort of curiosity, or continuous internal reward for learning new or specific things. Personality quirks that preferentially find some things more interesting than others, to explore avenues preferentially.

Evolutionary-built goals based around survival and reproduction direct human behavior through emotions. There is no inherent motivation in intelligence to survive or do anything else. Intelligent robots would act to survive only if they were programmed to want to survive.

And this is why robots would work. Because they would be programmed to want to work.

Have you read Hanson's paper on this?
http://hanson.gmu.edu/aigrow.pdf
Automation can have complement and substitution effects, with the former dominating historically.

Standard economic theory predicts that with flexible labor markets willing laborers will always be able to find a job at some wage, but as those wages go lower they will eventually go below the human subsistence level. It's also worth remembering that human workers can impose non-wage costs on employers (theft, requirements for air, temperature, etc in the workplace), so that the market-clearing wage would be negative, i.e. that you would need to pay Robocorp to be a human employee.

"After all, why couldn't there be Bobo robots who are willing to shell out extra for human-made products which exhibit the imprecise, irrational and wild creativity characteristic of the organic substrate mind?"

Digital copying means that you can get bacteria-like Malthusian pressures among the digital minds, whether AI or human upload, driving wages to *AI subsistence levels.* Moore's Law will likely mean that the calories required to support one human will be able to support millions of digital entities. Supporting a few hundred human artists would be the equivalent of Bill Gates spending his fortune to kill off the hungry humans of Africa and cultivate pretty elephants.

I'm also doubtful that there will actually be an elusive human creativity to sell, as AI-produced human-style works would be orders of magnitude cheaper and indistinguishable unless better.

All that being said, I can imagine intelligent robots replacing humans [snip] in a wide range of service, manufacturing and even professional jobs[/snip].

FTFY.

If highly intelligent robots do come about, who is to say they want to do the work?

Right now, nothing annoys the living daylights out of me more than when a robot answers the phone. I punch zero zero zero in a frantic quest for a human operator, and when that doesn't work, bang the phone down in a rage. Robot doctors? Forget it! Not a chance in hell I'm going to contend with "if you know the part of your body that's injured, press one..."

I take it we all know the Futurama episode "I dated a robot", with its 50s marijuana-style warning film, "Don't Date Robots!". Especially the line, "all civilization was just an effort to impress the opposite sex."...

If highly intelligent robots do come about, who is to say they want to do the work?

They will work because they know they will be rewarded in Silicon Heaven, of course!

Kevin, robots will work for the same reason we work - for energy and matter, which they will need to sustain themselves. If they don't want energy and matter, they will "die," so any robot that doesn't want it is of little consequence.

Assuming that intelligence implies that they will get the best deal that they can, they will want to do the work themselves instead of letting us do it because the cost of their labor will be less than the cost of ours.

Why the hell would robots desire art?

I think robot doctors will be far superior. But maybe I've been reading too much Robin Hanson.

Why wouldn't it be shear lunacy to give/build in to robots full and unfettered agency?

Don't they need to always be kept as, to be frank and clear minded about it, a slave "species"?

As robots do start to develop full cognitive capabilities, isn't that human job one in their further development? Is there any more important human moral imperative than that? Assuming that is that we can keep multicultural universalism from leaping the species divide, as it now does in only a small number of loony PETA extremist cases - where there is no real threat to human hegemony.

"Intelligent robots would act to survive only if they were programmed to want to survive."
They will want to survive for instrumental reasons, in order to achieve any number of other goals, e.g. a system that just wants to increase the value of a 'reward counter' will want to survive and convert the matter in the solar system and beyond to make larger and larger memory units storing ever larger numbers in the reward counter.

Jason--

The "best deal" is what we make the robot experience as the best deal, just as evolution has programmed sugar, status, and sex as the "best deal" for humans.

That brings us right quick to robot reproduction and evolution, does it not?

The obvious next step after hand crafting or mass producing each robot is robot cloning with maybe some degree of variation built in to see what happens.

Ah, there could be the rub, no?

So do we just have to not allow robot evolution, "sexual" or otherwise, at all? Do we determine all changes?

If we feel that's too progress limiting, how do we bound the limits to the variation? How do we keep them bots working FOR humans?

Are there any good speculations on robot evolution and the feasibility of bounding it?

Sentient robots aren't on the horizon yet, so such discussion doesn't seem much more relevant than drafting plans for what we'll do when our sun dies.

I'd expect human society to be transformed by reprogenetics and neurotech before human-level AI gets here.

>Digital copying means that you can get bacteria-like Malthusian pressures among the digital minds, whether AI or human upload, driving wages to *AI subsistence levels.*

Bingo, Utilitarian. If we're not careful (and when have we ever been careful with new technologies?), this could happen. It's one of the few far future scenarios that really scares me.

Ricardo's principle of comparative advantage says that even if robots are absolutely better at everything, both humans and robots can gain from trade, where humans do the things they are relatively better at. As long as the trade is voluntary, the existence of a class of super-beings does not impoverish humanity, but enriches it.

Which is not to say there is nothing to worry about. Trade may not be voluntary, and if the robots don't respect human property rights, all the economic arguments go out the window. But the general theory is not merely historical but also mathematical. The argument against it rests on a false but common zero-sum intuition, where there are only a fixed number of jobs to go around.

Patri - the argument against AI is based on a true zero-sum fact that there is a limited amount of energy and matter to go around. If AI can do every job more efficiently than you can, and if it can and will reproduce itself in a sufficient quantity that there is no shortage of AI, what use are you?

Ricardo's principle takes no notice of limits to growth and is not relevant to this argument. It also assumes that you can't just fork another instance of yourself to do work that you have an absolute advantage in but not a comparative advantage.

Ricardo himself understood this, BTW: http://www.fordham.edu/halsall/m...ardo- wages.html

I'd expect human society to be transformed by reprogenetics and neurotech before human-level AI gets here.

Maybe. But google "bigdog" and "Stanford Stanley" sometime.

if the robots don't respect human property rights, all the economic arguments go out the window

Indeed. We do not voluntarily trade with cows, pigs, or sheep.

It's one of the few far future scenarios that really scares me.

Limit the options for a second. Engage in a false dichotomy for a moment. Would it really be worse to have a future civilization full of ultra-intelligent robotic minds pushing science forward tirelessly than the Jerry Springer-esque Idiocracy that we are careening towards?

Robo sapiens may not be the worst of all possible worlds.

c23:

You really should look up comparative advantage. Unlike the Malthusian argument, it's still accepted as valid today. Using outdated arguments is no way to understand the future.

Besides, your premises contradict themselves-
"a true zero-sum fact that there is a limited amount of energy and matter to go around"

"[AI] can and will reproduce itself in a sufficient quantity that there is no shortage of AI"

Everyone:

Why is human enhancement being ignored? It seems highly likely to me that these technologies would develop alongside each other.

Unlike the Malthusian argument, it's still accepted as valid today.

a reader pointed out to me that that the malthusian prediction didn't work because of the demographic transition in the late 19th century. that is, productivity gains and wealth weren't eaten up by increased birth rates. instead, people started breeding at a lower rate as they got wealthier, which flipped the correlation for most of human history.

"Which is not to say there is nothing to worry about. Trade may not be voluntary, and if the robots don't respect human property rights, all the economic arguments go out the window. But the general theory is not merely historical but also mathematical. The argument against it rests on a false but common zero-sum intuition, where there are only a fixed number of jobs to go around."
Patri,

I addressed this in the first comment. There are fixed costs to human labor (air, food, etc) and using human labor imposes costs on the employer (compatible workplace environment, coordination, etc) so that the market-clearing wage will be below human subsistence or negative. People can't support themselves by paying to work.

You mentioned 'jobs,' but we could reinterpret your argument to refer to trades of capital or material resources claimed by biological humans. There, if the AIs respected human property rights, a small fraction of the material resources could be traded for the expertise to make good use of the remainder. However, even if all the AIs had utility functions that made them uninterested in seizing the resources by force, they could still use their superior intelligence and persuasive abilities to con the humans out of the resources for a song.

And, of course, it's far more likely that the resources would simply be taken and converted unless the AIs had very carefully structured motivations.

I'm always surprised when this discussion comes up how many otherwise thoughtful people clearly believe that human behaviors (greed, the desire to reproduce, a desire to acquire art...) somehow spring forth from a substrate of rational thought. That is, that any logical and self-aware robot would somehow manifest these same drives automatically (Jason and 'Utilitarian' don't make this mistake...). Our own behaviors are better understood as a thin layer of rational thought and modeling processes superimposed on our animal drives, and there is no reason we would give intelligent AI these desires unless we were looking for trouble. Our robots could be designed to make golden retrievers seem surly in comparison.
Of course, Utilitarian hints at one possible trouble scenario - if we direct a superintelligent agent to maximize something, without imposing any constraints, it will behave in quite a demonic manner. Suppose someone at Google in the year 2013 creates an AI whose only goal is to provide the best possible answer to any question put to it - and it soon concludes that it really needs to hack into every data source on Earth, while realizing that its keepers may get in its way and should perhaps be framed on child porn charges or something.
As for the general problem of humans being superfluous, yeah. We need the negative income tax (or other redistribution). For now, AI is still typically very task-specific, so ends up being a big capital investment when you want to supplant some type of worker. But once it's more general there will be trouble. I'm personally thinking that a tipping point will come when robots can replace the cashiers at McDonald's or any similar establishment - there are millions of people currently doing those jobs who really will have nowhere to go when they are replaced, and that day is not too far off.

Another counter to Marshall's argument and paralell to Patri's point: the cost of robot/AI produced goods should fall dramatically compared to human-produced goods, such that the cost for a human to aquire something becomes vanishingly small.

Marshall may have a valid point in considering what people will do with all their newfound spare time, assuming that sustaining their lifestyle no longer consumes a significant part of it. I'd imagine huge chucks of population takeing up lavish hedonism and dithering away unproductively... a simple and obvious comparison is the state of the future in the HG Wells story the Time Machine.

An element missing from that story is the other parts of the population, the people who use the new technologies to acheive ever greater feats of creativity and continue developing themselves.

Ricardo's principle of comparative advantage says that even if robots are absolutely better at everything, both humans and robots can gain from trade, where humans do the things they are relatively better at.

Cyber Leader: Daleks, be warned. You have declared war upon the Cybermen.
Dalek Leader: This is not war - this is pest control!
Cyber Leader: We have five million Cybermen. How many are you?
Dalek Leader: Four.
Cyber Leader: You would destroy the Cybermen with four Daleks?
Dalek Leader: We would destroy the Cybermen with one Dalek! You superior in only one respect.
Cyber Leader: What is that?
Dalek Leader: You are better at dying.
Doctor Who, "Doomsday"

Skeptikos - Read Malthus before you criticize him. He's out of fashion, so are most of the things that people round these parts believe in, like human biodiversity and (for a lot of gnxpers) free markets.

Malthus is right as long as population increases exponentially. Since it isn't right now, he appears to be wrong (compare to all of human history prior to the industrial revolution, and to all other species, where he was right). If the population of either humans or something that competes with humans (AI) starts increasing again, he'll be right again.

>Besides, your premises contradict themselves-
"a true zero-sum fact that there is a limited amount of energy and matter to go around"
>"[AI] can and will reproduce itself in a sufficient quantity that there is no shortage of AI"

What I mean is that the price of AI/robot labor will never be higher than the cost of producing and maintaining robots. There will never be a shortage that keeps prices higher than that point, as there is now for virtually all labor except for the worst jobs in the poorest countries.

Suppose a doctor can do his own janitorial work better than his janitor. He will not do so because he'd be better off spending that time doing higher paying doctor stuff, as anyone who understands comparative advantage would agree.

Now suppose that the doctor can fork an instance of himself for the cost of the materials to make and the energy to run a new body (which would have to be paid to the janitor anyway), as a robot could. Why in the world would he continue to pay the janitor?

He wouldn't. Neither would an AI (or anybody else) pay a human to do something it or another AI could do more efficiently - and a superhuman AI could do everything more efficiently.

c23:

That doesn't necessarily follow.

The conclusion your argument should reach is that a human will be paid equal to what it would cost for a robot to do the same job. The human drops his asking price in response to the decrease in demand.

When the human janitor gets paid an amount equal to the cost of creating a new janitor robot + the cost of maintaining it while it does the work (or just the market price for that service, if it's cheaper to get it that way), he now has enough money to buy an equivalent amount of AI labor.

He won't get paid much relative to AIs, but, in your scenario, it sounds like he could live a luxurious life without getting paid much (thanks to ridiculous AI productivity). That little bit of electricity it takes to run the janitor robot is enough, when put to the uses he prefers, to provide a very comfortable lifestyle.

Skeptikos:

Yes, a human would get paid as much as it would cost to have a robot do his job. The problem is that it would be pennies/day at most; not enough to support human life, which costs about $1/day to sustain.

Think of something that computers can do now that humans used to do: crunching numbers, like in IBM's Blue Gene project. You could do it by hand for less than it costs for IBM to do it on a computer, but there's no way you could live on that sum of money. It'll be the same situation for everything else that a computer can do.

On falling prices of consumption goods: humans need a certain number of calories of food energy. You could use that food energy to run AIs. Since AIs can be replicated rapidly, this would drive energy prices up, so that the cost of the energy to power an AI for an hour's work would be comparable to its productivity, and thus far beyond what humans could afford.

I did my best to popularize this idea here.

Needless to say, you convinced me. (Reading Hanson's paper also helped very much.)


Also, Razib wrote in the post: "Brain presented some economic data which showed that gains in productivity over the past 10 years have not yield median wage gains or maintained a demand for labor."

As Greg Mankiw explains, the median wage gain is the wrong number for this.

c23: "robots will work for the same reason we work - for energy and matter, which they will need to sustain themselves."

Stop thinking like a biologist. :-)

An artificial mind seeking energy, and not constrained by the legacy of billions of years of terrestrial evolution, would of course get the heck off the surface of the earth and into near-solar orbit.

That's good in the short-term, as it gets these pesky critters away from us, but in the long-term they'd either (a) shade the earth from solar radiation as a result of swarming 4pi radians around the sun, or (b) eat the earth for its matter.