In Chinese, there is a term called 矛盾. Individually, the word 矛 means spear and the word 盾 means shield. The story goes - once upon a time there was a merchant selling spears and shield. "This spear can pierce through any shield!" he says, "This shield can block any spear!" he says. An astute member of the audience asks "What happens if you hit your shield with your spear?". 矛盾 is the term for "contradiction".
So let me ask now, what happens if you get the robot hand to play with another robot hand?
Interesting. I speak Japanese and know the word 矛盾, but I had never paid attention to the meaning of its constituent kanjis (I don't think I've ever studied those characters separately).
they will draw with rock every time, because both will wait each other to change their hands.
they could draw with paper too, that will depend on how the robot was programmed.
First the robots both spot the other one appearing to go with rock and make a last millisecond change to paper. Then they both notice the other one making the change to paper, so they both go scissors... then they both go with rock again.
Typically in RPS you reveal your choice as your hand descends the third time. By the time your hand is at rest vertically, your move is already visible. However this robot doesn't make its move until afterwards. So the robot is effectively playing rock every time and then changing his play to beat his opponent, which is pretty blatant cheating. The only thing that makes this different than a human cheating is the robot is fast enough that you might not notice.
It is interesting to an A.I. researcher. I loved watching the slow-motion version at 50x
When I took Machine Learning, we studied different "bots" in various contrived game scenarios to illustrate the effects of imperfect information, non-deterministic outcomes, and other kinds of noise.
The most interesting outcomes were when the bots learned to cheat. Since the system doesn't punish cheaters, it becomes a winning strategy. I think humans have learned not to cheat - for the most part - because at the most basic level cheating reduces the group's overall survival rate. E.g. stealing food from another member of the group.
tl;dr if cheating is so obviously the best strategy, maybe the game needs to be revised? Because for this robot, cheat-really-quickly leads to 100% victory.
Look up Kristian Lindgren's work on using the Prisoner's Dilemma in a Sugarscape scenario. Among the results were islands of Always Cooperate strategies buffered by Tit-for-Tat strategies from Always Defect strategies.
Unfortunately, I've been unable to find any actual papers. :/
>I think humans have learned not to cheat - for the most part - because at the most basic level cheating reduces the group's overall survival rate.
I'm pretty sure that our anti-cheating behaviors are mostly instinctive responses, rather than something humans learned. We have an emotional concept of "fairness" that causes us to punish cheaters. There's also no need to invoke group selection to explain this (which is good, because group selection has been pretty thoroughly falsified). Game theory gives us perfectly adequate explanations in terms of individual self-interest for why cheater-punishment is an advantageous behavior.
I am not sure why you think group selection has been thoroughly falsified. From what I've seen on wikipedia and elsewhere, there's simply a strong bias against it being right, and many people react vocally and vehemently, suppressing research of group selection. For example, E. O. Wilson writes a book and all of a sudden he gets a barrage of angry correspondence and rebuttals.
I would say stuff like that discourages exploration in science, and pushes an echo-chamber agenda.
To me it's rather clear that groups dominate individuals. Let's look at the level of groups of complete organisms. The individual is dominated from early childhood and is taught learned behaviors which may actually impede its reproductive success. For example, non-alpha-males in many species simply don't reproduce. That may be ascribed to kin selection, but the mechanism is not kin-related: if a wolf is adopted, the same mechanism applies.
In humans this can be summed up by one word: CULTURE. All the morality, etc. is taught from a young age, and the kids are dominated by the culture, and as they grow up they don't think of doing it any other way. If the culture told you that you can't have more than one wife, then that's what most men will approximately end up with. The man isn't going to go out and get a harem because modern society has basically trained this man that a shitload of problems await him should he try to do it. In other societies, however, the harem is fine.
So basically, the individual behavior is dominated by the group. Selection is a byproduct of that. In men, it typically selects for those traits which the most prolific (reproductively) men possessed. Genghis Khan is one of the ultimate examples.
You're right that "group selection" doesn't exist; but our "instincts" have themselves evolved into what they are, for a reason? Therefore, "not cheating" must bring a benefit to the individual (or be correlated with some other behavior which itself brings a benefit). The question is: what is that benefit?
It would seem that group selection could be possible, depending upon time constants in the equation. E.g. if the group benefit accrued faster than the individual benefit.
Furthermore look at religeon, not a lot of benefit to the individual but the cohesive group survival is improved. Lots of cultural examples.
Heck, even look at cells cooperating to make an organism - thats group selection at the most basic.
>Furthermore look at religeon, not a lot of benefit to the individual but the cohesive group survival is improved.
There is a very strong individual pressure to conform to your tribe. Humans are extremely vulnerable on their own, but benefit hugely from being in a group. It is therefore of paramount importance for an individual to avoid being outcast from the group. I think it is likely that religion is more or less the name we give to the feedback loop that causes tribal conformation pressure to override rational analysis in an individual's belief selection process.
> Heck, even look at cells cooperating to make an organism - thats group selection at the most basic.
No no, that is kin selection in its most extreme form. Kin selection is very strongly backed by evidence, and differs from group selection in that the individuals have an above average statistical overlap in their genomes. For example, a brother and sister have an average of 50% shared genes, so genes that promote sibling cooperation can propagate more effectively than their complementary alleles, even if that cooperation causes some reduction in survival of the individual.
In the case of cells comprising an organism, the overlap in genetic material is 100% (excluding anomalies like mutation and HGT), so we see enormous amounts of individual cell sacrifice which prove advantageous in preserving those cells' genes by benefiting the other cells in the organism.
Groups that find themselves in the same area become kin a few generations later.
So I don't see why kin selection must be the sole mechanism to explain all kinds of group selection. The group dominates the individual by indoctrination from a young age, as well as by force. You can refer to all this as "culture". The group is smart enough to threaten the individual with things that the individual cares about, and the individual has evolved to be receptive to such correction by the group, and learn to stop doing something, even if it would give them a better reproductive fitness.
It just so happens that groups which live together start becoming genetically related down the line. But the mechanism that acts isn't kin selection.
The robot cheating isn't the interesting thing here. Implementing that would be a simple case statement. This is a demo of fast digital signal processing (DSP) combined with fast mechanics/electronics.
"Look, this is not what I do, but I got an idea for one of your commercials.
You could see a carpenter making a beautiful chair. Then one of your robots comes in and makes a better chair twice as fast. Then you superimpose on the screen, "USR: Shitting on the little guy. That would be the fade-out."
So I know the article isn't entirely serious, but the "score one for the robots" attitude seems pretty common and it's perplexing to me. Isn't this "score one for the humans", since they have created and now control a new kind of machine?
When I see robots doing new stuff I don't worry about humans becoming obsolete (whatever that means), I look forward to the day when humans are freed from the grunt work they do now.
Some people worry about losing their jobs in the rock-paper-scissors industry, but I think this is a social problem that has nothing to do with robots.
It is clearly a social problem, not a technical problem. But it clearly has a LOT to do with robots (and computing in general, and technology in general).
I am not a neo-Luddite (nor Luddite, nor primitivist, all of which are different things), principally in that I do not subscribe to normative Luddite beliefs (beliefs about What We Ought To Do). But the Luddite fears are legitimate fears, for those affected; when the work you currently know how to do is replaceable with (sufficiently-cheap) machinery, it makes you personally poorer. (Of course, when MY job is replaced by cheaper machinery, it makes YOU personally marginally richer, because you get better prices on the goods that I used to produce, and over time this appears to be a net win).
And socially, we observe that increasing automation has, for the most part, not freed humans from doing grunt work, as they once did. Except for those people that have been "freed" into poverty. Admittedly, the poverty of today seems to me to be quite a bit nicer than the poverty of 200 years ago (esp. urban poverty).
socially, we observe that increasing automation has, for the most part, not freed humans from doing grunt work, as they once did.
Sure it has. Much less people work on the land now. Much less people shovel dirt, till land and harvest crops. Also I don't need to wash ky hands by hand, wash my dishes, and empty my chamberpot anymore.
That's not to say there is no poverty (there is), but its wrong to say there is the same amount of grudge work as there was 200 years ago.
I agree that a shrinking percentage of society is doing agricultural work, and certainly many specific tasks of grunt-work-of-yore are less grunt-y and more automated.
But my assertion was not that "automation has not started doing any of the grunt work we once did". My assertion was that the segment of society that WAS doing grunt work is mostly not elevated out of grunt work TODAY. I assert that, rather, those people are mostly either still doing grunt work, or else are unable to find a useful role in society (a leftier turn of phrase would be "they are being abandoned by society"). I can certainly concede that there has been SOME improvement in this, but it's vastly less than starry-eyed optimists seem to suggest.
Remember that this followed mwd_1 stating a vision of the future, in which "humans are freed from the grunt work they do now", by which I assume (s)he meant not just today's variant of grunt work, but grunt work in general. Basically, my whole claim is that techno-utopianism is unrealistic, and it is unrealistic for social reasons, not technological reasons. Not that it stops me from wishing for it.
Oh, and a minor unimportant correction: the phrase is "grunt work", not "grudge work", at least in most English-speaking communities. Look it up.
I assert that, rather, those people are mostly either still doing grunt work, or else are unable to find a useful role in society (a leftier turn of phrase would be "they are being abandoned by society")
Ah, now you're moving the goal posts. From "200 years ago X% of society were doing grunt work, and today X% of society are doing grunt work" to "x% of society are doing grunt work or are just loafing around". I agree that it's wrong that there loads of people who are unemployed, or working crappy jobs, but it's much better than when they had to shovel shit for 10 hours a day.
Tell a serf in the middle ages about the terrible future where people don't work till they're 16ish (sometimes 20), then sit around in housing estates all day doing nothing. Tell them how horrible it is.
Things have gotten better. We need to continue to make things better.
> I agree that it's wrong that there loads of people who are unemployed, or working crappy jobs, but it's much better than when they had to shovel shit for 10 hours a day.
A lot of those crappy jobs involve the equivalent of shoveling shit for 10 hours a day. Jobs available to power in lower economic brackets and classes are demanding: retail, fast food prep, janitorial/custodial work, day labor, etc.
More importantly, just because a job doesn't require physical toil doesn't mean it isn't a soul draining experience. The fact of the matter is that increasing automation hasn't ended class division because automation schemes are employed to increase the profit of those in control of business. Improvements in technology do not automatically improve society, as most technologies get exploited to divert more wealth to few individuals.
Yes those jobs are shitty, but they are better than 200/500 years ago.
(a) There are more labour laws now. You don't have to work as much. No more shoveling shit for 10 hours a day, 6 or 7 days a week. (b) There are health and safety laws now, if you were to shovel shit now, you'd have to have a mask etc. (c) There are more labour laws now, so that the employer has to provide tools and training. No more "shovel shit for 10 hours, you have to bring your own shovel, oh you're shovel got broken, you're out of a job" (d) There is more of a safety net. No more shovel shit for 10 hours, oh you broke your hand? Hope you can live without food until it heals. etc.
Those jobs suck, yes. But ask anyone doing them if they think mediæval shit shoveling is just as good (or better), and they'll say no.
The present of labor laws does not mean that everyone has access to union work or that those laws are enforced in some situations. The same goes for health and safety regulations. There are plenty of jobs where you buy your own work equipment (I've had them). Also, the most economically disadvantaged work more than 1 job, so saying that people don't work 10+ or 6 or 7 days is not accurate.
The bottom line is that those at the top have always exploited technological advance for themselves while leaving others behind. This is a primary function of capitalism throughout history. That some conditions may have been worse in the past means nothing while economic oppression continues right now.
I think "grudge" might be a combination of "grunt" and "drudge", aided by the fact that "grudge" is a real word with a different meaning. I noticed it too.
This is not applicable to my comment. I was merely attempting to quickly and easily explain why treating "robots" as a distinct entity, rather than merely human tools, is an extant viewpoint.
If you heat the rock up enough you can burn thru the paper, melt the scissors and win everytime.
Interesting is the aspect that given 3 choices people will tend to fall into a pattern, albiet one influenced by what the other player does.
Also worth noting that variations of RPS exist for example Malaysians use water instead of paper, and in place of scissors is a bird, made by holding the fingertips together, forming the shape of a beak. The bird drinks up the water, the water sinks the rock, and the rock kills the bird." (from http://psychology.wikia.com/wiki/Rock_paper_scissors)
Funny thing is when computers cheat it is an achievement, almost feel descriminated against being a human who if cheated would not fair as well.
Well, those high-speed algoritms aren't investing -they're making money on split second changes... nothing long term, and in no way predicting long-term market changes.
People do that.
This is scary because the more you know about human perception, the more you realize that this kind of anticipatory system can be applied to all kinds of human decision making.
Actually, it anticipates your complete movement based on your initial movements. Recent neuro scientific research suggests that we consciously justify subconcious decisions (as opposed to consciously reacting to subconscious stimuli). Imagine a system that knew what you were about to say based on your context, body language and subtle facial expressions.
> Actually, it anticipates your complete movement based on your initial movements.
Well, it seems you still haven't read it because there is not anticipation according to the article:
> It only takes a single millisecond for the robot to recognize what shape your hand is in, and just a few more for it to make the shape that beats you, but it all happens so fast that it's more or less impossible to tell that the robot is waiting until you commit yourself before it makes its move, allowing it to win 100% of the time.
I suppose what you suggest might be possible eventually but that's not how it works at the moment.
I predict an eternal rock/rock tie - unless these robots are also learning as they play, in which case one of them will switch to paper, and win a few times, before ... actually, I have no idea. Let's do this.
There's a big difference - the chess one is playing the game honestly, whereas this (whilst still very impressive) is basically just very high speed cheating.
There was a funny tv segment I caught here in Japan about how to increase your chances at RPS.. If you let out a blood curdling yell or make a semi attack-like movement just as you throw down, people's instinct is to clench. You throw paper.
Given how fast it responds, and that most of that response time seems to be the time it takes for the physical mechanism, this gives me hope that the Kinects of the future will not be unusably laggy.
So what about robot vs robot? Will one be the first to make a decision based on the assumption the other will be rock, and then it will choose paper? Will this be the outcome 100% of the time?
Logically you would assume that both robots would tie. By making it no longer about a faster reaction time you would be reinstating the basic principles of the game. The interesting follow-up would be how would two bots that just passed Turing tests perform?
You could have gained a lot more nerd-cred by mentioning this monstrosity instead of that trivial rps5 version: http://www.umop.com/rps101.htm I would actually be impressed if they implemented 101 different gestures instead of just 3 or even 5. (They'd need a second hand though.)
I guess to win, all I need to do is to tell the robot to not win. Unless of course, someone threatens to murder a human hostage if the robotic hand does not win.
I somehow suspect that the asimov subroutines have yet to even have a comment marker, let alone any code. Also given the highspeed nature I'd be somewhat against testing them out with any human appendages of any kind, especialy after a certain Big Bang Theory TV episode.
So let me ask now, what happens if you get the robot hand to play with another robot hand?