I was talking with my sons about the war we lost in Vietnam, and how we lost 50,000 soldiers. My elder son said, "We should have used robots."
We have robots now, I said, referring to drones.
But we didn't have those then.
"What exactly is a robot?" My kindergartner asked. "How do you define it?"
"Two things: it has to be man-made, but also has to move by itself."
"Then the fan is a robot?" Julian asked. The fan in the middle of the room swiveled on an axis.
It seemed too easy to call it a robot, but I guessed that according to my definition, it was.
"People are also robots," he said. "They are man-made, and they move."
"Well, they have to be mechanical, too," I said.
"Then, that's three things, right dad?" He said.
"Right," I admitted.
"That was the most fun I had all day," Julian said, as he prepared to go to sleep.
He held up three fingers to indicate his victory, and then closed his eyes, and with a smile he went to sleep.
Bested by a kindergartner.
19 comments:
I admit that this post was a lob to Stu, from whom I expected a smash. However, he has declined to take the bait, even though he works in some kind of robot department. Therefore, I had to go to all the trouble of looking up robot, wikipedia, and got this tightening of my def:
"The more the control system seems to have agency of its own, the more likely the machine is to be called a robot. An important feature of agency is the ability to make choices. Higher-level cognitive functions, though, are not necessary, as shown by ant robots."
Ok, so Julian will be unhappy to discover that he is wrong, and that I am correct, if I add the stipulation that it seems to have agency of its own.
As we go to sleep this evening, I shall slip this notion in, and see what fancypants objections he can come up with next.
The fan just does the same thing over and over: the definition of psychotic, except the psycho does the same thing over and over even though it doesn't do him much good.
The fan, on the other hand, is doing me some good, and keeping me in a rather good mood, in fact, as I finish reading Bianca Lamblin's A Disgraceful Affair, about her affair with SdB and JPS when she was a teenage Jew in Occupied Paris.
sartre had terrific blackheads and didn't know he had stains all over his shirts and pants. They were both robotic to a degree: working hard on their literary reputations to the exclusion of all else, and sucking in their fans and students and turning it all into lousy literature: too ambitious even to have children.
Lamblin does them in beautifully.
This should be required reading. They come off almost like the Clintons, except the Clintons' books sold much better.
The fan doesn't appear to have agency.
If we said that a person was acting robotically, we would be saying that they seemed to not have free agency, right?
Determinism is therefore a robotic world view.
I just explained this all to Julian and he asked, "What does mechanical mean?"
I think I have to look it up. I know, or think I do, but I find the dictionary is often quite helpful.
Kirby and Julian,
I've had other commitments. I know that Kirby doesn't follow these things, but last night was the first game of the Stanley Cup finals, and the Hawks are in it. :-). And tonight was Church Council.
Anyway, I work in a computer science department, not a robotics department per se. But let me try to tackle this anyway.
I see the distinction between a robot and a mere machine as resting in the robot's far more sophisticated ability to respond autonomously to its environment. Mere machines can exploit various feedback mechanisms to try to sustain some sort of equilibrium, but they lack the ability to really surprise us. Whereas robots, properly, should be complex enough that they can surprise us.
Kirby raises the point, and I think it is an interesting one, about how saying that a person is being "robotic" is a pejorative, which essentially stands the distinction I'm making above on it's head. This does seem right, because we believe that robots ought to be predictable in principle, even if they're not always predictable in practice, and we believe we as humans always reserve the possibility of being unpredictable in principle as well as practice. One way of seeing this is that we believe that if you replaced one robot by another, intended for the same purpose, they'd do the same things in response to various stimuli, but we don't believe people would do this. Or even that we would necessarily do the same thing ourselves presented with the same stimuli twice. But I think this view is in error, as robots can incorporate randomness, and so show the same sorts of variation in behavior that we do.
The "incorporation of randomness" is a good trick, and is perhaps a nice definition of the chaos of human lives.
Julian got into the fun of definitions last night and started asking me questions:
What is the difference between a quiz and a test?
A test is longer.
AND?
It counts more heavily for or against you?
Right, Julian said.
We'll talk about the incorporation of randomness.
I think in hockey there is a certain randomness. Goals are scored off the legs of defenders by deflection, or off the post.
It's not all randomness, but randomness plays a part in higher level sports.
My wife is Finnish so we watch the Finnish team when it's playing a high-level game. I used to go to games when I taught in Finland. There wasn't any fighting, but aside from that it was very high level soccer.
I don't know why but the Swedes and Finns didn't fight in their home countries. Which nationality is most responsible for fighting?
Is it the French Canadians?
Kirby,
My wife is Finnish so we watch the Finnish team when it's playing a high-level game. I used to go to games when I taught in Finland. There wasn't any fighting, but aside from that it was very high level soccer.
I think that we could use the direction that various comment threads go on this blog as a high-quality source of randomness.
Fighting is less common in hockey now than it was in the 70's, when Hockey succeeded in getting itself kicked off of TV, to its enduring injury. Most of the improvement comes from truly vindictive penalties against people who join a fight in progress. This has tended to make fights small, usually just two guys, and easily controlled. Hockey is a physical game, and contact is going to lead to fights. I don't see this as being a nationality issue, but team psychology does matter. And these days, even the two-person fights aren't all that common, and the texture is very different. In the old days, two guys would drop gloves just because they wanted to know who was tougher. The fights that I've seen this year have pretty much all come out of the passion of goal rushes and the resulting scrum.
I don't know why but the Swedes and Finns didn't fight in their home countries. Which nationality is most responsible for fighting?
Is it the French Canadians?
Philadelphia sports fans are a particularly hard and unforgiving lot, and while they've had some successes (Phillies), they've had more than their fair share of meltdowns too (Eagles, 76ers). The chippiness rubs off on the teams, but somehow especially the Flyers, who earned the nickname "the Broad Street Bullies." The Flyers are still have more than a bit of that 70's era "enforcer" mentality in players like Carcillo and Laperriere. The Hawks aren't cream puffs, but they build their game around speed and discipline than intimidation.
So think it's the Flyers.
In addition to randomness I think that logic asks us to put a different emphasis on certain values that lead us into any activity. Feelings underlie many of our values, and feelings are notoriously ephemeral. If our values are upheld by how we feel about something, then those values will always be in a state of flux, leaving us to act differently as the feelings ebb and flow.
I couldn't get many of the cartoons to load. I did get the Tibetan/Chinese Olympics to load, and also the one about the older man looking through a fence as he waited for Al Salah, I think was the name of the person (is this the guy who kicked out the medieval christian armies?): Saladin I think we call him.
At any rate, there IS humor here of a kind.
I think too there IS humor in the Sufi tradition -- all the jokes surrounding Mullah Nasrudin.
Nasrudin is a wisdom figure who is genuinely funny. In the Gurdjieff literature the writers are always telling Nasrudin jokes.
I was once reading the newspaper about some event in Kabul, and an Afghan told an amusing version of the events, and gave his name as Mullah Nasrudin, which was reported as factual in the paper.
I laughed as I read it.
If robots have to surprise us, then isn't it true that what was once considered a robot will always be considered a machine in the future? A constant moving of the goalposts.
A.I. deniers do this. "A computer will never beat a master human at chess."
Done.
"Well, chess is a rote game. A computer will never be able to translate language realtime."
Done.
"Well, language is too systematic. A computer will never pass the Turing Test."
Wait and see, mon frere. Wait and see. When a computer does pass it, they'll be saying that the Turing Test is meaningless.
Brett:
Well, sort of.
The Deep Blue games were deeply flawed and I've yet to see consistently reliable translations, though I'm certainly all for that.
But those are false constructs regarding A.I.--the assumption that knowledge is sapience.
Brett says: "'A computer will never be able to translate language realtime.'
Done."
Not so fast, Brett. Machine translation from one natural language to another is fraught with problems, difficulties and errors (some quite funny, as anyone who's tested Babelfish, Intertran et alia can testify), except perhaps for tasks like doing weather reports, where context, vocabulary, and syntax are quite limited and structured.
On what I know about AI and the Turing Test (following Searle et alii), I tend to the agnostic, though it'd be interesting to know what others here think, especially stu.
I wonder about Deep Blue. Kasparov said two things:
1. The computer was trained to defeat him, personally, and he was not allowed to play a prep match to get a sense of the computer. That is, the computer had the aid of years of spying on Kasparov.
2. Also, Kasparov saw Karpov and others in the Deep Blue War Room, possibly giving human help to the computer.
This was not, in other words, and clean and clear defeat. It was highly suspicious, and even at best unlike any other chess match between very high-ranking players, in which the games of the others are available to all.
Deep Blue was immediately retired after the match so that it couldn't be defeated. However, if it was left to play on its own for a few years even amateurs would learn its flaws and play against them and crack it.
My own chess computer is fatally prone to queen traps. I can beat it, and am the rankest of amateurs.
This, I think, is the problem with robotism: the predictability. Even if you program it for randomness, it will be a highly specific randomness that others can suss out eventually, and prey upon.
Humans are also logical, which makes them easier for hit men to get, but they can adapt, and change, from within, whereas I don't think robots can do this, although they are working on it.
Brett writes,
If robots have to surprise us, then isn't it true that what was once considered a robot will always be considered a machine in the future?
Part of the problem here is that robot has a number of distinct meanings, and some of those cover fairly simple machines. I took Kirby and Julian's discussion to be using the word robot in approximately the sense of the first gloss in OED: "One of the mechanical men and women in Čapek's play; hence, a machine (sometimes resembling a human being in appearance) designed to function in place of a living agent, esp. one which carries out a variety of tasks automatically or with a minimum of external impulse." If this gloss is uninformative, Čapek's play is R.U.R., i.e., Rossum's Universal Robots, wherein the word "robot" itself was coined.
I didn't approach the word technically, and "surprise" was intended to express a level of complexity that captures my sense that when "machine" and "robot" are being used in opposition to one another, the intended distinction is one of complexity (perhaps even intelligence, although this makes the bar higher than the state of the art) and/or human-like appearance and some reasonable set of human-like behaviors.
As such, "surprise" is slippery, as experience with a particular machine gives rise to a better understanding of how it works, and therefore improved ability to predict its behavior. I suppose I was channelling Arthur C. Clarke's more famous aphorism, that "Any sufficiently advanced technology is indistinguishable from magic."
JADL writes to question Brett's claim that machines can do realtime language translation. I think there's justice to both Brett's claim and to JADL's questioning thereof. We seem to be at a point where facilities like Babel fish can provide useful realtime translations of "ordinary writings." I think that a lot of the success has to do with the fact that "ordinary writings" tend to be both robust and redundant, and that the readers of these translations are intelligent and flexible enough to make sense of them despite their limitations. But once you move to texts which are more tightly engineered, e.g., legal or scholarly writings, then machine translation (realtime or not) is going to let you down.
As for the Turing Test itself, I think it is an eminently sensible test based on an eminently sensible philosophical position: once it becomes empirically impossible to distinguish two things, it becomes pointless in practice to do so. That said, I'm sure that Brett is also right. If and when computational systems begin to pass the Turing Test, there will doubters of two types: those who doubt that the passes were valid (c.f., the discussion of Blue Gene in this thread), and those who decide that the test itself is inadequate. The rear-guard efforts to defend the notion that humanity has a privileged place within creation are hardly going to lay down their arms at this juncture.
Kirby,
This is about hockey. Clearly in last night's game, it was the Hawks (esp. Byfugulien, who has more than a bit of the enforcer personality) who were chippy. So let me suggest the hypothesis that visiting teams, especially in stadiums with many aggressive fans, are more likely to initiate fights.
This makes sense if you think about it.
Stu, I admit that I do think there is something special about life, as opposed to machinery (of which robots are the most advanced toward something like life).
You can fix a machine. If one dies, you can usually get it going again.
With life, you can't, unless you're Jesus, and the dead person is Lazarus.
If we get our lives back after death, it is in another format, from what I understand.
But with machines, you can generally just replace the parts, and they are good to go.
With Deep Blue, I am relying on a video I saw about the match series which told the story primarily from the Kasparov viewpoint, although it did have interviews with the makers of Deep Blue.
I stand by the important objection that Deep Blue had off site handlers who could have an impact on the play of the programming during the matches (in the show I saw Kasparov claimed that the moves were Karpov's, and he later saw Karpov in the hallway after another match, and Karpov turned away from him).
But there is also the inability of Kasparov to study Deep blue's games, or to have a rematch.
I think if I could blindside Mike Tyson I might be able to hurt him.
But if I tried a second time, I would get creamed.
If I kicked him in the face as hard as I could while he was asleep and not expecting it, I think I could hurt him.
But if he knew I was going to do that, I couldn't do that. And the second time I would get my butt kicked.
Deep blue is something like that: it's handlers say it's stronger than Kasparov based on a spurious set of circumstances which favored it.
I could go around saying too that I had kicked Tyson's butt (or face) if the circumstances were arranged for me to do so.
But if we had a series of 100 boxing matches, I think my initial victory would be paid out with 99 serious losses.
I think Kasparov would be able to do this to Deep Blue.
But I'm an amateur, and I freely admit, biased toward humanity.
Given that, I still think the human chess player will win over the machines over a period of a thousand games.
But it's the human PLUS the machine that might beat any human, given that speed is an essential part of the play.
Deep Blue had Karpov behind its curtain.
This would be like me against Tyson only I have a machine pistol.
the analogies aren't perfect, but are meant to be suggestive of the imbalance?
(Can't comment on the hockey -- although I do intend to watch the first game of the NBA finals this evening with Julian and Tristan, if they can remain awake that long. I don't care who wins, but I suspect that Paul Pierce and Kevin Garnett will defeat Kobe and whoever steps up on his team over the series.)
IF Deep blue's makers weren't more or less certain that I am right -- that a sustained series between Kasparov and Deep blue wouldn't go to Kasparov, then they should let the machine have a go. As it is, I can only ascertain that they know as inevitably as I do, that the risk is too great against them. Once Kasparov started to see the patterns of thought of the machine, he could adjust, and kick its butt.
Robots may have speed and strength, but they don't have ingenuity, or the ability to adapt.
Kasparov would have won. I think Deep blue's makers are more or less admitting this by taking it out of contention. Deep Blue's a crybaby that kicked Kasparov when he was half-asleep (didn't know the background or the strengths of the programming), and now doesn't want to get reamed.
Can't blame them, but I see the refusal of a rematch as tantamount to concession of defeat.
stu:
On machine translation (MT), there's little doubt some translators of some kinds of texts find MT as a useful first step before the inevitably heavy word-for-word, sense-for-sense editing process--not omitting extra-textual glossing of culturally-bound terms, collocations and neologisms, marking ironies and class registers, avoiding false cognates (faux amis), and the like. And this is merely to scratch the surface.
Not sure what you mean by "ordinary writings," but if accuracy is valued in a given translation, MT "gisting" (which is usually what extended MT translation provides) is quite inadequate.
While the Turing Test may satisfy you on some pragmatic level, philosophical difficulties abound in implying (or even proclaiming, as do some materialists and mechanists like Dawkins, Dennett, Churchland, et alii) that computers can "think" and are capable (now or at some future time, as promoters of the "exact-science-just-around-the-corner" ploy confidently assure us) of "consciousness." Computational activity should not be confused with conscious thinking, and while computers can help us think, they're not in themselves thoughtful.
Some of the anthropomorphising verbiage used to describe computers and what they do may be at fault here (as well as metaphors linguistically mechanising human thoughts, feelings, intentions, etc.), but "the mind is what the brain does" set cannot solve the mind-body duality problem by sub-philosophical fiat.
Kirby,
Brett's claim was that "A computer will never beat a master human at chess," not that "a computer will beat the reigning chess world champion." Thus, questions as to whether Deep Blue acted in isolation in beating Kasparov are immaterial, since Kasparov was much, much stronger than the weakest grandmaster. There are many computer chess victories over grandmasters. There's a decent survey article on Wiki:
Wiki: Human-computer chess matches
And returning the Kasparov-Deep Blue match, it is important to remember that match play isn't like game play. Karpov could have been legitimately involved, per the rules of the match, in adjusting the program between games. To draw a couple of threads together, a useful analogy can be drawn between a chess master in a match and a goalie in a hockey game. Sometimes, a few lucky goals can "break" a goalie, making him far more likely to give up goals on subsequent shots. This can also happen in chess, where an early loss can cause a master to lose confidence in his ability to prevail, and therefore more losses. Kasparov cites this phenomenon in the wiki article op. cite. Computers, which don't preserve a psychological sense of being dominated, can't be broken in the same way, merely beaten. And there is a distinction here.
JADL,
By "ordinary writings," I meant web pages, fiction as might appeal to the man-on-the-street, etc. And my point was not that the translation was accurate in and of itself, but rather that the source texts contain enough redundancy that an intelligent reader can usually reconstruct from an imperfect translated text the sense of the original. In effect, redundancy in the material coupled with intelligence on the part of the end reader makes up for defects in the translation itself. Moreover, the intelligent reader corrects the infelicities in redundant texts so effortlessly that he is often unaware of the repairs he's making, and gives the translation credit for some of the intelligent analysis that he himself has provided.
As for your philosophical objections to the Turing test, let me say that I think you're falling into the vitalist trap of assuming that our minds are somehow more than manifestations of our physical bodies. Where is the proof? I am not aware of any evidence that human intelligence is anything but a manifestation of our physical bodies. Are you? Such evidence would overturn the central dogma of computation neuroscience, and would be a very big deal indeed.
It seems to me that the capabilities of humans represent entirely convincing plausibility arguments for the capabilities of engineered systems. This does not mean that we know how to build systems that exhibit human intelligence, nor that we understand in full how human intelligence works, for neither is the case. But there seems to be no barrier in principle. If our bodies can do it, then it can be done. Although not necessarily by systems that we can build today.
I'll note that this does not require anthropomorphizing language, indeed rather the opposite. I don't need to falsely ascribe to existing machines human characteristics, but instead merely make the perfectly valid point that our bodies are no more than wonderfully complicated machines.
Stu, I'm confused about Brett's claim because he puts it in quotes and yet you quote it as if he's saying it.
My mind will try to understand what happened on this in the morning.
I think he's saying something differently than what you're saying he's saying, but shall have to check again in the morning. I have just a minute here to push comments through.
stu:
Yes, intelligent readers of texts both in the original languages and in translations can correct some errors in what they read as they read, but it's considerably more difficult to detect the errors in translations if one has little or no competence in the original languages and usually doesn't even have available the source language texts. I should think intelligent readers might generally require more accuracy in translations than less and be more likely to read translated texts than your hypothetical monoglot man-in-the-street. But again, it depends on the kind of texts and the readers' purposes in reading them.
You're right to say that I think (!) minds to be more than mere manifestations of our physical bodies, or more specifically, brain-parts. Though of course brains are necessary for the consciousness of minds--whether directed toward rule-following or instrumental behaviour, sensations, reflection, memory, scientific discovery, artistic creation, reverence, exercise of free will, etc.--it doesn't follow that minds are sufficiently explained as computational machines blindly engineered by natural selection, even less so as identical to such machines. What some philosophical materialists seem to do is to reduce subjective experiences or qualia to objective or quantitative events (whereupon the "I am" disappears into the "me") determined by evolutionary biology and fully described only in terms of neural events "in" the brain (Putnam's "brain in the vat" conception). Leaving open the question of how brains transcend themselves in human experience and form communities of minds seems more warranted to me than yielding to the arguments of materialist neurophilosophy.
You may call this scepticism about minds as computational machines "vitalism" (with its pejorative connotation as "folk psychology"), but since we're discussing human consciousness and intelligence, "dualism" seems more appropriate.
Post a Comment