The current generation of "consumer robots" is driven mostly by robot-love: people enjoy things which move around on their own, especially if they can build or tinker with the gadgets themselves. That much became clear at a recent symposium on Robots, which I described here last month. The consumer robot business today is …
The niche market for robotic replacement of humans is high-attention repetetive tasks, which humans do pretty well, but lack of "boredom" capacity in robots means they'll be able to do better.
> We lack a common mathematical language for generic sensory input
The original meaning of 'common sense' is exactly this; the philosophical construct which posits that different sensory inputs must all be "brokered" somewhere into a common sense which models the external world (once we have decided to be existentialist, and assume that there is one). I am unaware that any robotic efforts are focussed on common sense.
Bacteria can't see...
Nice article...but bacteria can't see or walk, that's the province of complex, multicellular organisms; they can't hoover rooms either - but that's because they can't get brushes small enough, and anyway they're so little it would take too long ;)
robots a stupid idea
There is no idea dumber than a robot smart enough to replace a human being at anything. I mean that very seriously. This idea is stupid for the same reason that pointing a gun at your head and pulling the trigger is stupid. What do we do with the replaced human beings? Get it?
I am continuously amazed at the number of earnest people wanting to beaver away at building their own replacements. What will we do to make food money? Robots will always have many moving parts. They will be expensive to purchase, so only the rich will be able to afford them. They will replace the poor without making the poor go away.This will displease the poor.
So, robot mars rovers? Cool. Robot reactor repair techs, great. Robot drivers and cooks and housemaids? Uh uh, no way. Only you don't get one without the other. After replacing astronauts, replacing you and me will be a piece of cake.
Now, where did I put my robot soldiers. Gotta keep civil order until the poor finish starving. Oh yes, there they are...
robot problems falling one by one
We make autonomous mars rovers and autonomous cars nowadays. We are solving problems like dynamic balance (walking), scene analysis and route planning. Yeah, there's no Windows API for it yet, but the problems are falling, one by one. Intelligent behavior is just harder than coding a database for your personal contacts, or gluing together an R/C airplane, so it's not surprising that tinkerers haven't mastered it yet. But it's coming.
re: Bacteria can't see...
Neither can a roomba either, not in the same way as we expect. Bacterial intelligence in reference to a roomba is dead on. Bacteria basically eat and move towards more stuff to eat. They may not see anything but are aware of their environment. If it didn't know where more food-stuff is it would quickly die shortly after the 1st bite. So a roomba works the same way, it knows where it is, and where it's been, and that where it's been is clean so more dirt is only where it's not been yet. The added bit of smarts is the unstucking logic as it were. Unlike bacterium, there is usually only a single roomba so it has to know how to get unstuck. Bacterium just reproduce to get around obsticles--"junior will catch that spot next to me."
re: Bacteria can't see...
And you're right, they can't hoover a room, but they can clean up oil spills. But is it cheating if they just do it because they're hungry?
Blue Screen of Death again?
"We make autonomous mars rovers and autonomous cars nowadays. We are solving problems like dynamic balance (walking), scene analysis and route planning. Yeah, there's no Windows API for it..."
Of course there's no Windows API for it. Can you imagine the chaos that would result from releasing robots which used Windows as their "brain kernel?" Would you want a Windows robot performing surgery on you or a loved one, or "fixing" a nearby nuclear reactor, or piloting an orbital reentry vehicle above your home?
But the basic "ability to learn" *has* been built, and the pilot "brain" is under study; Google for it. Of course, the problem right now is that it takes as long to train the robot as it does to train a human baby...
It's a philosophical problem
Before we can recreate intelligence we need to know what it is, which is a problem for philosophers.
We maybe need a coding Kant before we can make the most of an engineering Einstein.
Re: robot problems falling one by one
Kurt, your examples "dynamic balance (walking), scene analysis and route planning" all have one thing in common. We know when we've got the right answer. For problems like balance, there is even a mathematical algorithm that we can start from. The final implementation of the walking robot may not solve the dynamical equations explicitly, but you can be sure its designers started from that.
For intelligent behaviour, not only is there no "brute force" method for a starting point, it is debatable whether there is even a way of checking the right answer. (Turing's test being somewhat inapplicable to hoovers.) How do you even start to solve a problem when you've no way of knowing whether some answers are more correct than others?
Since I don't believe in the soul, I expect we'll figure out AI eventually. When we do, it will change society beyond recognition. Happily for those who prefer their humanity in the traditional "flesh and blood" form factor, right now no-one has the slightest idea how the brain does it.
Scene analysis is solved?
According to Kurt Guntheroth, scene analysis is solved. I beg to differ. If it were anywhere near solved, the winner of the DARPA grand challenge wouldn't have had to supplement very basic image analysis (basically finding the road coloured part of the image) with radar and GPS.
And route planning? Well, if you call finding the quickest path through a pre-digested network of nodes and arcs route planning, then yes, it was solved decades ago. Unfortunately, that won't really cut it for consumer robots. A consumer's house does not come with such a representation. Real-world route planning will mean the robot doing the hard part - creating that representation in the first place. Very far from solved.
Robot problems are not falling. Rather, the appreciation of how hard the problems are steadily grows. The dominant attitude of AI has, until recently, been characterised by this sort of wild optimism. General problem solvers a la Simon & Newell were thought to be just around the corner in the 50s because we solved chess. It is very interesting to re-read Hubert Dreyfus's scepticism of the 1970s state of AI because it still largely applies today.
I think a better marker of sophistication is needed, mice are relatively complex sure, but you'd need the combined intellect of say 11 or 12 of them to match the likes of Rolf Harris, to me the true marker of intelligence and versatility.
Dons flame retardant suit
2 problems with AI
Once you program the simplest of frameworks for your robot baby, ie don't let the power meter reach zero, EVER! and it's mission statement to build a copy of itself and protect that copy until it seems likely that IT will make a copy of itself on its own, what's left, except storage for all it's sensory inputs and outcomes of it's actions and the time it takes to achieve the mission statement?
That's if we want robots to be like humans of course.
Lucky for us that we seem to have an unlimited amount of storage, and foraging for fuel to keep the power meter from reaching zero is simple for the more fortunate of us. Apart from that, we learn from trial and error and nurture.
I've been asked before, "but what makes a baby feel loved?" Does it matter? Do robots have to feel loved in order to survive? Maybe they would evolve to become beings who understood love (as an advanced evolutionary tool). Did the first humans understand the concept of love? Did they even feel it? The drive was/is to survive and recreate. Love helps us do that, but isn't compulsory.
My point is I guess, that whenever I read about Artificial Intelligence, it seems to focus on details that seem irrelevant. The robot will never be human. How many years of evolution did it take for humans to have the luxury of considering ourselves great thinkers and indulging ourselves in love and war and art etc?
Seems to me, our baby robot only needs the simplest of instincts (ones we should choose very wisely) and some nurturing in order to grow up and develop the skills necessary to make a copy of itself. HOW HARD CAN IT BE?
Prepares fire extinguisher ;)
We don't know what intelligence is, but we'll know it when we replace it. We don't know how to extract route planning from TV camera vision, but we can do it (sorta) by adding additional senses. (Actually, I was thinking of the mars rovers new route planning software, which uses only image data).
A problem is solved when it stops being a problem. A robot doesn't have to solve a given problem the same way you do. It wouldn't be much of an invention if it did. It would just be a human made of silicon, and probably not a very good one. Robots can use sensors to solve problems that you solve with intelligence, and a gnat solves by flying.
A robot may never be properly intelligent, but if it can respond correctly when you say, "Fetch me some more champagne", or "Shoot anyone but me who tries to enter my walled compound", that will be enough. A robot doesn't have to learn like a college student. It can "learn" like a block of RAM, like a tree, or like the stock market.
Natural intelligence is very complex and generalized, perhaps because an organism doesn't choose the environment into which it is born. A robot is placed into an environment, so such generalization isn't always necessary. All my robot car has to do is obey laws, avoid obstacles and reach its destination. I'll get out and fill the gas tank. I'll decide where to go. Yeah, it's not as cool as an android servant who can drive, cook, and fight. But it's enough to put my (hypothetical, sadly) driver out of work, if not my butler and my bodyguard.
If we build robots, they will replace us. More and more of us every year. Apparently we're too stupid to live, even if robots are dumber.
What do we hope to achieve with ever more sophisticated AI? Robots that can interact with us socially? Any robot capable of this would surely be entitled to any rights that we are, and would therefore to all intents and purposes be equipped to make its own decisions on aspects such as occupation and lifestyle. Without definate criteria for intagibles such as setience and feeling where do you draw the line between design function and slavery? Any robot capable of say performing open heart surgery would need advanced sensory and decision making capabilty that would put it far above your average dolphin in terms of the sophistication ladder. Does this then mean that after a hard days work at the operating theatre, surgeonbot is entitled to go home, put his feet up and listen to some Bruce Hornsby whilst he recharges his hydrogen wotsit. Would you send a dog in to defuse a suspected explosive device? No, because they aren't equipped with the requisite tools for the job, in fact, barring hunting, your average dog isn't really equipped to do anything, yet we all feel sad when we hear reports of animal cruelty on the tv, because we feel that every living thing has certain rights. Imagine a very basic device, it has a simple electronic sensor inside that puts an LED on when it is struck or kicked; the LED is labelled "pain". Ok i know its a very vague association to make but when you take sophistication out of it, the mechanisms are the same. I know we're a long way from surgeonbot at the moment, but shouldn't we resolve these issues before we delve too deeply into the robot psyche!
Call of Duty...
"To break that logjam we need an Einstein of engineering."
I'm on my way.
2 (3?) things to make learning possible (maybe 4)
1.b) and its converse, pleasure (I actually think I do mean "converse" here)
3) forgetting: you must have the ability to discard unimportant or detrimental (to learning) information (and feedback).
And embarrassment (as a sort of self inflicted fear/pain) is an important part of the "learning to forget" feedback loop. Until then, the memory and microcode constraints will tend to mire the whole world of automatons in a sort of 1950s backwater of, "Computer, everything I say to you is a lie." nonsense.
Registering AI niche market creche....... Virtual Bourses
"The niche market for robotic replacement of humans is high-attention repetetive tasks, which humans do pretty well,..."
Actually, surely, the niche market for robotic replacement of humans is high-attention repetitive tasks, which humans do badly. And there is none, more in need of replacement, than Leadership.
But it is IT ReSearch and dDevelopment, well in hand and a lot further advanced than is currently imagined/documented....... but then again such is the Live Sensitivity of the ITs MkUltra UltiMate Programs ........ which an Axis of Evil would subvert to a Failed Pogrom, marking them out as not Fit for Leadership Purpose.
" Since I don't believe in the soul, I expect we'll figure out AI eventually. When we do, it will change society beyond recognition. Happily for those who prefer their humanity in the traditional "flesh and blood" form factor, right now no-one has the slightest idea how the brain does it."
I agree, Ken Hagan, but disagree that "right now no-one has the slightest idea how the brain does it." with that being simply your brain's response.
"Did the first humans understand the concept of love? Did they even feel it? The drive was/is to survive and recreate. Love helps us do that, but isn't compulsory." Hmmm...?
Some might tell you that, although not compulsory, it is necessary to to understand the concept of love to understand the drive to survive and recreate.
And how much further along that journey would modern man be, as opposed to first humans? Some might say pitifully pathetically they have not progressed at all, and in all too many cases, have even regressed.
""To break that logjam we need an Einstein of engineering."
I'm on my way." ...... With what, Rudiger?
Re: It's a philosophical problem
"Before we can recreate intelligence we need to know what it is, which is a problem for philosophers.
We maybe need a coding Kant before we can make the most of an engineering Einstein."
Bull. The problem is completely the opposite. There is TOO MUCH philosophy being done in the field of AI. Even the perceived need for formal structures with which to "comprehend" the data - referenced in the article - is part of the problem.
We do not come into this world with the paradigms of comprehension hard coded into our brains, and we shouldn't expect a human-like robot to do so either. These things are learned - no less so than the data that they organise.
Real progress in the field of AI will be completely hamstrung as long as it is dominated by people who engage in philosophical debates about poppycock like "what is a mind?", or "what does it mean to be self-aware?". When you look at what progress these debates have actually produced, it is painfully obvious that the emperor has no clothes.
Thankfully the field of fully genetic algorithms should soon grow past infancy into more maturity. There's a field that actually has a chance at producing something resembling truly functional "intelligence" (whatever that means!)
Re:Blue Screen of Death again?
Meanwhile, Apple are beavering away at producing a robot that has one button, cleans only the bathroom, uses a broom, doesn't quite have good enough sensors, so keeps on walking into things, is made of shiny white plastic, costs three times as much and needs to be scrapped for the new version every two years (or else the battery dies).
The Linux robot is incredibly efficient, cheap, very fast, and prevents your neighbours from using it with incredible security features, but you need to spend a month to get it to understand what your bathroom is, and it looks like a penguin.
The Dos robot fell over because its cache wasn't big enough.