A.I. and Robotics Take Another Wobbly Step Forward

CWmike writes to tell us that artificial intelligence and robotics have made another wobbly step forward with the most recent robot from Stanford. "Stair" is one of a new breed of robot that is trying to integrate learning, vision, navigation, manipulation, planning, reasoning, speech, and natural language processing. "It also marks a transition of AI from narrow, carefully defined domains to real-world situations in which systems learn to deal with complex data and adapt to uncertainty. AI has more or less followed the 'hype cycle' popularized by Gartner Inc.: Technologies perk along in the shadows for a few years, then burst on the scene in a blaze of hype. Then they fall into disrepute when they fail to deliver on extravagant promises, until they eventually rise to a level of solid accomplishment and acceptance."

Comments

New Algorithm ?

[daveime]

Bearing in mind that this new robot is called STAIR, does that mean it is using gradient descent algorithms ?

Re:New Algorithm ?

[Chris Burke]

Bearing in mind that this new robot is called STAIR, does that mean it is using gradient descent algorithms ?

Naw, it means that in a fit of irony, they named it after the one thing it couldn't handle.

Hype?

[AnthropomorphicRobot]

This article both points out the problems of over-hyped advances in robots, while also claiming this robot has transitioned away from narrowly defined domains?

The voice recognition & language processing component alone would be years ahead of anything else if it worked well outside of a "narrow, carefully defined domain". It seems like they are yet again over-hyping new research.

Re:Hype?

[sleeponthemic]

Robotics is a lot like graphical photorealism/virtual reality. People have been predicting it for decades, but the actual nuance of such an achievement is much more complex than most are able to comprehend. As such, we're perpetually surrounded by the hype.

Obligatory film reference....

[GPLDAN]

From the article: "Because these small [software] agents don't have a complete representation of the world, they are uncertain about their actions. So they learn to understand the probabilities of various things happening, they learn the preferences [of users] and costs of outcomes and, perhaps most important, they becoming self-aware."


I sure as hell hope they left out the lip-reading module.

Surprised it hasn't been said yet

[philspear]

I for one welcome our new learning, seeing, navigating, manipulating, planning, reasoning, speaking, and natural language processing Stair overlords.

Re:People perception

[Chabo]

Even my AI professor in school pointed to Data as really the end-goal of AI research (as well as a character from Battlestar Galactica, though I don't watch that show). I think many people are aware that modern AI has roughly the intelligence of an animal. That's much improved on AI from when the character Data was made, where the intelligence was more like that of a single-cell organism.

Of course, considerations must always be made for disaster...

I'm always amazed how broad a field AI really is; algorithms started in AI theory for moving robots around a room can be applied nearly everywhere.

Re:People perception

[sleeponthemic]

The generals population of AI is the Data, or Terminator. Some how superior to us humans who will not make mistakes.

Clever use of first sentence to invalidate second :-)

Other Robots

[troll8901]

Recent articles on robots.

• The Open Source Humanoid Robot and Its Many Uses (Sep 08, 2008)
• Inventor Builds Robot Wife (Dec 11, 2008)
• Denver Couple Unveils Homemade Service Robot (talks to bartender and fetches beer) (Idle) (Dec 22, 2008)
• The Best Robots of 2008 (Jan 19, 2009)

Yeah, but..

[karlwilson]

If you squeeze its chest, does it slap you? If not, they'll just have to take this one back to the drawing board. http://www.youtube.com/watch?v=3l6buDfU9AY

Re:People perception

[religious freak]

But being that a standard computer has the brain power of a bug, it isn't surprising that AI meets the hype.

I was thinking it was a clever use of the first part of the last sentence to invalidate the last part of the last sentence.

Surely there's money in this?

[Lexta]

As an undergraduate Computer Engineer who intends to major in AI I would be really interested in knowing if anyone knew whether the companies mentioned in this article (Google, Walmart etc...) actually do hire people with a specialization in AI? And if so how do u get recognized?? Or is it all just copied and pasted from researchers who work in all the Universities they mentioned?
I guess the root of my question is, by pursuing AI are you pushing yourself into becoming an academic for the rest of your life?

Re:Surely there's money in this?

[MichaelSmith]

I think AI is mainly used in spamming these days. Maybe William Gibson is right. It will be a crime to conspire to enhance an artificial intelligence.

Re:People perception

[im_thatoneguy]

It could even be argued that the ability to navigate a room is the same set of problem solving skills that informs all other intelligence.

It's spacial understanding combined with analyzing the capabilities of the agent to complete the task. Include a door and you have extremely complex problem solving and learning abilities.

Re:Yay!

[MRe_nl]

And we could call the offspring of STAIR and MASTER...
the enslavement machine!

Re:People perception

[Nigel Stepp]

Then again, some single cell organisms are pretty smart.

Seriously though, I don't think AI has yet reached the point of being as smart as your typical animal (which means low-level mammal I'm assuming). Not without substantial loans of intelligence on the part of the AI operator/designer.

Hyper Cycle

[computechnica]

We are still waiting for V.R. to reemerge from the Hype Cycle.

OK. I'll say it

[PPH]

Real Daleks don't climb stairs; they level the building.

Re:People perception

[sean4u]

You said something unpopular about AI. It's a good job there's no -1 sceptic modpoint, or I wouldn't even have seen your comment.

As far as I can see, AI has reached the point of being as smart as a snail that's really, really good at chess.

...if I've offended any snail slashdot readers, I apologise profusely.

The three types

[linhares]

Disclosure: I am a Hofstadterian, so I am biased here.

There are basically three types of AI-peoples: The neverlands, the hype masters, and the hope monks. The neverlands, like Searle, deny that intelligence is a product of information-processing. Searle has made it into a sport the claims that AI will never happen because it does not have the "causal powers of the brain".

Then there are these types, like those reported here. Hypeware at its best. Look, it's alive, it's (F*CKING GASP) becoming self aware, etc hype hype hype ad-nauseum. But look at its innards _very_ closely, and it's pretty empty in there.

There are so many pitfalls involved that it's impossible to mention all faulty premises involved in each project. But just for starters, consider this: when we program a machine to deal with the number 2, it usually goes into binary form 10 and there it stays, ready for manipulation. But how plausible is this psychologically? NOT AT ALL! When _we think_ of a "2", hordes of disparate, subliminar images come to mind, such as the gestalt of the digit, the sound of it, the fact that it's a prime (if you're math inclined), a couple (if you're a therapist), even-ness, odd-ness, the words "two" "too", and a huge number of semi-visible mental imagery.

Whenever you see a hyped AI project, just consider how it deals with the numeral 2. Most likely it's a _fake_. The process through which it goes through is not psychologically plausible. Which means that it will fail to understand human concepts.

Some projects with machine learning actually make it a habit of finding _meaning_ in highly correlated words (i.e., words that tend to occur together in documents). That is a _joke_. Meaning NEVER comes from correlation. If it did, "lawyer" and "telephone" would have much more to do than "lawyer" and "vampire", or "politician" and "scumbag".

Sorry for the rent, but I work hard to understand fucking hard issues and to see these folks being slashdotted with nothing to show for just begs for a rant. If you want to see really serious research, take a look a Douglas Hofstadter's "Fluid Concepts and Creative Analogies" and/or google for Kemp's MIT thesis.

Re:The three types

[Simon Brooke]

Disclosure: I am a Hofstadterian, so I am biased here.

There are basically three types of AI-peoples: The neverlands, the hype masters, and the hope monks. The neverlands, like Searle, deny that intelligence is a product of information-processing. Searle has made it into a sport the claims that AI will never happen because it does not have the "causal powers of the brain".

Then there are these types, like those reported here. Hypeware at its best. Look, it's alive, it's (F*CKING GASP) becoming self aware, etc hype hype hype ad-nauseum. But look at its innards _very_ closely, and it's pretty empty in there.

I think you'd find that if we could look into the functioning of the human mind in as much detail as we can look into an AI program, you'd find that it's pretty empty in there, too. I believe that intelligence is an emergent property of a lot of fairly simple processes. Yes, that's a matter of belief, not of proof.

There are so many pitfalls involved that it's impossible to mention all faulty premises involved in each project. But just for starters, consider this: when we program a machine to deal with the number 2, it usually goes into binary form 10 and there it stays, ready for manipulation. But how plausible is this psychologically? NOT AT ALL! When _we think_ of a "2", hordes of disparate, subliminar images come to mind, such as the gestalt of the digit, the sound of it, the fact that it's a prime (if you're math inclined), a couple (if you're a therapist), even-ness, odd-ness, the words "two" "too", and a huge number of semi-visible mental imagery.

Whenever you see a hyped AI project, just consider how it deals with the numeral 2. Most likely it's a _fake_. The process through which it goes through is not psychologically plausible. Which means that it will fail to understand human concepts.

A machine does not have to reproduce the mechanisms of the human mind in order to display intelligence; it has to emulate the performance. If the inputs are similar and the outputs are similar what happens in the middle is unimportant.

Re:The three types

[Simon Brooke]

A machine does not have to reproduce the mechanisms of the human mind in order to display intelligence; it has to emulate the performance. If the inputs are similar and the outputs are similar what happens in the middle is unimportant.

There is this general faulty reasoning that _understanding is the property of a representation_. That's just wrong. Just as temperature is not a property of molecules, understanding is not a property of a representation. It is a property of a process. In order to display the same "intelligent" behavior we do, machines have to go through the same process.

There is no ghost in the machine. The human brain is at best a Turing complete computing engine - at best, because we can prove that is not possible to be more. And we can prove that (modulo limited store, which is also an issue for human brains) our computers are also Turing complete. So it is not possible that our computers cannot do what a human brain can do - although admittedly we don't yet know how to program them to do it.

But we will find out, and when we do, I predict we'll look at the trivial little programs and say to ourselves 'is that really all?'

In the mean time I suggest to you that, apart from the purely academic interest of finding out how people tick, is isn't nearly as useful to program machines to do what people can do as to do what people can't do.

Re:solid accomplishment and acceptance?

[FiloEleven]

Seriously. We've been doing this HTML thing for what, 20 years now</a>? And still we can't accomplish the proper use of the closing tag. ;)

same stuff elsewhere

[sixtuslab]

There's a BBC documentary "Where's my robot" out since 16 dec 2008 with STAIR and others in it.