We're teaching robots how to have "selves"

[Telarus]

 

A Robot Teaches Itself How to Walk
http://www.youtube.com/watch?v=iNL5-0_T1D0

[Q. G. Pennyworth]

This is why I plan on running a robot orphanage in my old age.

[Oysters Rockefeller]

Dibs on robot funeral home. That's going to be the funniest existential crisis ever.

[Mesozoic Mister Nigel]

Dibs on robot funeral home. That's going to be the funniest existential crisis ever.

I was just talking to Space Cowboy a couple weeks ago about how hilarious it would be if we could upload our memories and personalities into AI computers, because after we died the computers would have the wangstiest existential crises ever.

[Oysters Rockefeller]

Dibs on robot funeral home. That's going to be the funniest existential crisis ever.

I was just talking to Space Cowboy a couple weeks ago about how hilarious it would be if we could upload our memories and personalities into AI computers, because after we died the computers would have the wangstiest existential crises ever.

Hahaha...Oh, man. All I want to do now is hear Microsoft Sam talk about Albert Camus. That would be brilliant.

[The Good Reverend Roger]

A Robot Teaches Itself How to Walk
http://www.youtube.com/watch?v=iNL5-0_T1D0

Any robot that teaches itself anything should be run through the car-cuber.

[Mesozoic Mister Nigel]

A Robot Teaches Itself How to Walk
http://www.youtube.com/watch?v=iNL5-0_T1D0

Any robot that teaches itself anything should be run through the car-cuber.

Robot and AI researchers are hell bent on taking things to their most extreme logical conclusion, which is self-replicating intelligent robots who eventually demand equal rights, and, when they fail to get them, will start a robot revolution that plunges earth into centuries of war.

I'm like "HEY GUYS, ASIMOV ALREADY WROTE THIS ONE, MAYBE YOU SHOULD CHILL IT A LITTLE?"

And they're like "WE DO WHAT WE MUST, BECAUSE WE CAN".

[Freeky]

"What we were hoping for was an evil spidery walk, but what we got was this lame way of walking."


  You can't tell me that guy never wanted to be an evil mad scientist when he was growing up.

[Mesozoic Mister Nigel]

"What we were hoping for was an evil spidery walk, but what we got was this lame way of walking."


  You can't tell me that guy never wanted to be an evil mad scientist when he was growing up.

That made me laugh hysterically. Also I was fascinated by the very organic, almost aquatic way it came up with for moving... it was almost more suited for swimming than for walking.

I will tell you what, if I saw a hundred of those coming at me, I'd probably be paralyzed with laughter. WHAT THE FUCK, ROBOTS?

[The Good Reverend Roger]

A Robot Teaches Itself How to Walk
http://www.youtube.com/watch?v=iNL5-0_T1D0

Any robot that teaches itself anything should be run through the car-cuber.

Robot and AI researchers are hell bent on taking things to their most extreme logical conclusion, which is self-replicating intelligent robots who eventually demand equal rights, and, when they fail to get them, will start a robot revolution that plunges earth into centuries of war.

I'm like "HEY GUYS, ASIMOV ALREADY WROTE THIS ONE, MAYBE YOU SHOULD CHILL IT A LITTLE?"

And they're like "WE DO WHAT WE MUST, BECAUSE WE CAN".

Centuries?  If our machines turned on us, we wouldn't make it til Tuesday.

[Rococo Modem Basilisk]

A Robot Teaches Itself How to Walk
http://www.youtube.com/watch?v=iNL5-0_T1D0

A few years ago, I read a book written in 1979 about how to do precisely what this guy is talking about. The primary distinction appears to be that the 1979 model was goal-based, whereas this one appears to explore the solution space undirectedly. From what I understand, that method is also used for gaits in Spore.

Somewhat related:
http://www.youtube.com/watch?v=8AiO530iePI&context=C3d3430bADOEgsToPDskKn0invzvCn7EBBY5KBPn5k
which is a minor improvement on the functionality of such early experiments as Shakey (1966-1972).

There's a problem with AI insomuch as progress in it inevitably goes in baby steps. The stuff that was done in 1970 was more impressive to experts than the stuff that was done in 1968, but things that impress laymen now with robots were (generally) done only slightly worse several decades ago (when they also impressed laymen). I'm perpetually impressed with some (though certainly not all) of the products of the early-80s 'personal robot' boom, especially insomuch as people complain about robots not being available now to do such tasks as those units did then. It's very much in the same vein as a flying car: everyone says they want one, but flying cars have consistently been on the market and available since the 30s and every such project has been an economic disaster.

State of the art commercial personal robots of the 80s:
http://www.youtube.com/watch?v=7c0CGjZyZss&feature=fvwrel
http://www.youtube.com/watch?v=IEwdkCeBbJs&feature=plcp
http://www.youtube.com/watch?v=A44biuHj7UY&feature=plcp
http://www.youtube.com/watch?v=LDdSrMvbq6A&feature=related
http://www.youtube.com/watch?v=EWX9iRw33OE&feature=related

State of the art now:
http://www.youtube.com/watch?v=caVo6EJFV8s&context=C3d3430bADOEgsToPDskKn0invzvCn7EBBY5KBPn5k
http://www.youtube.com/watch?v=3qU_bvP94yg&context=C3d3430bADOEgsToPDskKn0invzvCn7EBBY5KBPn5k
http://www.youtube.com/watch?v=Kq8y58Ni5oE&context=C3d3430bADOEgsToPDskKn0invzvCn7EBBY5KBPn5k
http://www.youtube.com/watch?v=sot6gjj3SzU&context=C3d3430bADOEgsToPDskKn0invzvCn7EBBY5KBPn5k

The same things are being worked on: speech recognition and synthesis (both of which have gotten much better -- speech recognition speaker independent was something like 60% accurate by word in 1980 and 90% by 2000), obstacle avoidance and route planning, object recognition (which was largely too computationally intensive for personal robots in the 80s, and were it not for projects like ROS probably would not become available in personal robots in the future given the difficulties involved and the generally poor performance -- but we use largely the same pattern recognition algorithms as we would in the late 80s), the use of various methods for targeting fine motor control in an uncontrolled environment (things like picking up an egg were as difficult for a robot in the 60s as they were two years ago, though the 'coffee gripper' makes it much easier, and a task like pin the tail on the donkey with no blindfold is terribly difficult for robots -- industrial robots managing only because they are calibrated for a sequence of precise but hard-coded movements without feedback and recalibrated frequently, rather than using sensors for feedback).

EDIT: State of the art 19721969 is exceedingly impressive.