Robot Demonstrates Self Awareness, appearantly

[wolveraptor]

A few years ago I think there was a story of a robot in a compound running away, and that the programmers (after they caught it) could only explain its behaviour as that of someone trying to escape captivity.

I'll try and find it, but I find this story very interesting.

This seems much more interesting than the the current article of this thread. Wow, a robot can tell if it's looking at itself after being designed to do so? I'm not impressed. Or at least, I'm not awed at the awesome mental capacities of this creation. It's still a pretty cool doohicky. Still, one wouldn't need to ascribe the action to self-awareness. It could just be reacting to the presence of a sheet of glass.

[Dooey Jo]

Hardly. The robot is designed to acknowledge the behavior of other robots, and imitate said behavior. The different colored LEDs flash depending on whether the robot thinks its imitating what its seeing, or actually recognizes that what its seeing is just a reflection of itself.

How else would it recognise itself in a mirror than from concluding that the mirror image is doing the same thing itself is doing at the same time? It probably checks how different the behaviour it sees is from its own behaviour and from that decides wether or not it actually is itself (that's not exactly how neural networks work but it probably compares the pattern of the firing recognition-neurons to the movement-neurons and maybe runs them as input on others, the output of which is linked to the blue LED, which in turn lights up if the movement is similar but not perfect). So yes, I still think you could fool it with a robot that is much better at imitating than the self-robot is.

It would be the same thing as plugging an interface into a cat or dog that shows whether its brain thinks that what it sees/hears is another cat/dog, or the cat/dog itself, seen from a strange point of view.

Would it? Cats and dogs are very different from this robot, not just because of the LED interface. They might not have any means at all to realise that they're seeing themselves, but this robot does. In fact it has to, otherwise it would imitate itself forever if it saw a mirror, whereas cats and dogs don't always mimic everything they see so they might not need such a mechanism...

[Xenophobe3691]

How else would it recognise itself in a mirror than from concluding that the mirror image is doing the same thing itself is doing at the same time? It probably checks how different the behaviour it sees is from its own behaviour and from that decides wether or not it actually is itself (that's not exactly how neural networks work but it probably compares the pattern of the firing recognition-neurons to the movement-neurons and maybe runs them as input on others, the output of which is linked to the blue LED, which in turn lights up if the movement is similar but not perfect). So yes, I still think you could fool it with a robot that is much better at imitating than the self-robot is.

You can fool people with that as well. Ever seen one of those shots that pans out and turns out to have been facing the mirror the whole time? We're not trying to fool it, we're trying to see if it recognizes the image in the mirror as itself, and we're trying to do it in as similar a fashion as possible to how we test living creatures.

Would it? Cats and dogs are very different from this robot, not just because of the LED interface. They might not have any means at all to realise that they're seeing themselves, but this robot does. In fact it has to, otherwise it would imitate itself forever if it saw a mirror, whereas cats and dogs don't always mimic everything they see so they might not need such a mechanism...

In the experiment, the robot is what's acting, and its reflection copies. The robot we're testing isn't reacting in any way.

[Dooey Jo]

You can fool people with that as well. Ever seen one of those shots that pans out and turns out to have been facing the mirror the whole time? We're not trying to fool it, we're trying to see if it recognizes the image in the mirror as itself, and we're trying to do it in as similar a fashion as possible to how we test living creatures.

I'm sure that's true. However, it still isn't a matter of having damn good sight, just damn good pattern recognition abilities (or was that impled when you wrote "sight"?).

In the experiment, the robot is what's acting, and its reflection copies. The robot we're testing isn't reacting in any way.

I see. Well, future version are surely supposed to be reacting as well? That's what I figured the technology would be for; so that it wouldn't mimic things that already mimics what it's doing...

[Xenophobe3691]

I'm sure that's true. However, it still isn't a matter of having damn good sight, just damn good pattern recognition abilities (or was that impled when you wrote "sight"?).

It was just an expression, "It's a damn sight better/worse/insert adjective here".

I see. Well, future version are surely supposed to be reacting as well? That's what I figured the technology would be for; so that it wouldn't mimic things that already mimics what it's doing...

I think the future versions are supposed to be able to tell itself from another robot, or from a person. I'm not sure, though.

[Qwerty 42]

It does seem to be a bit of a "jury-rigged" experiment (see the banning thread about animal training for an example,) but I'll trust the scientists judgment on this one in that there is a difference.

It's still not sentient, but as mentioned, it's not pocket change either.

[ArmorPierce]

If it's able to recognize its surroundings and act accordingly, I'd say that it is sentient.