Robot Learning To Recognize Itself In Mirror

First time accepted submitter Thorodin writes in with a story at the BBC about scientists at Yale who have built a robot that they hope will be able to recognize itself in a mirror. "A robot named Nico could soon pass a landmark test - recognizing itself in a mirror. Such self-awareness would represent a step towards the ultimate goal of thinking robots. Nico, developed by computer scientists at Yale University, will take the test in the coming months. The ultimate aim is for Nico to use a mirror to interpret objects around it, in the same way as humans use a rear-view mirror to look for cars. 'It is a spatial reasoning task for the robot to understand that its arm is on it not on the other side of the mirror,' Justin Hart, the PhD student leading the research told BBC News. So far the robot has been programmed to recognize a reflection of its arm, but ultimately Mr Hart wants it to pass the "full mirror test". The so-called mirror test was originally developed in 1970 and has become the classic test of self-awareness."

Comments

Self-Awareness

[guttentag]

When it can tell the difference between a human and a metallic exoskeleton with glowing red eyes, it's time to pull the plug. And put on your 1,000,000 SPF sunscreen.

Didn't QBO do this already?

[Anonymous Coward]

QBO Robot in front of a mirror.

Laugh

[koan]

It isn't "self awareness" there is no true AI.

Re:Laugh

[Kell Bengal]

Actually, it's more about recognising the auto-motion structure in the scene. I'm familiar with Justin's work (Go Team Scazlab!) and it's a lot deeper and more interesting than the article gives it credit for.

AI claims from the 70s ruined a generation of people for machine intelligence (which is why we now have to sell it as 'machine intelligence' or 'machine learning'). Knowing what part of the camera scene is moving because something is happening, and knowing what part of the scene is moving because you're waving your end-effector is useful. If you can extract your own state from indicators in the environment, then you have more information to work with - that's why we use a mirror to do our hair and straighten our ties.

Well... those of us that wear ties...

Re:Laugh

[Anonymous Coward]

What makes you think your DNA wasn't a complex self-executing program?
Hint, it is.

And it has had millions of generations of evolution behind it that has resulted in useful "code" being the baseline of what makes it a human, makes it breathe, speak and type silly things on Slashdot.

Re:Laugh

[timeOday]

It isn't "self awareness" there is no true AI.

Hard to disagree with that logic. Thanks for settling the issue.

Re:Laugh

[VortexCortex]

which is why we now have to sell it as 'machine intelligence' or 'machine learning

I actually prefer the term Machine Intelligence to Artificial Intelligence. There is nothing artificial about a neural network's intelligence. The network may be artificial (man made, or existing as a simulation), but the degree of intelligence is not artificial; It's a function of the network's complexity. Intelligence emerges due to the properties complex interactions naturally have.

Cars do not create Artificial Movement. Machine Learning does not create Artificial Knowledge. Machine Intelligence does not provide Artificial Intelligence, it simply yields a measure of intelligence. A house fly, dog, or penguin doesn't have as complex a neural network as you likely do, but this does not make them Artificially Intelligent simply because their degree of intellect and awareness is less than your own. When we train the lesser minds to communicate with us, and perform tasks, they are not artificially performing the tasks.

I find the term A.I. to be racist, and indicative of the chauvinistic attitude some humans have about their own mental prowess -- Your brains are not special. Any sufficiently complex interaction is indistinguishable from sentience, because that IS what sentience is. Once cybernetic systems attain (and surpass) the level of complexity present in humans brains, Artificial Intelligence will be a derogatory term: "Oh you pass yourself off as being smart, but you're just Artificially Intelligent -- You don't actually understand anything!"

Also: Not that it matters, but I don't personally believe that a god created the race of men. However, some do consider this to be true, and yet they do not call themselves Artificial Life...

Re:Laugh

[Baloroth]

I find the term A.I. to be racist, and indicative of the chauvinistic attitude some humans have about their own mental prowess -- Your brains are not special. Any sufficiently complex interaction is indistinguishable from sentience, because that IS what sentience is. Once cybernetic systems attain (and surpass) the level of complexity present in humans brains, Artificial Intelligence will be a derogatory term: "Oh you pass yourself off as being smart, but you're just Artificially Intelligent -- You don't actually understand anything!"

First of all, thats a pretty horrible misuse of the term "racist", and second, the term "artificial" means, by definition, created through art (art here being the broad sense as any product of human activity, rather than the fine arts): i.e. created by human intention and design. By definition "Machine Intelligence" is "Artificial Intelligence", at least so far as we have created it. That intelligence is designed and a product of human work. It's intended, and is brought about not because of some emergent behavior found naturally in existence, but because humans arranged it that way and brought it about. That's not in any way racist, it's just the meaning of the words.

It would be completely irrational and contradictory to the very meaning of the term to call humans "artificial life", since we were not created by human art. You'd destroy the meaning of the words to call humans "artificial", just as we wouldn't call the sun "artificial" even if you said it was created by a god, since that's not what the word means. Long story short, you are trying to destroy the meaning of words. Don't do that: it's bad for everyone.

Re:Laugh

[Tarlus]

Even the trolling AC's are becoming self-aware. Fascinating...

Re:Laugh

[grouchomarxist]

Note that part of the problem in the discussion here is that the word "artificial" has several meanings. The grandparent is referring to one meaning which is something like "contrived or false", while the parent is referring to a related meaning which is "made or produced by human beings rather than occurring naturally, typically as a copy of something natural". I think both of these meanings of artificial are in play when we talk about AI, although the meaning used by the parent is probably the more fitting for the context.

Re:Laugh

[TheMathemagician]

No they're not really self-aware, it's just a Chinese Room.

Re:Laugh

[michelcolman]

It would be completely irrational and contradictory to the very meaning of the term to call humans "artificial life", since we were not created by human art.

You mean sex is not human art? You haven't been doing it right, then.

Re:Laugh

[TapeCutter]

Intelligence emerges due to the properties complex interactions naturally have.

Precisely, further, "intelligence" is in the eye of the beholder. My favorite example is an ants nest, each individual ant follows some very simple rules, so simple it doesn't need a brain to carry them out, it's nervous system alone provides enough "intelligence". The ant and the neuron both display automata like behavior that can be expressed as a state machine. Ants and neurons live in colonies (nests and brains), unlike the individuals the colonies do display what most people would call "intelligent behavior", yet nests and brains are also just state machines all the way down.

Your brains are not special.

My brain uses it's knowledge to inform itself that it will cease to exist, but deep down in the brain stem it's not really buying it's own story. And it's certainly not buying the idea it's not unique or special. I think programmers can see the idea that the human brain could be expressed as a state machine more readily than most because they are in the business of producing intelligent behavior from simple rules. However don't underestimate the impact that a deeply rooted acceptance of ones own morality can have (meat starts @ 3:55), non-existence is a fear that comes from the brain stem, it's the emotional driver for the "fight or flight" response. All humans recoil instinctively from the idea like ants instinctively find the sugar bowl. The existential question can be a deep dark rabbit hole with some side routes leading to depression and insanity. Of course if you can avoid (or get past ) all that, you may eventually lose the fear of not knowing, the moment of genuine acceptance is an experience many have described as "religious" - as in the natural buzz one gets from surviving "a leap of faith".

Disclaimer: I've been an atheist since my mum quit teaching Sunday school in the mid 60's and started reading me Aboriginal dream time stories, Greek fables, etc, as "stories that some people think are real". In my late teens I was sucked in bad by Uri-Geller for a couple of years. He fixed my broken watch, it didn't matter that he did it by staring at the TV with a face like a constipation sufferer, the proof was right there, the watch ran for days!!! A couple of years later I had a book shelf jammed full of "alternative science". James Randi set me straight on the real meaning of skepticism in his (short) 1980 book debunking Geller (that's HS science for you, both then and now), ironically I had picked up Randi's book from the bargain bin because I thought I knew enough to easily debunk it, in one night he had convincingly debunked my entire bookshelf.

Later still dad confessed to winding the watch with tweezers while I wasn't looking.

Re:Laugh

[koan]

"The experiment is the centerpiece of Searle's Chinese room argument which holds that a program cannot give a computer a "mind", "understanding" or "consciousness",[a] regardless of how intelligently it may make it behave. The argument is directed against the philosophical positions of functionalism and computationalism,[2] which hold that the mind may be viewed as an information processing system operating on formal symbols. Although it was originally presented in reaction to the statements of artificial intelligence researchers, it is not an argument against the goals of AI research, because it does not limit the amount of intelligence a machine can display.[3] The argument applies only to digital computers and does not apply to machines in general.[4]"

Re:Laugh

[Kell Bengal]

If you disagree with Justin and Prof Scassellati's approach, I'd like to hear your thoughts as to how you'd solve the problem differently. If you're familiar enough with their work so as to dismiss it in such straight-forward terms, I presume you can provide specific criticisms that will help them improve their work?

(And yes, I know both Justin and Scaz personally, and I volunteered for one of their social robot-interaction studies.)

nobody's said it yet?

[jehan60188]

robot overlords, etc

what's the physical robot for ?

[cathector]

seems like a physical simulation and a renderer could get the same job done.

hm. i guess the challenge must be in getting it to happen in realtime w/ portable hardware.

Oh, great.

[Farmer Tim]

As if kernel panics weren't enough, now my computer will be able to get depressed over its body image too.

Re:Oh, great.

[Farmer Tim]

Helium is indeed a noble gas.

Mirror, mirror, on the wall...

[macraig]

My mirror tests me every morning now. Incidentally I fail every morning. Tomorrow I'm gonna try wearing a Guy Fawkes mask to see what happens.

And its first words during the experiment were...

[sidthegeek]

"Do these bolts make me look fat?"

if a robot thinks in the wilderness...

[globaljustin]

"a step towards the ultimate goal of thinking robots"

**sigh** I thought we were past this stuff, even in mainstream media...."Thinking robots" is not a coherent concept or benchmark that can be accomplished.

"thinking robots"....most people mean 'artificial intelligence' when they use these words, but the idea of AI as independent thought is irrational. It is all programmed responses at some level. Even machines that are programed to process new data into existing algorythms for feedback processing are **still** doing that 'learning' according to a human-programed way of processing and integrating data...its all just machines executing complex instructions at the core!

Commander Data...some people contextualize "thinking robots" as a technical level at which a machine is so like beings with Sapience that it is immoral to deny them the rights of a humanoid. This is science fiction. It is helpful, but it is a scenario based in a world with several assumptions. Its not fit to apply to computing directly. We do not know how the human mind ultimately works...unless we have that, then there is nothing to accurately compare a non-human brain to consistently.

Ultimately, if neuroscience and AI converge, meaning we can map every thought in the human brain **AND** have the technical ability to construct an artificial system that enables what we know as 'free will' and 'thought' and 'choice' and especially 'self awareness'....THEN and ONLY THEN have we made something...

And what have we thus made? IMHO, its a **new** third thing. Not human, but at least equal to human and bound within the same social contract all humans are bound to.

Re:Training to the test

[camperdave]

When you teach to the test, what exactly has the student learned?

Whatever is on the test... Duh!

When you're training a neural net, you feed it positives, then you refine that with exceptions. It's like when the FBI trains people in counterfeit detecting. They spend the bulk of their training with real money. They learn the feel. They learn the security features. They learn the artwork, and the quirks, and the smell. They don't see counterfeit money until they are well and truly acquainted with what the real thing looks like, and how it behaves.

pigeons have been taught to do this already

[0-9a-zA-Z_. !*'()123]

and no explanation in terms of self-awareness was used to explain it:

Citation:
https://www.sciencemag.org/content/212/4495/695.short

Full:
http://drrobertepstein.com/downloads/Epstein-Self_Awareness_in_the_Pigeon-Science-1981.pdf

So now robots can do what pigeons can do. Self-awareness is a hypothetical construct http://psychclassics.yorku.ca/Skinner/Theories/ which may not be very useful.