Self-aware machines won't be science fiction for much longer

[Shawn Knight]

In brief: Humans are unique in that we are self-aware and can envision ourselves in future scenarios – daydreaming about winning the lottery or starting a family. It’s an advantage we have over machines as most robots still use laborious trial and error to learn about the world around them and their place in it.

Self-aware machines have been the topic of science fiction for decades but thanks to a recent advancement from Columbia University, they’re now becoming reality.

Researchers at the school have created a robot capable of learning what it is with zero prior knowledge of geometry, physics or motor dynamics. It starts out “babbling” but after about 35 hours of collecting movement data and training it with machine learning, it created a self-simulation that can be used to adapt to situations, learn about itself and even detect damaged components and work around them.

Hod Lipson, professor of mechanical engineering and director of the Creative Machines lab where the research was done, said that if we want robots to become independent, then it’s essential that they learn to simulate themselves.

Lipson and his PhD student, Robert Kwiatkowski, are fully aware of the potential ethical implications associated with their creation. “Self-awareness will lead to more resilient and adaptive systems, but also implies some loss of control,” they said. “It’s a powerful technology, but it should be handled with care.”

If that isn’t frightening enough, the researchers are now looking into whether or not robots can model their own minds – that is, if they can think about thinking.

Lead image courtesy Phonlamai Photo via Shutterstock

Permalink to story.

https://www.techspot.com/news/78519-self-aware-machines-wont-science-fiction-much-longer.html

 

[Uncle Al]

Creating a device that can go through any number of decision sub-routines is not "self aware". The topic has been covered any number of times and the conclusion is always the same ....

 

[dirtyferret]

Fantastic! ...someone else to ask us "what are you thinking about?"

 

[mbrowne5061]

This isn't new, not the way Techspot describes it at least. MIT's CSAIL lab did something something similar a few years ago (a robot that was able to figure out what 'it' was in terms of geometry, range of motion, etc)

 

[VitalyT]

whether or not robots can model their own minds – that is, if they can think about thinking.

Thinking in itself does not denote any limitation on the subject, which makes thinking about thinking an empty expression.

 

[DelJo63]

Using ECC techniques, a machine can diagnose itself and pinpoint precisely which card has failed and even where it is located.

Self-awareness is not "oh gee, I've got a broken part", but ever so much involved in the current emotional state of the entity -- My feelings are hurt -- and the machine as no emotions does it Mr Spock!

 

[Reachable]

Yes, I suppose it's accurate to say that the machine is self-aware, but it's an accomplishment that's profoundly minute, not to be confused with human self-awareness.

We are all born without self-awareness, but after a year or two of constant attention by many people to every aspect of our staggeringly complex physical, mental and emotional being, we develop an awareness of being a "person" and all that that involves, including relationships to other people.

This machine is nothing but a toy, a collection of gears and levers and wires and circuits. There is certainly an accomplishment to having it be able to develop a coordinated system that collects its entirety into a single entity and to function on that basis. I suppose that's self-awareness, but it's toy self-awareness. The scientists are engaging in a little overenthusiastic hyperbole if they make people think that the machine regards itself as a person. It doesn't.

 

[wiyosaya]

I suppose that's self-awareness, but it's toy self-awareness. The scientists are engaging in a little overenthusiastic hyperbole if they make people think that the machine regards itself as a person. It doesn't.

Agreed! Its unfortunate that some scientists think this much of their own work. To me, it takes the science they practice outside of the realm of credibility. Yes, the machine can figure out what it can do, but like you said, what the machine has accomplished is far from anything that even remotely approaches human self-awareness.

 

[Underdog]

Most modern cars can self diagnose faults. This doesn't make them self aware. Genuine self awareness would be like a sex doll with AI saying NO because it has a headache

 

[JaredTheDragon]

It's just marketing hype. Tech is no closer to AI than it was 50 years ago, or 50,000. Machine learning isn't intelligence.

 

[wiyosaya]

Most modern cars can self diagnose faults. This doesn't make them self aware. Genuine self awareness would be like a sex doll with AI saying NO because it has a headache

IMO, the sex doll topic is creepy to begin with, but having an AI sex doll say NO because it has a headache, well that would be extremely creepy.

 

[Aux101]

The video only demonstrates one thing we already knew: self-learning is working!
Now, to say that self-learning is the same as self-awareness is... well... totally wrong.

The code behind the magic could be as simple as this:
1-Goal is to 'get the ball' {GET is restriction bigger than 1mm}
2-Is this move good{+1mm} or bad{0mm}? If good then A; If bad then B
3-Choose A over B {memorize 'Move'}
4-{write down sequence A1+A2+...}
5-Repeat
: /

 

[PcePce]

It's funny how scientists are going out of their way to do experiments which produce clickbaity headlines because it creates a financial incentive to fund their research. It might not mean much to the progress of science, but it does generate interest and may result in some useful technologies.

This article's headline is pure BS, though, unless you willfully interpret "self-aware" as broadly as possible.

 

[redgarl]

It's just marketing hype. Tech is no closer to AI than it was 50 years ago, or 50,000. Machine learning isn't intelligence.

ML and DL are not AI, I totally agree with you. Only people not familiar with the field think they are the same.

 

[quadibloc]

The fact that a computer program in a machine includes a model of that machine, even if that model is created naturally by a learning method, doesn't mean that the machine actually feels or experiences what it is doing - and that it matters ethically how we treat the machine, because it has feelings. A machine could do what the article describes and still have no more feelings than an adding machine. Consciousness is still a mystery to science.