Thoughts on Artificial Intelligence

[David]

I just read the two articles on AI on Wait But Why and it's pretty interesting and a bit scary.

Here's the first one: The AI Revolution: Road to Superintelligence - Wait But Why

It's a long read but sure gets you thinking.

The premise is there are 3 levels of AI: Artificial Narrow Intelligence (ANI) which are things like Google, Spam Filters, etc., stuff we already have; Artificial General Intelligence (AGI) which would be on par with human intelligence, and Artificial Super Intelligence (ASI) which would be smarter than humans on about the same range we're smarter than ants.

Experts in the field expect us to reach AGI in this century, perhaps in the first half of this century. The scary part is a computer with AGI could potentially increase its intelligence so that it reaches ASI within a few years, days, or even hours of reaching AGI. And an ASI computer may not see much use for humans after that and have no moral reason to keep us. Or it may allow for humans to be immortal. Pretty interesting stuff.

[MBAtarga]

skynet....

[scottmandue]

We were looking at some videos here at work of a the robot dog running up and down hill... even climbing stairs... the tech on robots is really moving.

A.I. however is another can of worms... they have some very, very clever machines, but they still are just machines.

[JD159]

Originally Posted by MBAtarga

skynet....

When I used to work for Futureshop / Best Buy, a thing called Skynet controlled the scheduling and heat of the building...

Not even kidding...

[Z-man]

In theory, yes - it is pretty interesting stuff. But practically speaking, an AI computer is essentially a relational database with computational algorithms that can make connections via the database and generates output based on certain criteria. Yes, a database can grow (provided there is sufficient storage available) with more input, but the computational algorithms can be a bit difficult to grow autonomously to the point where increased intelligence (or rather improved algorithms) can be measured. Therein lies the rub.

Too many people confuse the issue by perceiving a computer mimicking human behavior as intelligence. The computer is only following a set of algorithms, which is not what intelligence is.

I would not worry about a Terminator knocking on your door just yet...

-Z

[930addict]

Meh. If the computers get out of line just unplug them. This also works for windows os. Haha.

[vash]

this week? i'd be happy to see some "Actual" intelligence.

[motion]

I want a computer that does MI. Money Intelligence.

[ckelly78z]

Sarah Conners currently is doing military training in the desert, and maybe would should as well !

[GH85Carrera]

The entire AI bit is interesting.

People are legitimately concerned about computers and machines taking over. They all run on electricity or HAVE to have electrical connections to make the computer functional.

We can always unplug them, or turn off the electricity.

Terminator will never happen until there is a several thousand orders of magnitude improved power source developed. It defies physics in several way to have that much power in a small power source. In the movies the power cell used to run him would fit in the palm of your hand and power him through feats that would require several large generators with a long power cord and many gallons of gasoline.

[rcooled]

[David]

The comments about unplugging them and they're just machines are covered in the articles. Of course it seems easy to stop but the problem is we're only seeing the issue with human intelligence, just like a monkey can understand that a building exists but can't comprehend that it was built by a human. We can't even fathom what a super intelligent being or computer could invent or do.

[GH85Carrera]

In the book, HAL functioned as designed. He was programmed and told flat out, nothing is to stop him from investigating the monolith. He just followed the instructions he was given. Just like James Bond, he had a license to kill and a mission to accomplish.

[stomachmonkey]

Originally Posted by Z-man

In theory, yes - it is pretty interesting stuff. But practically speaking, an AI computer is essentially a relational database with computational algorithms that can make connections via the database and generates output based on certain criteria. Yes, a database can grow (provided there is sufficient storage available) with more input, but the computational algorithms can be a bit difficult to grow autonomously to the point where increased intelligence (or rather improved algorithms) can be measured. Therein lies the rub.

Too many people confuse the issue by perceiving a computer mimicking human behavior as intelligence. The computer is only following a set of algorithms, which is not what intelligence is.

I would not worry about a Terminator knocking on your door just yet...

-Z

Thanks for saving me some typing.

The difference in sentient intelligence and AI is sentient beings brains are adaptable and to some degree self wire, Neuroplasticity.

Think of your brain as a CPU. Adaptable, tunable.

[Crowbob]

Originally Posted by Z-man

In theory, yes - it is pretty interesting stuff. But practically speaking, an AI computer is essentially a relational database with computational algorithms that can make connections via the database and generates output based on certain criteria. Yes, a database can grow (provided there is sufficient storage available) with more input, but the computational algorithms can be a bit difficult to grow autonomously to the point where increased intelligence (or rather improved algorithms) can be measured. Therein lies the rub.

Too many people confuse the issue by perceiving a computer mimicking human behavior as intelligence. The computer is only following a set of algorithms, which is not what intelligence is.

I would not worry about a Terminator knocking on your door just yet...

-Z

Seems like there's allot of confused human behavior perceived by people as mimicking intelligence, too. But as far as the definition of intelligence goes, has anyone determined that intelligence isn't following a set of algorithms? Couldn't learning be described as following a set of algorithms with increasing efficiency?

I also wonder at what point mimicry becomes authentically-learned behavior. Seems like at some point it becomes a semantics issue. Some machine winning at Jeopardy sure looks like intelligence and acts like intelligence to me. At one point memory was considered to be integral to human intelligence. Now with a capacity for gazilliobits of memory in machines, the definition of human intelligence had to change.

[Z-man]

Originally Posted by Crowbob

Seems like there's allot of confused human behavior perceived by people as mimicking intelligence, too. But as far as the definition of intelligence goes, has anyone determined that intelligence isn't following a set of algorithms? Couldn't learning be described as following a set of algorithms with increasing efficiency?

I also wonder at what point mimicry becomes authentically-learned behavior. Seems like at some point it becomes a semantics issue. Some machine winning at Jeopardy sure looks like intelligence and acts like intelligence to me. At one point memory was considered to be integral to human intelligence. Now with a capacity for gazilliobits of memory in machines, the definition of human intelligence had to change.

Memory (ie the capacity to store data) is only the starting point of intelligence. What actions are done based on the data is the next step, and the step after that is improving the actions. The first two - a computer can do. The third one -- improving the action (ie: cognitive learning) is where computers are inefficient in many respects.

Memory is a very interesting concept, and there are two main schools of thought on it -- take a simple object like a chair. In your mind, you can easily picture what a chair is. But what is stored in your mind? Is it a collection of all types of chairs that you have seen? Or is it more of a relational memory - you understand a chair to be something with 3-4 legs, a platform to sit on, and an optional backrest and armrests. Computers can be programmed to store memory in both ways: the former takes up more memory (storage), but the latter requires more computations to arrive at some conclusion.

So - let's say an AI robot is able to recognize a chair. Fine. And he is programmed to sit in a chair once he finds one. That is the total extent of his knowledge -- once he sees a chair, he sits in the chair. Great - this can be easily accomplished even today. However - it would be very difficult for this robot to learn other things: what if the chair is on its side? How does the robot sit in the chair then? If he is programmed to maneuver his body to the chair's location, then he will likely lay down and align his body to the chair, rather than stand the chair upright and then sit on the chair. What if the chair is in pieces? Unless programmed to recognize the pieces and assemble the chair, the robot would not be able to construct a chair. What if there is no chair? How would the robot learn to sit instead on a ledge, table, or stoop? These are types of simple intelligence that are not trivial to solve.

-Z-man.

[Crowbob]

Great discussion, thank you Z.

What you are describing are the limitations we currently face. However, I am reminded that not long ago it was pretty much definitive that no machine could produce beautiful and novel art. Say, for discussion, a painting that satisfies our current notions of esthetics, even as individual and subjective as esthetics are.

Today, a drone could easily and remotely take a photo of a landscape, a woman or even a chair. Connect that image up to a printer and voyla an esthetically pleasing, novel image is created. Were one to show that image to someone from a century ago that person quite likely would not be able to comprehend that that beautiful, novel image was not produced by the hand of a human being.

Just as today, you ask how a robot could 'learn' to sit on a ledge or reassemble a chair from pieces. My friend from the past would ask very similar questions about the painting.

[GH85Carrera]

When I think of the computers having real artificial intelligence I think of that as having a self awareness, like Commander Data on Star Trek. Even the ship's computer could understand commands and complicated questions but it did not have an identify and was not cognizant.

While it is certainly possible to program a computer to control a camera and photograph a sunset by pointing west and the right time and even send it to the printer, it has no sense of what the photograph is. It might just be a sunset on a rainy day, nothing pretty at all.

I guess we can hope that when real self awareness is developed they can program in Asimov's three laws of robotics at the basis of the program.

[Crowbob]

Careful, Glen. You may be swerving back into a discussion of the soul. That issue having been resolved many times right here on PARF!

But back to the artificial artist.

There are many artists and a whole lot of art that isn't pretty at all or so I'm told, not being either an artist or pretty much to my dismay.

[Z-man]

Originally Posted by Crowbob

Great discussion, thank you Z.

What you are describing are the limitations we currently face. However, I am reminded that not long ago it was pretty much definitive that no machine could produce beautiful and novel art. Say, for discussion, a painting that satisfies our current notions of esthetics, even as individual and subjective as esthetics are.

Today, a drone could easily and remotely take a photo of a landscape, a woman or even a chair. Connect that image up to a printer and voyla an esthetically pleasing, novel image is created. Were one to show that image to someone from a century ago that person quite likely would not be able to comprehend that that beautiful, novel image was not produced by the hand of a human being.

Just as today, you ask how a robot could 'learn' to sit on a ledge or reassemble a chair from pieces. My friend from the past would ask very similar questions about the painting.

That, in my opinion, is not AI -- it would be the next action that would be the starting point for AI. Let me explain...

Once the drone has established that a sunset picture falls within the parameters of esthetically pleasing landscape, how does the computer learn about other esthetically pleasing landscapes? Let's assume the computer realizes that a mountain is in his sunset picture, and extrapolates that mountains are esthetically pleasing (big stretch here, I know...). Ok - so the drone starts taking pictures of all types of mountains with the assumption that they are esthetically pleasing. Unfortunately, not all mountains and mountain shapes are pleasing, and the drone computer has no way of discerning between Mt. Rainier, Mt Fuju, a mound of garbage and a pile of steaming poo. All it knows is that a mountain shape is pleasing. One would have to program the AI with more parameters to resolve this -- but that's not AI, that's just humans improving the code...

Over time, humans have developed more intelligent machines. But that's not AI -- that's just humans improving the code that runs the machines.

And we haven't even begun to discuss topics like emotions, morality, ethics, violence and so on...

-Z