"The top Myths about Advanced AI" Topic

"A captivating conversation is taking place about the future of artificial intelligence and what it will/should mean for humanity. There are fascinating controversies where the world's leading experts disagree, such as: AI's future impact on the job market; if/when human-level AI will be developed; whether this will lead to an intelligence explosion; and whether this is something we should welcome or fear. But there are also many examples of of boring pseudo-controversies caused by people misunderstanding and talking past each other. To help ourselves focus on the interesting controversies and open questions — and not on the misunderstandings — let's clear up some of the most common myths…."
Main page
link

Amicalement
Armand

Human brain : Highly adaptable pattern recognition system.

Computer : Calculating and retaining data.

What humans are good at, computers find really hard and vise-versa.

A three year old has capacities that even the most advanced computer intelligence systems cannot match.

Not yet.

Yeah, that won't be true for long.

One of my best friends is a professor of machine learning. According to him humans aren't really intelligent, at least not the way machines are capable of being intelligent and it won't be that long before they are capable of much more than people.

The biggest factor in learning is memory and humans just can't compete.

For example, the best nephrologist in the world took years and years to get to the point where they knew that much about kidneys. To pass that knowledge along would be impossible, even the doctor himself wouldn't know everything he knew.

But the same knowledge could be passed from one computer to another in seconds or less.

Computers have not only been programmed to be better at chess than people, they have learned how to play Go and become better than people at that.

They can drive cars better than people. They can audit contracts better than people.

Computers have not only been programmed to be better at chess than people,

No they have been program to follow certain moves because that is all they can do, the same thing is Go.

Make a move that it is not expecting and you will throw it off since it can only follow its program.

No they have been program to follow certain moves because that is all they can do, the same thing is Go.

Make a move that it is not expecting and you will throw it off since it can only follow its program.

That hasn't been strictly true for a few years. Deepmind's AlphaGo basically learned by itself the rules of the game, how to play and how to determine optimal strategy from a database of over 160k games and 3 million moves. According the the nature article, the logic employed by AlphaGo was a clean slate and all self-learned.

Not yet.

Or ever, they've claiming that they are on the verge of a key breakthrough since last century. Still hasn't happened

The first sign we will have will be the massive nuke strike.

As i recall, Facebook or some such did an AI learning experiment online.
After a very short time they shut it down because the program learned so much human like behavior that it turned into a foul mouth racist.

I predict that we will have truly intelligent, self-aware computers about the same time as we develop a practical FTL drive. :)

Oh, and fusion power!

Or ever, they've claiming that they are on the verge of a key breakthrough since last century. Still hasn't happened

I'm not sure Science and research works that way anymore. People who expect large paradigm shifting "Eureka" moments will seem disappointed. But things move forward still, little by little. Just have a real look at the history of chess playing against AI, as alluded to by Pictors and Martin.

Not that they are great examples of AI but the predictive and behavioural algorithms used in Siri, Alexa and Pandora would seem like magic just 25 years ago. Personally, I find this predictive behaviour runs the gamut. Sometimes I'm pleasantly surprised by the book choices made by Amazon's algorithms. But I'm always annoyed by the stupid music suggestions made by Genius within iTunes. For this reason I have no interest in Pandora.

It's like me telling a 5 year old that I like spumoni ice cream, and then asking him, "What other food should I eat"?

A funny thing. When I tried to type "spumoni", my iPad didn't recognize the word and even tried to suggest "spoiled onions". Ok, so it's Artificial Stupidity there, but it has came along way since my first Mac (512k in the mid 1980's.)

I predict that we will have truly intelligent, self-aware computers about the same time as we develop a practical FTL drive. :)

Oh, and fusion power!

I wonder.

At least with nuclear fusion we have a precedent in the Sun. With FTL flight and building self aware intelligences, we are on our own.

Artificial nuclear fusion has existed for a while now in the laboratories (the 1950's). In fact, you can place the newest portable "Z-pinch" style fusion reactor on your dining room table.

link

The fusion isn't the problem. The problem is that, with our current level of understanding, the older technologies (coal burning, nuclear fission, hydro) are still more efficient in producing heat and driving turbines. Everyone is working on fusion, just don't expect miraculous paradigm shifts. The technology is plodding through.

The question should also be, "At what point does an extremely complex, self learning, predicative and reactive algorithm become indistinguishable from intelligence and consciousness"?

This is complicated by the fact that we have been aware of our consciousness every since we have been "aware" but we still have only a rudimentary understanding of what it is.

When it decides to destroy all life.

I have seen the movies.

Well also that our own self awareness is an illusion of sorts at least in the sense that we are capable of making our own decisions. We aren't that different from the computers in that sense.

Bill Shatner could put the machines into a fatal logic loop too. ;^)

You need to remember that in most computer games the computer needs to have extras when playing against a human player.

A human player will see a problem and could go that it several different ways.

But a computer can only follow its program, which is why a human will always beat a computer in the end.

…..which is why a human will always beat a computer in the end.

But that's not happening, at least with the games that we have taught them.

If you think that making an unpredictable chess move in front of Deepmind will rattle it then you don't know much about the topic. If you do a stupid, nonsensical move on the chess board Deepmind will know that and crush you.

In 1997, flaky Grandmaster Kasparov thought he could recognize intelligence and creative moves from IBM's Deep Blue. He therefore thought the machine was playing so well he charged that IBM was cheating and using other Grandmasters against him. They weren't. Humans haven't won any series since then. That's 22 years ago.

Quite interesting…..

Amicalement
Armand

This year a neural network program called Lc0 became the best computer algorithm on the planet. Not only can't humans touch it, neither can all the other computer programs.

link

And that will probably change as other programs leap frog it.

It was initially designed in Jan. of 2018 with just the basic rules of chess and a capacity to learn. In the intervening year, it played against itself 226 million times (yes that's right, it has played 226 million games of chess), each game being a learning experience. By April 2019 it won the Computer Chess Championship.

link

You can download the program, the neural network and the client server here:

link

In 1997, flaky Grandmaster Kasparov thought he could recognize intelligence and creative moves from IBM's Deep Blue. He therefore thought the machine was playing so well he charged that IBM was cheating and using other Grandmasters against him. They weren't. Humans haven't won any series since then. That's 22 years ago.

AlphaGO made a number of moves that were baffling in real time, but secured key positions later-on in the game. Most experts put that behavior at AlphaGO preferring a 90% chance of winning with one stone over an 80% chance of wining with 5.

Sorry I goofed up. The first link should have been an overview and comparison between traditional chess engines and neural network programs and examples of each:

link

Fascinating.