Multi-Functional AI

[Chesterton]

I don't know much about programming, so I hope this question even makes sense.

I hear about different kinds of AI that does different things. AI can be used to play chess, in financial transactions, advertising, judicial sentencing, and lots more. Can there be one AI program which does more than one thing? For example, could there be a single AI program which plays chess with me, diagnoses my cancer, and kills me with a drone strike?

 

[Of the Kingdom]

There doesn't seem to be much point in that particular combination of capabilities! Joking aside, with today's best AI methods, any one of the three features is a fairly daunting task. Perhaps in the future there will be "complete personality" automatons which are human-like in their ability to simultaneously have many "interests", and are able to multi-task in ways humans cannot.

The different "skills" you suggest are in effect separate computer programs, but nothing stops us from building a multi-programming computer that can run them simultaneously. If all three "tasks" are somehow focused on you, then the three programs controlling those tasks will presumably interact with each other.

 

[crossnote]

I don't see why not since our computers are doing 100s of tasks simultaneously.

 

[Strathos]

Artificial general intelligence - Wikipedia

 

[timewerx]

The ultimate goal of AI is AI that can learn to do everything.

An AI that knows it all. Google's goal is AI that can write other programs (Artificial Programmer) and even possibly, modify its own code - Google must be extremely annoyed with its human programmers!

 

[Chesterton]

The different "skills" you suggest are in effect separate computer programs, but nothing stops us from building a multi-programming computer that can run them simultaneously. If all three "tasks" are somehow focused on you, then the three programs controlling those tasks will presumably interact with each other.

So as of now I guess there isn't the kind of thing I'm talking about, although maybe there's no real need for it.

Could three programs interacting be considered one set of code? I'm just curious if one program could perform unrelated tasks.

 

[timewerx]

Sometimes I think Humanity is God's own AI project.

 

[Nithavela]

I don't know much about programming, so I hope this question even makes sense.

I hear about different kinds of AI that does different things. AI can be used to play chess, in financial transactions, advertising, judicial sentencing, and lots more. Can there be one AI program which does more than one thing? For example, could there be a single AI program which plays chess with me, diagnoses my cancer, and kills me with a drone strike?

In that order?

Real AI hasn't been invented yet. What we have is a bunch of algorythms that nobody really knows how they work, only how to create and train them.

 

[Chesterton]

In that order?

lol

Real AI hasn't been invented yet. What we have is a bunch of algorythms that nobody really knows how they work, only how to create and train them.

Yeah I know, but everybody calls it "AI". And yeah, I've heard when it starts producing weird results it can sometimes be extremely difficult for the people who designed it to figure out why.

 

[Of the Kingdom]

Could three programs interacting be considered one set of code? I'm just curious if one program could perform unrelated tasks.

Yes, what constitutes a "single program" is a lot more vague than, for example, what constitutes a single human brain. It is possible to have multiple processors and multiple pieces of computer code organized so that it is not clear how many "programs" or "computers" you have.

 

[usexpat97]

My short answer would be "no". Machine learning targets a very pointed question you are trying to ask. And you have to train it with everything you might already know about the data, if you are to have any chance of getting a decent probability of accuracy out of it. If you are simultaneously trying to ask it 3 different questions, for which the data you already know has zero correlation with one another, then it's basically 3 different programs. It's like the difference between programming the computer to win 3 games of 3-dimensional tic-tac-toe, vs. winning one game of 100-dimensional tic-tac-toe. How in the world would you do THAT?!??

 

[timewerx]

In that order?

Real AI hasn't been invented yet. What we have is a bunch of algorythms that nobody really knows how they work, only how to create and train them.

AI doesn't always have to be very machine-like in capability.

Strategies for developing AI and its application can be many and one or more of these could produce a result very close to human thought.

Our mind's shortcomings when it comes to retaining, retrieving, and manipulating information is actually a feature designed for a purpose, not a limitation.

It allows a person to have a different perspective over the same event, thought or idea or beliefs / principles over a period of time having gained valuable experiences.

It helps prevent us from latching onto false, misleading, corrupt, and wasteful information or beliefs. Of course, not always successful due to our exploitable weaknesses such as physical limitations, survival instincts, emotions, hormones, animal desires / behaviors, etc.

Without those weaknesses, we are essentially very machine-like and much closer to our intended purpose and of course, the more we encourage, promote, and even take pleasure in our weaknesses, the more we are at risk of accepting false/misleading information.