PDA

View Full Version : Need input from computer scientist



Evan Henry
01-19-2015, 01:39 AM
Hi, everyone:

I'm currently working on a series of short stories dealing with artificial intelligence. My approach is something of a hybrid between traditional "hard" science fiction and a more character-driven story, but I am trying to be as realistic as possible. I'm scientifically literate but by no means an expert.

For the stories, I've "invented" a few (possibly absurd) ideas for how a powerful AI system would function, and I'd like to get some input from someone who actually knows what they're talking about. Rudimentary (or better) knowledge of electrical engineering, and maybe even quantum mechanics couldn't hurt.

Drop me a PM if you'd be interested in reading! I feel like it's a bit in-depth to spell it all out here.

morngnstar
01-19-2015, 04:51 AM
I have a bachelor's degree in computer science. I've never been quite comfortable with calling anyone a scientist (computer or otherwise) unless they have a Ph.D. But I am a software engineer. I touch on AI a little bit in my work and studies. So I will try to answer your questions.

Cath
01-19-2015, 05:10 AM
Do you have a specific question to ask, Evan?

Evan Henry
01-19-2015, 07:11 AM
Do you have a specific question to ask, Evan?

Yup! It's just long and complicated and involves several passages from multiple stories, so I'd rather do it by PM, if possible. :)

T Robinson
01-19-2015, 08:05 AM
Yup! It's just long and complicated and involves several passages from multiple stories, so I'd rather do it by PM, if possible. :)

I think that part of the benefit of the forums is that people see the replies and learn something, even if it does not directly apply to them. No one wants lengthy passages, but it should be possible to summarize what you are trying to determine.

I agree with what appeared to be Cath's intent <can't speak for her>, ask a specific question. You might be surprised at some response that sparks an idea. Many people have read stories involving AI and might be able to tell you what has been used, even if they are not scientists.

Amadan
01-19-2015, 08:24 AM
FWIW, electrical engineering and quantum mechanics might have some bearing on how an AI would be built (I am skeptical about the quantum part, since "quantum mechanics" seems to be a synonym for "magic" in a lot of SF), but not much relevance to how it would function. If you're wondering how powerful a hypothetical AI could be, anything past the very near future is going to be completely speculative.

Cath
01-19-2015, 04:18 PM
I think that part of the benefit of the forums is that people see the replies and learn something, even if it does not directly apply to them. No one wants lengthy passages, but it should be possible to summarize what you are trying to determine.

I agree with what appeared to be Cath's intent <can't speak for her>, ask a specific question. You might be surprised at some response that sparks an idea. Many people have read stories involving AI and might be able to tell you what has been used, even if they are not scientists.

Yep, exactly this!

Dennis E. Taylor
01-19-2015, 07:44 PM
FWIW, electrical engineering and quantum mechanics might have some bearing on how an AI would be built (I am skeptical about the quantum part, since "quantum mechanics" seems to be a synonym for "magic" in a lot of SF), but not much relevance to how it would function. If you're wondering how powerful a hypothetical AI could be, anything past the very near future is going to be completely speculative.

Quantum computing, on the other hand, would be absolutely essential for an AI, if it was to run in anything like real-time. You can run neural networks on digital computers, but you basically ask the computer a question then go for lunch.

For some fun light reading, look up qubit.

Amadan
01-19-2015, 08:37 PM
Quantum computing, on the other hand, would be absolutely essential for an AI, if it was to run in anything like real-time. You can run neural networks on digital computers, but you basically ask the computer a question then go for lunch.

For some fun light reading, look up qubit.


I'm quite familiar with quantum computing, and "quantum computers" still tend to get used as shorthand for "magic" - i.e., everything that we currently lack the technology or computational power to do now, a lot of writers think will just magically become possible with a quantum computer. Certainly there's reason to hope quantum computers will vastly increase computing power, but functional Artificial Intelligence does not necessarily follow.

kuwisdelu
01-19-2015, 08:41 PM
Quantum computing, on the other hand, would be absolutely essential for an AI, if it was to run in anything like real-time. You can run neural networks on digital computers, but you basically ask the computer a question then go for lunch.

For some fun light reading, look up qubit.

No. Neural networks aren't really anywhere close to AI, and more computational power won't change that.

We need better algorithms first. Quantum computing won't change that.

Dennis E. Taylor
01-19-2015, 11:27 PM
No. Neural networks aren't really anywhere close to AI, and more computational power won't change that.

We need better algorithms first. Quantum computing won't change that.

Well, actually they will. Running a neural net on a digital computer requires simulating multiple continuous inputs to produce output, and it has to be done for each neuron. With quantum computing, multiple inputs simply superimpose themselves on the qubit and you get a real (or more real, anyway) neuronal output. No question the algorithms will have to improve as well, but so much of the effort right now is in just replicating the physical models, without even getting anywhere near how the input is actually used.

T Robinson
01-19-2015, 11:34 PM
Do you see what I mean about the interactions developing just from your initial post?

Other people bring bits of knowledge to the table (cliche alert) and the sum of the whole is greater than the individual parts. Someone somewhere has the knowledge you need or may spark the idea in you.

It is a proven system and it works. Try it. Ask a specific question and see what happens.

Amadan
01-19-2015, 11:36 PM
Well, actually they will. Running a neural net on a digital computer requires simulating multiple continuous inputs to produce output, and it has to be done for each neuron. With quantum computing, multiple inputs simply superimpose themselves on the qubit and you get a real (or more real, anyway) neuronal output. No question the algorithms will have to improve as well, but so much of the effort right now is in just replicating the physical models, without even getting anywhere near how the input is actually used.


The problem is you are assuming a neural net is actually a simulation of how the brain works and that a fully functional neural net, working fast enough, would equal AI. That was the theory behind them when they were first developed, but while neural nets have proven useful for some applications, most computer scientists are now very skeptical of the NN-as-model-for-the-brain theory, and in machine learning and AI in general, neural nets are a bit passe.

robjvargas
01-19-2015, 11:37 PM
Sort of like what Angry Guy said, we need a new order of magnitude in size reduction to achieve human-level intelligence in anything approaching the same space as the brain. It may or may not affect the computing, but it definitely is part of the engineering.

Taejang
01-20-2015, 12:16 AM
Without a better understanding of the real question(s), may I suggest studying Watson (http://en.wikipedia.org/wiki/Watson_%28computer%29) as a great example of current technology and its capabilities and limitations?

Furthermore, if you want your AI to perform like a human, perhaps as an android, the speed of human thought (http://en.wikipedia.org/wiki/Nerve_conduction_velocity) is also a good study. A fully-functional AI must be capable of functioning at least this quickly. An AI will probably have at its disposal a vast amount of information, possibly much more than any human, and thus requires very good algorithms and methods for searching and good processing time once data is found.

Dennis E. Taylor
01-20-2015, 12:25 AM
The problem is you are assuming a neural net is actually a simulation of how the brain works and that a fully functional neural net, working fast enough, would equal AI. That was the theory behind them when they were first developed, but while neural nets have proven useful for some applications, most computer scientists are now very skeptical of the NN-as-model-for-the-brain theory, and in machine learning and AI in general, neural nets are a bit passe.

This is turning into a derail. You're focussing on my mention of neural nets (which I brought up simply because that's what we have right now) while completely bypassing the question of quantum computing, which regardless of what software methodology we use, or what we happen to be calling it at the time, is going to be a make-or-break.

Cath
01-20-2015, 01:00 AM
Gently, peeps. It's an interesting discussion.

morngnstar
01-20-2015, 01:05 AM
I agree with the common "quantum computing as magic" misperception. It does not provide an automatic speed-up factor for every computing task. However, it does seem plausible that neural nets would get a speed-up. Neural nets are a bit out of fashion, but it's possble they'd get more results if you could run them at brain-like scales.

benbradley
01-20-2015, 01:39 AM
Yup! It's just long and complicated and involves several passages from multiple stories, so I'd rather do it by PM, if possible. :)
In that case, maybe it would be appropriate for Share Your Work (as that forum is protected by password, posting there won't be "publishing" anything on the Internet), with specific instructions to critique the technical aspects of the story.

You could then post the title of that thread here, and those interested could go find it.

Quantum computing, on the other hand, would be absolutely essential for an AI, if it was to run in anything like real-time. You can run neural networks on digital computers, but you basically ask the computer a question then go for lunch.

For some fun light reading, look up qubit.
I'm tempted to jump into this, but I should really order a Parallella board.

stephenf
01-20-2015, 02:08 AM
Hi
The only thing you need to know is AI is fiction .

Amadan
01-20-2015, 02:31 AM
Hi
The only thing you need to know is AI is fiction .


That's hardly helpful. It depends on how "hard" he wants his SF to be.