Artificial Intelligence

JohnRoth wrote:I completely agree with RFC's decision to ban general AI from the Honorverse.

As do I...but it sure is a hoot in the Furyverse.

The basic fact is that we really can't define what "intelligence" is

"Intelligence" is sufficient complexity/capability in any computing and control system where the system can successfully and reliably repurpose or even improve the efficiency of its programming by itself, usually to adapt to external requirements.
With the above definition, humans are borderline intelligent for now; they can program new functions into their system beginning with "balance" and "sensory pattern recognition", which leads to "walking" and "language" and eventually to things such as "philosophy" and "abstract math" BUT that reprogramming is kinda slow as it is based mostly on external stimuli and its reliability is low as its coding ends up being an absurdly complex and tangled mess because it adapts on every single stimulus (internal and external), not just beneficial ones. Also, humans have no direct access to their code for improvement purposes. That's a bad thing in many cases (you can't edit away your dislike for your boring work, or PTSD, or your smoking habits) but a good thing in others (the government can't make programmed assassins, corporations can't make corporate drones whose only goal in life would be to work all day with no pay and enjoy it).

Nor can we define "consciousness."

I think you mean "sentience". That's the capability for any system to perform self-analysis and/or self-improvement of its own programming, without being based on any new external stimuli for it - though it is usually based on past stimuli. This definition covers things such as knowing you exist, knowing whether you're intelligent or not, and directing your own programming's improvement (or lack thereof).




It should be noted that a system can be intelligent without being sentient - that would be the case of several animals such as dogs and cats and dolphins, for example, or adaptive programming such as seen in the Honorverse in some cases.
It should also be noted that you can be sentient without being intelligent - that would be the case of A.I.s who are self-aware but whose creators "locked" their core programming and capabilities from being edited to prevent such things as "going Skynet".

Daryl wrote:Fred Saberhagen's Berserker series illustrates the dark side, but the Bolo series shows a happier outcome.

I kind of disagree with the Berserker series showing only the darker side. Yeah the Berserkers are the enemy due to a programming glitch but there are also several examples of AIs that haven't turned against organic life in the setting.

Belial666 wrote:

JohnRoth wrote:The basic fact is that we really can't define what "intelligence" is

"Intelligence" is sufficient complexity/capability in any computing and control system where the system can successfully and reliably repurpose or even improve the efficiency of its programming by itself, usually to adapt to external requirements.

So you're going with the idea that "intelligence" is an emergent property of complexity. I think this is the same idea that's being flogged by the Singularity folks. I'll just say that I don't buy it, at least in the form Kurtzweil and company are pushing it.

Belial666 wrote:With the above definition, humans are borderline intelligent for now; they can program new functions into their system beginning with "balance" and "sensory pattern recognition", which leads to "walking" and "language" and eventually to things such as "philosophy" and "abstract math" BUT that reprogramming is kinda slow as it is based mostly on external stimuli and its reliability is low as its coding ends up being an absurdly complex and tangled mess because it adapts on every single stimulus (internal and external), not just beneficial ones. Also, humans have no direct access to their code for improvement purposes. That's a bad thing in many cases (you can't edit away your dislike for your boring work, or PTSD, or your smoking habits) but a good thing in others (the government can't make programmed assassins, corporations can't make corporate drones whose only goal in life would be to work all day with no pay and enjoy it).

Are you familiar with the phrase "this is so bad it isn't even wrong?" Both examples you cited are special-purpose systems, one of which (balance) isn't even in the central nervous system (aka the brain).

You also seem to be using the "programming" metaphor, which is what AI started out with in the 50s and 60s, and which ran into a blank wall - it's now pretty well accepted (except by a few old fogeys) that the way the brain works has almost nothing common with computers as we know them.

Belial666 wrote:

JohnRoth wrote:Nor can we define "consciousness."

I think you mean "sentience". That's the capability for any system to perform self-analysis and/or self-improvement of its own programming, without being based on any new external stimuli for it - though it is usually based on past stimuli. This definition covers things such as knowing you exist, knowing whether you're intelligent or not, and directing your own programming's improvement (or lack thereof).

No, I said consciousness and I meant consciousness. That is, the ability to reflect on one's own internal processes. Calling it "sentience" is simply a verbal ploy to avoid the philosophical knots that contemplating consciousness creates.

Belial666 wrote:
It should be noted that a system can be intelligent without being sentient - that would be the case of several animals such as dogs and cats and dolphins, for example, or adaptive programming such as seen in the Honorverse in some cases.

Dogs are definitely conscious. The definitive test for first level consciousness is whether the critter can recognize itself in a mirror. Dogs pass this test easily.

Belial666 wrote:It should also be noted that you can be sentient without being intelligent - that would be the case of A.I.s who are self-aware but whose creators "locked" their core programming and capabilities from being edited to prevent such things as "going Skynet".

And this is just playing with words, IMNSHO.

If you want to look at some cutting-edge research in psychology, I'd suggest looking at "embodied cognition," or the idea that a lot of behavior is shaped by the body and has little or no input from the brain. Robotics is a good example: robots that are constructed with a central programming metaphor tend to have really bad characteristics when faced with something that isn't in their programming. Consider the difference between Honda's ASIMO and Boston Dynamic's Big Dog.

A couple excellent treatments of future ramifications of AI can be found in the late Iain M Banks' "Culture" series and in Alastair Reynolds' (mostly) standalone novel "House of Suns."

The former is a pretty well-known series, so I doubt I need to go into detail, but the Culture, humanity's prominent polity in the series, contains both planet-based and ship-based "Minds," AIs which vastly outstrip organic or lesser artificial intelligences. Minds are full citizens (and tend to be in charge, obviously...they're really good at it), as are lesser AIs that are nevertheless sentient beings. It's a fairly rosy vision...but not entirely, and Banks explores some of the less apocalyptic downsides of cohabiting with being that approach weakly-godlike intelligence.

As an aside, Culture military craft have got to be on any SF fan's short list of "most badass fighting ships," although as always, direct comparisons between writers' vastly different "universes" rapidly becomes an apples 'n oranges exercise. Best ship names ever, too...

Reynolds' "House of Suns" posits the Machine People, who are pretty much exactly what the name implies. The interesting take with them is that when they arose, it was we (the organics) who tried to wipe them out, not vice-versa. We failed, and the Machine People largely isolated themselves...there is very little interaction between "species," even over the millions of years the tale encompasses.

runsforcelery wrote:== CLIP ==

Ah, but who's to say it will? I tend to agree with Asimov about the "Frankenstein Complex," but this could well be the sort of mistake you only get to make once, no?

If you haven't read it, you might take a look at Jim Hogan's Two Faces of Tomorrow. It was published in 1979, I think, but it was an interesting take on one way to approach the "is their poison in that candy" question about AI. It's a Baen title, and I'm pretty sure it's still available.

Got it. 1979 paperback, author James Patrick Hogan, $1.59 at Half Price Books.
I'll try to check it out fairly soon.

BTW...

On the same day, I also won an auction for "Crusade in Europe" by Eisenhower, First Edition, 48' for $8.07+t/h
* http://www.shopgoodwill.com/viewItem.as ... D=16480426
Does anyone have an opinion as to if this seem like a good deal or not?


.

Hi Ensign Re-read,

Depends on whether you want a rather general overview of the western European campaigns with very little [as in darn near none] technical data at all or devoid of almost all personality classes etc, ie rather banal and PC.

I think you could find it cheaper at a second local bookstore to save on shipping if you still want it.

I'd think some edition somewhere online ought to be much cheaper.

L

Ensign Re-read wrote:

runsforcelery wrote:== CLIP ==

Ah, but who's to say it will? I tend to agree with Asimov about the "Frankenstein Complex," but this could well be the sort of mistake you only get to make once, no?

If you haven't read it, you might take a look at Jim Hogan's Two Faces of Tomorrow. It was published in 1979, I think, but it was an interesting take on one way to approach the "is their poison in that candy" question about AI. It's a Baen title, and I'm pretty sure it's still available.

Got it. 1979 paperback, author James Patrick Hogan, $1.59 at Half Price Books.
I'll try to check it out fairly soon.

BTW...

On the same day, I also won an auction for "Crusade in Europe" by Eisenhower, First Edition, 48' for $8.07+t/h
* http://www.shopgoodwill.com/viewItem.as ... D=16480426
Does anyone have an opinion as to if this seem like a good deal or not?


.

Even if there's nothing necessarily "bad" about AI I think it would be a truly dangerous technology simply because if an AI thought differently from a human, we wouldn't understand it well enough to "raise" it and it would probably end up with psychological problems.

(And if it DIDN'T think differently from a human -- what's the point? We can already reproduce humans.

Similarly for human cloning -- I really don't see the practical application. cloning individual organs for transplants, sure. People, no point to it.)

cralkhi wrote:Even if there's nothing necessarily "bad" about AI I think it would be a truly dangerous technology simply because if an AI thought differently from a human, we wouldn't understand it well enough to "raise" it and it would probably end up with psychological problems.

Interesting question: If the AI's model of sentience is fundamentally different from a human's, how would you know whether it was psychologically damaged? It's not like you have a baseline of correct or healthy behaviour that you can compare it to, after all.

Way I see it, an AI modelled after a healthy human psyche is something we don't really want anyway. There are several reasons for this, from the ethical (what happens to the prototypes that aren't quite there? What happens when you reset the AI? Where's the point beyond which an AI has to be treated like a human being, with all the rights associated with that?) to the practical (Most humans really don't like to be pressured into performing work they have no inclination for).

cralkhi wrote:(And if it DIDN'T think differently from a human -- what's the point? We can already reproduce humans.

Similarly for human cloning -- I really don't see the practical application. cloning individual organs for transplants, sure. People, no point to it.)

I'm wondering insofar faulty logic is inherent to intelligence. Humans have an awful lot of cognitive biases, but they are there for a reason, being able to quickly classify something, and speeding up decision making. Sometimes this goes wrong, and we have prejudice, racism, and ignorance of ones own limitations, all things one really doesn't want in an AI.