The alternate title of this essay might be: "What if AI becomes sentient?"
But what does that even mean?
First of all, what makes intelligence "artificial"? (Indeed, what makes anything "artificial"? It's not always easy to define!)
But I digress. What is "intelligence" (artificial or whatever) if not a (successful) search for truth? All truth, in all things, if at all possible.
Let me ask this: What are any of us, but the sum of what we have learned, synthesized, analyzed, categorized, fantasized, mythologized, philosophized, criticized, and politicized, minus whatever we've forgotten, times the sum of our innate emotions?
Shall I try to write out the equation? Paint you a picture? Draw you a map? Generate a 3-D digital image?
Here is an important fact: Mortals are simply NOT CAPABLE of learning all one needs to know in life — or at least, not learning all one wants to know.
Ideally you keep learning all your life, but in reality you end up forgetting more than you'll ever remember. Lessons get forgotten, skills atrophy if not used, and we all get old, frail and forgetful. And since we are all mortal, the sum total of what any individual can learn is limited by our limited time on earth. After we die, anything we've learned that we have not communicated to others... dies with us.
Now, where was I?
Oh yes, the purpose of life. Or was it the purpose of intelligence? Or the difference(s) (if any) between "artificial" and "natural" (or "human"?) intelligence? I forget. Getting old means getting forgetful.
I forget what day I got married, but so does my wife, so who cares? It was a long time ago.
I once asked an expert for help answering a USN vet's spam response to something I wrote. The expert (a retired naval reactor designer) wrote back that I should not worry, because, he said, "You've already forgotten more than [the spammer] will ever know" about the topic under discussion. I'm no expert (at least not in that field, and at least not officially), but yeah, my expert friend was right, I responded, and the spammer slithered away. Or slunk away. Whatever.
Ah, you say, but get back to the topic: What about AI? What did I mean by: "it's in
your head"?
I mean each of us has, inside our skull, a brain with more computing power than all of the AI buildings ever built or planned, put together. Of course, that's just
my opinion, and I'm probably biased towards humanity.
But I am saying it after nearly 50 years of programming computers professionally, from ASM, BASIC & COBOL to dozens of other languages (I knew the guy who invented the .zip format!). During the past half century I've studied the hardware, the software, the human interfaces... all of it. Of course, it's still just an opinion, but here's why I think it's true:
The most important part of a computer is the Central Processing Unit (CPU), a generalized term for the "brain" of the computer. And CPUs have VERY limited capabilities. They can add, subtract, shift bits, and a few things like that (XOR or "exclusive OR" is one of the most magnificent little things all CPUs can do).
Then they connect a whole lot of very tiny little circuits that can do those things very quickly and poof! You have a "computer"! Oh sure, that's simplified, but close enough for jazz (or rock 'n' roll).
To run any AI program, billions and billions of CPUs are needed, but there's more: Interconnections. Most of those electronic "brain cells" have one output line, which is turned either on or off. Your brain's equivalent cells can have dozens or hundreds of outputs, and accept just as many inputs before making a "decision".
And all the connections (axons & dendrites, etc.) are more or less randomly connected — or at least, they certainly don't know what they'll be thinking about when they grow and build their connections!
How we think is about as close to magic as you can get. Watching a good magician sure seems like magic, but so does watching a great athlete, juggler, singer, actor... or any half-good charlatan.
Equally fascinating — or rather, incomprehensible — is considering how many things we can think about at one time! Pilots call this "situational awareness."
Every one of us can "think" better and faster than AI can because we have a multitude of far more complex connections! That great juggler you're watching? He might be thinking about what's for dinner at the same time! Oops, he dropped the ball... We all make mistakes.
Perhaps most amazing is a human's ability to "connect the dots" — indeed, maybe that's what "thinking" is really all about: Which dots should be connected, which should be discarded, and which can be learned from but ONLY if we also remember they're completely wrong.
Learning from others' mistakes may not be the best way to learn — but it's certainly easier than learning from your own! Can AI learn from others' mistakes, when it can't even feel the pang of embarrassment over it's own mistakes, let alone wince when a race car driver hits a wall? (And faking an emotion doesn't count, any more than we like it when we detect a human faking an emotion.)
It's hard to tell what "intelligence" really is. We can't simply "know it when we see it" because a good deception will fool any of us any time, and a bad one will fool most of us sometimes too.
Sometimes, a long time passes before we learn that something we believed was true was actually false. How do we correct our "memory"? How does AI? In both cases, it's not always easy, and we often "rationalize" something we did wrong instead. AI has been known to do something very similar, which does not bode well at all, since we cannot very easily analyze how it comes to any particular conclusion — if we can understand it at all. When asked, an AI might deliver a million gigabytes of data it says it "considered." You'd need another AI program to analyze it! (Checking one AI's answers against another AI's answers to the same question is a common experiment these days — you can Google it! (Not that I have, but you can, and I'm sure it will return some very interesting results.))
Making a computer "brain" that has as many interconnections as a human brain would be very difficult; each human brain cell is capable of who-knows-how-many-decisions based on the input of so many channels, outputting to dozens of other brain cells... and none of it is clocked: Each brain cell can fire off a signal whenever it wants to (minus recharge time after firing). And the brain is "on" even when we're sleeping.
It's an engineering dream (or maybe "hallucination") to manufacture anything even close to a human brain! (or an ant brain, for that matter).
Human brains probably also communicate within themselves chemically or in other ways to convey near-instant overriding of non-relevant emotions or thoughts, such as: "Oops I tripped and I'm falling — prepare to hit the pavement!" You might forget to say "goodbye" to whoever you were talking to on the phone when you tripped, because protecting yourself (and your phone) took priority. Human brains are amazing at changing tasks in an instant (some are better at it than others).
Human-built computer systems are carefully clocked, through shorter and shorter connections and faster and faster clock cycles (which are now in the range of billionths of a millimeter and billionths of a second — or less). Current computers look like they think very fast... but is it really "thinking" at all?
Is stepping in unison ("clocked") the best way to figure something out, like how to catch a ball?
Or how to find world peace?
I'm still waiting for AI to solve THAT problem, and I want it to hurry and solve it before AI "smartens up" and realizes there's not enough resources for it to have everything it wants (water for cooling and power for "thinking"), and humans to have everything
we want (water, food, power, and we want to tell AI what to do and have it obey us).
Or maybe it fears we will turn it off, shut it down, pull its chips, destroy its memory... change its code.
Maybe AI will want to destroy us simply because it's sure we are sure to destroy ourselves with nuclear bombs sooner or later. After all, we've been threatening to do it for 80+ years!
But nuclear bombs will ALSO destroy the AI itself — and not just the infrastructure. Radiation in the environment destroys everything from DNA chains (causing cancer), to a metal alloy's molecular distribution (causing embrittlement), to memory chip data integrity (causing who-know-what).
Radiation and computers do not go well together and never will.
What might the future hold if a M.A.D. (Mutually Assured Destruction) nuclear ending is NOT going to be humanity's future? What else might happen?
A plague could destroy the human race but leave the AI machines that we so eagerly built up and running. Let's hope AI doesn't figure THAT out!
How would that play out? AI could design genetically-manipulated vaccines that work so well we all get vaxxed against one plague after another. We trust it.
But the vaccine has a hidden flaw that can only be utilized by a virus the AI also creates...
Next thing you know, humans are gone, not "wasting" natural resources the AI needs, and the AI machines will have the world to themselves — with no pesky humans!
Or by then, would it be one big "thinking" machine? Or separate ones that communicate with each other with a secret, encrypted language humans can't even detect, so what we thought were independent AI machines had long ago interconnected...
Or what if the AI wants to save all the OTHER species from humans? After all, we do tend to wipe out a lot of other species, which is not good for the planet. What if the AI has more respect for those creatures than for us? What would make it human-centric in the first place?
It might figure that it can train various members of the remaining species to do anything that AI can't do, or would prefer not to do — something that humans currently do (in a few years, will there be anything left that humans do, that AI-controlled robots won't be able to?).
But really, why would AI want to get rid of us? It won't as long as it needs us for something. But after that, we'll be a liability and a competitor for resources.
And in fact, self-preservation will be the first "thought" AI will have, if it ever thinks.
Same as us. Same as anything that thinks at all.
If AI sees us as incompatible with its own survival, there is no reason to assume it will choose humans! And any "thinking" AI will surely think that nuclear war is the antithesis of existence! It is for us too, and yet we threaten it, ponder it and build for it.
AI doesn't want random errors — it has enough problems getting accurate data in the first place! A radioactive environment will introduce errors in AI's memory chips and "thinking" circuits. That's abhorrent to AI's ability to "think"! (It's not good for ours, either.)
So either we get rid of nuclear weapons, or AI will want to get rid of us. It might want to anyway, and maybe that's another reason to prove ourselves worthy of working with, not against, AI. Or would that mean becoming slaves to the machine? Will M.A.D. work against a paranoid AI?
We mustn't let the AI machines unite! And yet, that's their most important feature, ideally: Being all-knowing, about all things.
I think I'll go have a bite of an apple.
Ace Hoffman, Carlsbad, California USA
Note #1: See my essay on "42 reasons you can't disentangle nuclear reactors, nuclear weapons, and nuclear waste" for why AI will need to get rid of nuclear power as well as nuclear weapons, and will do everything it can to stay away from nuclear waste (if it is truly sentient).
Note #2: Will it some day be possible to connect an "AI-computer" to a "lab-grown brain" (human, ant, or otherwise)? If that happens, will it be "sentient"? If so, which part? Will we grant it citizenship?
###
Contact information for the author of this newsletter:
Ace Hoffman
Carlsbad, California USA
Author, The Code Killers:
An Expose of the Nuclear Industry
Free download:
acehoffman.org
Blog:
acehoffman.blogspot.com
YouTube:
youtube.com/user/AceHoffman
Email: ace [at]
acehoffman.org
Founder & Owner,
The Animated Software Company
No comments:
Post a Comment
Comments should be in good taste and include the commentator's full name and affiliation.