Peter Watts says (edited slightly to pull out email headers, etc):

Latest volley. I thought of just posting it as an anonymous comment to your LJ account, but I didn't know if you'd rather stream these as a single thread or as standalone postings-- format as you see fit.

Anyway:


At 08:09 PM 12/27/2006, you wrote:

Peter, in his reply (posted earlier today) says "planning doesn't require sentience." He sites as an example computerized chess programs. He's correct. I was in error in my post, which focused on planning. (He's been ruminating on this subject for years, I just got started last week.) Before I move on, I will point out that my day job for the last ten years has been herding computer networks large and small, and I know exactly how that computer chess program works - it calculates every possible move. It's a brute force approach that requires at least a 33 MZ chip (33 million calculations / second) to work.

Really? *All* of them? There aren't any fuzzy-logic, general-rules-based algorithms out there? Way back in the early nineties, Carnegie-Mellon had neural nets that learned how to drive vehicles on public roadways, based solely on experience/reinforcement. They didn't preprogram any conventional rules of the road at all, if I understood the writeup correctly. They only programmed rules on how to derive new rules. So I'd be really surprised if here in 2006, smart systems were still totally reliant on brute-force calculations for their decisions.

(Chris Gerrib) Peter points out that the conscious mind gets inputs "predigested" (my words) from the subconscious. Therefore, what exactly does the conscious mind do? Let's stick with the computer chess example. The computer can calculate all possible variations of moves and the probabilities thereof. But until a computer reaches up and tweaks the player's nose, it ain't sentient. The computer can't plan outside the rules of chess - it can't think outside the rules of the game - or "think outside the box." I would argue that the first time we can prove humans thought outside the box was when they controlled fire. There have been others - agriculture, metalworking and animal husbandry come to mind.

Again, I'd cite the heuristic elements of neural nets; these things *can* make up their own rules Further, it's pretty straightforward to see how such rule-creating behaviors could arise through brute, blind natural selection. All you need is time, and a large enough population whose individuals vary in the degree to which they use the time-series buffers in their brains. If long-term strategic thinking results in greater fitness (and cockroaches provide sufficient evidence that this not always the case), then those who happen to think longer term will leave more offspring to the next generation than those who don't. Lather, rinse, repeat a million times: so long as intelligence is selected for, intelligence will result. With no requirement for self-awareness, as far as I can see.

(Chris Gerrib) My contention is that computers and non-sentient beings are very good at following rules. But not following rules get you rocketships and the Internet. Don't get me wrong - once the problem gets narrowed to a "crank the numbers" level, the subconscious may be faster. (Or, it could be that going to sleep frees up enough clock cycles in the brain to do the calculations.)

Then how do you explain recent findings (I cite 'em in the Blindsight endnotes) showing that complex decisions are better handled by the *subconcious* mind, that the conscious self simply can't hold as many variables or perform as complex calculations? We're not talking the rote mechanics of playing the piano or taking a golf shot here: we're talking novel problems the subjects hadn't had to solve before. The group who were prevented from consciously thinking about the problem actually made better decisions than those who were allowed conscious cogitation. Conscious processing proved equal to unconscious processing only for simple problems (1 or 2 variables); for more complex problems (7-11 variables), the subconcious mind kicked consciousness's ass.
Editorial interjection - those are "crank the numbers" calcuations.

Scientific breakthroughs routinely come to people in dreams, slot into conscious awareness fully-formed. The solution to my own M.Sc. came to me in a dream, when I was (by definition) unconscious. The structure of benzene came to its discoverer in a dream This is a well-known phenomenon. These aren't rote behaviors we're talking about: These are conceptual breakthroughs.

(Chris Gerrib) Here's another way to look at it. Humans can sleepwalk, even sleep-eat. These are fairly complicated but routine behaviors, being performed without the conscious mind being engaged. Ever hear of anybody sleep-cooking? Successfully sleep-driving? If you want complex activities, living things seem to need consciousness.

Yes! Yes! And no! People have driven across town, committed murder, driven back home and climbed into bed again, asleep the whole time! (At least, IIRC, one such person managed to convince a jury of as much). For that matter, I'd say we *all* sleep-drive, even when we're awake. Who among us is consciously aware of all the times they turn the wheel, ease off on the brake, hit the gas? Am I the only one who's only realised that he's driven to the wrong destination *after* he's arrived, because the brain went on autopilot during the trip itself? (Yes, this is an example of an ingrained, rote behavior-- but it's also exactly the same example that you just cited as a behavior too complex for nonconscious control.)

(Chris Gerrib) Again, Peter's forgotten more about evolution then I'll ever know, but everything I've read suggest that the smarter the animal, the bigger and more developed the cerebral cortex. If the cerebral cortex is the source of "consciousness" maybe animals are more "conscious" then we thought. But if you have to snap a line in the sand, my vote is for controlling fire.

I actually think we do vastly underestimate the extent of animal cognition. Anybody who's ever looked at a cat probably shares my very strong impression that there's something autonomous looking back: Descartes was an idiot to describe all nonhumans as nonconscious. And the brain structures most fundamentally associated with the conscious state are present in everything from boney fish on up (sharks and rays, not so much).

I actually think that nature is *full* of sentient species. But the thought-experimenter in me can't help asking how all that self-awareness actually increases fitness in a Darwinian sense, and I can't come up with a satisfactory answer. I'd really like to. There may be no other issue on which I would more like to be proven wrong. But I keep postulating an organism every bit as smart as us but without self-awareness, and those thought-experiments always end up with us getting our asses handed to us.

One might even wonder about the sociopaths who seem to rise to the top of such high-profile professions as law, medicine, and politics. These folks are highly intelligent -- the "poor impulse control" aspect doesn't necessarily characterise the whole group, just the less-competent psychos that we baselines are able to detect-- and lack any sort of capacity for empathising with another human being. They appear to bne neurologically incapable of putting themselves into someone else's shoes, of feeling what it's like to *be* someone else. Maybe that's because they have less of a sense of what it is to be *themselves*, hmm?

And according to some recent pop-science studies I've encountered, they have greater reproductive success than the rest of us. They are fitter, in a Darwinian sense. So maybe that experiment is being played out even as we speak.

It would certainly explain a great deal about the current US administration...

In your court...

P.

I think we'll have to agree to disagree.

A housekeeping note - I've re-tagged all entries related to my ongoing debate with Peter Watts as Blindsight so I can track them.

Peter, in his reply (posted earlier today) says "planning doesn't require sentience." He sites as an example computerized chess programs. He's correct. I was in error in my post, which focused on planning. (He's been ruminating on this subject for years, I just got started last week.) Before I move on, I will point out that my day job for the last ten years has been herding computer networks large and small, and I know exactly how that computer chess program works - it calculates every possible move. It's a brute force approach that requires at least a 33 MZ chip (33 million calculations / second) to work.

Peter points out that the conscious mind gets inputs "predigested" (my words) from the subconscious. Therefore, what exactly does the conscious mind do? Let's stick with the computer chess example. The computer can calculate all possible variations of moves and the probabilities thereof. But until a computer reaches up and tweaks the player's nose, it ain't sentient. The computer can't plan outside the rules of chess - it can't think outside the rules of the game - or "think outside the box." I would argue that the first time we can prove humans thought outside the box was when they controlled fire. There have been others - agriculture, metalworking and animal husbandry come to mind.

My contention is that computers and non-sentient beings are very good at following rules. But not following rules get you rocketships and the Internet. Don't get me wrong - once the problem gets narrowed to a "crank the numbers" level, the subconscious may be faster. (Or, it could be that going to sleep frees up enough clock cycles in the brain to do the calculations.)

Here's another way to look at it. Humans can sleepwalk, even sleep-eat. These are fairly complicated but routine behaviors, being performed without the conscious mind being engaged. Ever hear of anybody sleep-cooking? Successfully sleep-driving? If you want complex activities, living things seem to need consciousness. Again, Peter's forgotten more about evolution then I'll ever know, but everything I've read suggest that the smarter the animal, the bigger and more developed the cerebral cortex. If the cerebral cortex is the source of "consciousness" maybe animals are more "conscious" then we thought. But if you have to snap a line in the sand, my vote is for controlling fire.

So I emailed yesterday's post to Peter Watts. Here (edited slightly to remove the greetings and email headers) is his reply:

I *welcome* debate. I love arguing. And truth be told, I find my own ruminations on this whole sentience thing to be extremely unpalatable. I really hope you're right and I'm wrong.

But if you are, it ain't because of fire.

The way I see it, what you've done is show that some kind of higher reasoning process must have been necessary to override intinctive fire-aversion responses. Fair enough. You also make a good case that those reasoning processes have to be able to look several steps ahead, i.e, recognise that long-term gain sometimes necessitates short-term pain. What you *haven't* done is show me why those processes would have to be self-aware.

This is what you said:


But fire is different. The instinct of every land animal (except man) is to avoid fire. Fire kills. So, to master the use of fire, somebody (actually, probably a lot of somebodies with burnt fingers) had to make a conscious decision to override their instinct. Not only that, they had to do so with a plan. The ape-man who said "fire pretty, me want" and grabbed it died. The ape-man who had a vision of fire as a useful tool (sometimes) succeeded. Consciousness, then, gives humans the ability to make long-term plans. You don't build a spaceship without the ability to do long-term planning.

What you've done here is slip consciousness into the argument as though it were impossible to make a "plan" or have a "vision" without it. True, you don't build a spaceship without the ability to do long-term-planning: but who ever said that planning process had to involve sentience? The simplest commercial chess program is smart enough to sacrifice its own pieces in pursuit of checkmate; chess is the archetypal long-term planning scenario. Yet no one has ever claimed that these programs were sentient.

Perhaps more fundamentally, recent peer-reviewed research strongly suggests that even *we* do our best complex problem-solving on an unconscious level, that our brains perform better at multivariable analysis when the conscious parts are actively excluded from the process.

The unconscious mind can handle more variables, perform more complex analyses, is far faster, and metabolically much cheaper than the conscious mind. Further, the conscious mind seems to get its input not from the sensory systems directly, but premassaged after those inputs have already been interpreted by nonconscious modules. So what does consciousness actually do, other than *observe* the results of the subconscious mind's analysis?

And why wouldn't the nonconscious parts of the mind figure out uses for fire much more rapidly than the self-aware homunculous? How do we know it hasn't?

Cheers,
P.

My reply is coming soon.

In my December 23 post, I raved about Peter Watt’s new novel Blindsight calling it “Wicked Good” and “a novel of ideas.” One of the ideas in the book is that sentience (or consciousness) is not needed for intelligent life. The book doesn’t explicitly define “intelligent life” but seems to agree with the commonsense definition of “building technological artifacts” like spaceships. I disagree with the idea that sentience isn’t needed for intelligent life, and in fact argue that it is vital.

Now, Watts has a Doctorate in Biology, making it safe to say that his big toe forgot more about biology then this History / Business major will every know, so I approach this argument with trepidation. Watts makes a very vigorous and cogent argument at his web page in the “long-winded version” of the book’s endnotes. (footnote 19 is especially interesting, leading to various optical illusions pointing to how flawed consciousness is.) But to summarize the argument (any errors are mine) he says that consciousness has no evolutionary advantage. My response is simple: fire.

Fire, or rather the mastery of fire, gave the early humans (Homo Erectus, starting about 2 million years ago) a HUGE evolutionary advantage. You don’t get spaceships without fire, and you don’t get fire without consciousness. First, let’s look at the non-controversial part of that statement (fire is good), then I’ll discuss why I think consciousness is required to get there.

Control of fire gave humans a number of evolutionary advantages, which I will discuss in no particular order. First, it allowed humans to cook food. This increased the types and digestibility of foodstuffs, and (since heat kills bacteria) made the food safer to eat. Dysentery kills right quick. Second, animals avoid fire, because it’s dangerous. So fire helped keep humans off the dinner menu. Third, humans evolved in tropical Africa. We can still die of hypothermia in weather as warm as 50 degrees F.

Moreover, fire requires fuel. Since there are only so many dead branches on the forest floor, tools were needed to cut trees and shrubs for fuel. This leads to improved tool making skills. Adding to the pressures for expansion and tools were the increase in population. More mouths meant you needed to get good at hunting and gathering. In short, there is no way Homo Erectus was able to expand out of Africa without fire.

Watts might use the bee argument at me. Individual bees have an instinct to spit out wax while turning in a circle. Since they’re all doing the same thing, these circles get squashed into hexagons, which are ideal for maximum space with minimum structure. In short, the individual bees have no idea why they do what they do, but it works.

But fire is different. The instinct of every land animal (except man) is to avoid fire. Fire kills. So, to master the use of fire, somebody (actually, probably a lot of somebodies with burnt fingers) had to make a conscious decision to override their instinct. Not only that, they had to do so with a plan. The ape-man who said “fire pretty, me want” and grabbed it died. The ape-man who had a vision of fire as a useful tool (sometimes) succeeded. Consciousness, then, gives humans the ability to make long-term plans. You don’t build a spaceship without the ability to do long-term planning.

To answer the question posed in the header, if you want to create high tech, you need sentience.