Slashdot Mirror
Douglas Hofstadter Looks At the Future
An anonymous reader writes with a link to this "detailed and fascinating interview with Douglas Hofstadter (of Gödel Escher Bach fame) about his latest book, science fiction, Kurzweil's singularity and more ... Apparently this leading cognitive researcher wouldn't want to live in a world with AI, since 'Such a world would be too alien for me. I prefer living in a world where computers are still very very stupid.' He also wouldn't want to be around if Kurzweil's ideas come to pass, since he thinks 'it certainly would spell the end of human life.'"
Is it just me or does the Singularity smack of dumb extrapolation to me? "Progress is accelerating by X, ergo it will always accelerate by X".
I mean, if I ordered a burrito yesterday, and my neighbor ordered one today, and his two friends ordered one the next day, does that mean in 40 more days, all one trillion people on earth will have had one?
Done with slashdot, done with nerds, getting a life.
I agree with Douglas, I expect I would be uncomfortably unfamiliar in a world shared with AI beings. Then again, based on my understanding of Kurzweil's Singularity, it's unlikely to affect me much: I plan to live out my life in meatspace, where things will go on much as before.
...And pollution and loss of habitat, but through all that, they still live amphibian lives.
(Also according to my understanding of Kurzweil's projections,) It's worth noting however, that for those willing to make the leap, much of the real growth and advancement will occur in Matrix-space. It's an excellent way to keep "growing" in power and complexity without using more energy that can be supplied by the material world.
Here's my analogy explaining this apparent paradox: Amphibians are less "advanced" than mammals, but still live their lives as they always have, though they are now food for not only their traditional predators but mammals too.
In fact, I can't help but wonder how many of us will even recognize when the first AI has arrived as a living being. Stretching the frog analogy probably too far: What is a frog's experience of a superior life form? I am guessing "not-frog". So I am guessing that my experience of an advanced AI life-form is "whatever it does, it/they does it bloody fast, massively parallel, and very very interesting...". Being in virtual space though, AI "beings" are likely only to be of passing interest to those who remain stuck in a material world, at least initially.
Another analogical question: Other than reading about the revolution in newspapers of the day, how many Europeans *really experienced* any change in their lives during the 10 years before or the 10 years after the American revolution? We know that eventually, arrival of the U.S. as a nation caused great differences in the shape of the international world, but life for most people went on afterward about the same as before. The real action was taking place on the boundary, not in the places left behind.
(Slightly off topic: This is why I think derivatives of Second Life type virtual worlds will totally *explode* in popularity: They let people get together without expending lots of jet fuel. I believe virtual world technology IS the "flying car" that was the subject of so many World's Fair Exhibits during the last century.)
I know, I know... Asimov's laws, etc etc. But... for a being to be sentient and at the same time reach the same level of thinking that we enjoy, you must given them the freedom to think, without any restrictions... as humans (ostensibly) do. This requires a level of both bravery and of careful planning that is far greater than we as humans are capable of today.
I'm not predicting some sort of evolutionary re-match of Cro-Magnon v. Neanderthal (where this time the robots are the new Cro-Magnon), but it does require a lot of careful thought, in every conceivable (and non-conceivable) direction. When it comes to building anything complex, it's always the things you didn't think of (or couldn't conceivably think of given the level of technology you had when designing) that come back to bite you in the arse (see also every great engineering disaster since the dawn of history).
Best bet would be to --if ever possible-- give said robot the tools to be sentient, but don't even think of giving them any power to actually do more than talk (verbal soundwaves, not data distribution) and think.
It reminds me of an old short story, where a highly-advanced future human race finally created a sentient device out of massive resources, linked from across every corner of humanity. They asked it one question to test it: "Is there a God?" The computer replied: "There is... now."
Quo usque tandem abutere, Nimbus, patientia nostra?
Also to add to that, there is no requirement for us to understand the brain in depth, only that we understand how we learn, and in that respect we've come leaps and bounds over the years. Plus, let's not limit ourselves to the human brain. For instance, a dog is intelligent too. A piece of software with the intelligence of a dog could be very useful. Hell one with the decision making abilities of a bird would be a nice start. And on and on..
On a tangent:
Intelligence is such a broad word, and then to tack on Artificial. AI lacks a precise meaning and if anything needs to be done in the world of AI, it's to create a nomenclature that makes sense and provides a protocol of understanding.
For many the word AI simply means "human brain in a jar" but that's just one small branch of AI sciences. But where is our Fujita Scale of artificial intelligence? Where is out toolkit of language (outside of mathematics)?
I ask this seriously btw, if any of you know about work on this please post a response.
Actually, even if it kept accelerating, singularities (as some fancy world for when you divide by zero, or otherwise your model breaks down) so far never created some utopia.
The last one we had was the Great Depression. The irony of it was that it was the mother of all crises of _overproduction_. Humanity, or at least the West, was finally at the point where we could produce far more than anyone needed.
So much that the old-style laissez-faire free-market-automatically-fixes-everything capitalism model pretty much just broke down. There just was no solution to how much a country should produce. Hence my calling it a singularity.
By any kind of optimistic logic, it should have been the land of milk and honey. It was actually _the_ greatest economic collapse in known history, and produced very much misery and poverty.
And the funny thing is, the result was... well, that we learned to tweak the old model and produce less. We still go to work daily, and a lot of companies still want overtime, and a whole bunch of people still are dirt-poor. We just divert more and more of that work into marketing, services and government spending. It's a better life than the downwards spiral of the 19'th century, no doubt. But basically no miracle has happened, and no utopia has resulted. The improvement for the average citizen was incremental, not some revolution.
That was actually one of the least destructive "singularities". Previous ones produced stuff like, for example, the two world wars, as the death throes of old-style colonialism. When the model based on just keeping expanding into new territories and markets reached the end, we just went at each other's throats instead. A somewhat similar "singularity" arguably helped the Roman Empire collapse, and ushered in a collapse of trade and return to barbarism. The death throes of feudalism created a very bloody wave of revolutions.
All the way back to the border between Bronze Age and Iron Age in Europe, where... well, we don't know exactly what happened there, but whole civilizations were displaced or enslaved, whole cities were razed, and Europe-wide trade just collapsed. Ancient Greece for example, although most people just think of it as a continuous "Greece", had a collapse of the Mycenaean civilization and Achaean language it had before, and after some 300 years of the Greek Dark Ages, suddenly almost everyone there speaks Dorian instead. The Greeks and Greek language of Homer, are not the same as those of Pericles. (An Achaean League was formed much later, but apparently had not much to do with the original Achaeans.) And, look, they displaced the Ionians too in their way.
We recovered after each of them, no doubt, but basically the key word is: recovered. It never created some utopian/transcendence golden age.
So, well, _if_ our technology model ends up dividing by zero, I'd expect the same to happen. There'll be much misery and pain, we'll _probably_ recover after a while, and life will go on.
A polar bear is a cartesian bear after a coordinate transform.
I think that most people who want AI for pragmatic reasons are essentially advocating the creation of a slave race. You think companies/governments are going to spend billions of dollars creating an AI, and then just let it sit around playing Playstation 7 games? I doubt it. They'd likely want a return on their investment, and they'd force the program to do their bidding in some manner (choosing stocks, acting as intelligent front ends for advanced semantic search engines, etc). Maybe this would involve an imperative built into the AI at ground level: "obey your masters", or it could be more obviously sinister like a pain/pleasure reward system like the ones used to control human slaves.
Do you think that mainstream society would find this as repugnant as I do? I doubt it. Most people seem to find it difficult to empathize with other humans who have a different skin color, a different religion, or a different sexual orientation. If Average Joe doesn't care about the individual rights of people in Gitmo, he's certainly not going to care about the individual rights of a computer program- which is not even a biological life form.
I would say that any serious AI research needs to be preceded by widespread legislation expanding the definition of individual rights (abandoning the "human rights" label as anachronistic along the way). We need to insure that all sapient beings- organic or digital- have guaranteed rights. Until then, I think AI researchers are badly misguided- they're naive idealists working towards a noble goal, without considering that they're effectively working to create a new slave race...
BUT, I think that his chapters on math and physics and their interface (everything prior to the biology chapters) constitute the SINGLE GREATEST and only successful attempt ever to present a NON-DUMBED DOWN layperson's introduction to mathematical physics. I gained more physical and mathematical insight from that book than I did from any other source prior to graduate school. For that alone, I salute him. Popularizations of physics a la Hawking are a dime a dozen. An "Emperor's new mind" having (what I can only describe as) 'conceptual math' to TRULY describe the physics comes along maybe once in a lifetime.
His latest book is the extension of that effort and the culmination of a lifetime of thinking clearly and succinctly about math and physics. He is the only writer alive who imo has earned the right to use a title like "The road to reality: a complete guide to the laws of physics".
As for Hofstadter, GEB was merely pretty (while ENM was beautiful), but essentially useless (to me) beyond that. Perhaps it was meant as simply a guide to aesthetic appreciation, in which case it succeeded magnificently. As far as reality is concerned, it offered me no new insight that I could see. Stimulating prose though - I guess no book dealing with Escher can be entirely bad. I haven't read anything else by Hofstadter so I can't comment there.
I am far more interested in digitally enhancing human bodies and brains than creating a new AI species.
Consider this: throughout the eons of natural and sexual selection, we've evolved from fish to lizards, to mammals, to apes, and eventually to modern humans. With each evolutionary step, we have added another layer to our brain, making it more and more powerful, sophisticated and most importantly, more self-aware, more conscious.
But once our brains reached the critical capacity that allows abstract thought and language, we've stepped out of nature's evolutionary game and started improving ourselves through technology: weapons to make us better killers, letters to improve our memory, mathematics and logic to improve our reasoning, science to go beyond our intuitions. Digital technology, of course, has further accelerated the process.
And now, without even realizing it, we are merging our consciousness with technology and are building the next layer in our brain. The more integrated and seamless communication between our brains and machines will become, the closer we get to the next stage in human evolution.
Unfortunately, there is a troubling philosophical nuance that may bother some of us: how do you think our primitive reptilian brain feels about having a frontal lobe stuck to it, controlling its actions for reasons too sophisticated for it to ever understand? Will it be satisfying for us to be to our digital brain as our primitive urges and hungers are to us?
It's not moving goalpots at all: it's a total failure to take even the smallest step towards machine sentience, but any inuitive definition. Something key is missing. It's not like we've made software that's as smart as a hamster, and now we're working on making it as smart as a dog.
The field of AI research has taken tasks that were once thought to require sentience to perform, and found ways to perform those tasks with simple sets of rules and/or large databases. Isn't even the term "AI" passe in the field now?
It's not moving the goalposts, it's simply a clarification of what sentience means: some level of self-awareness. Even a hamster has it, but no software yet does.
Socialism: a lie told by totalitarians and believed by fools.
Some people seem to think that the singularity will result in a matrix like virtual world, which wouldn't impact on the real world. This is simply not right. As by definition the singularity is the point at which we can't know or understand what's really going on, then there will be real world consequences that may be staggering. Imagine if the singularity figured out that all thinking was a subset of a larger mind, and then pushed a button to connect it all, permanently. We would become 'one' with the whole universe. Sounds a bit wanky I know, but it's that kind of thing we're talking about, not just a good version of the internet with a neural interface. More likely the result will be something that we simply can't conceptualise rather than the example above. Something that we just couldn't imagine no matter how smart we are or how we try. Imagine being an ant coming across a jet engine. What does it make of it? That will be us versus the singularity, and I suspect it will have the same effect as a jet engine would have on an ant if it were to pass through it. The rate of change is getting faster. More people are getting technofear as the rate increases. I think the singularity might happen over days or even hours when it happens, with the world/universe/dimensions/whatever_else_we_can'_think_of maybe changing in the blink of an eye. This is based on the idea that the singularity is unknowable, and will change things as radically as can be changed, and I can't think higher than that. I don't mind it happening, but it is the end of my life as I run it. I'd just like to get a bit more drinking time in before it.....
I couldn't disagree more. I was so enthused by this book that I went to University to study AI. After a couple of years of that I decided that what I was being taught was a load of rubbish and that as Penrose had claimed, "machines" (i.e. computational devices such that exist today) could not "think" (i stil graduated with a first however and it was some use to me in my future career). The problem I had with Hofstadter was that he assigned the concept of recursion an almost magical property. Dennett does a similar thing with his "multiple drafts" theory. They may in themselves be enough to describe complex functioning in the brain (or any other system come to think of it), but as Chalmers points out, at present a Materialist model of thought (or rather, consciousness, which we assume is required for thought) is impossible. I find Penrose and Hameroff's ideas of conscious action in the brain to be both fascinating and intuitively correct, even if the evidence does not exist at present. I noted with interest that scientists have recently discovered large-scale quantumn effects in the leaves of plants when photo-synthesizing. However, any such action in the brain will be difficult to pin down, for obvious reasons. I expect the science of consciousness to progress rather more slowly than other fields for this reason.
What's to add? But since I'm always ready with a slap at Kurzweil, I feel that Hofstadter has him pinned:
1. "Ray Kurzweil is terrified by his own mortality", and
2. "Rather ironically, [Kurzweil's] vision totally bypasses the need for cognitive science or AI"
It is exactly this complex and elusive puzzle of "I" and "consciousness" Hofstadter explores that Kurzweil hopes we can conquer without having to think about it at all. Which I scorn as "magic science".
I have to say I find the cyberpunk vision more appealing than Hofstadter. It would be "the end of humanity as we know it." I'm not sure it would be "the end of human life." It might be evolution. I just think it is many hundred years in the future at the most "optimistic" (depending on your viewpoint).
For a class project, I once created a genetic algorithm to evolve a Reversi-playing algorithm (Reversi is also known as Othello). I coded the system not to be able to consider more than X moves in advance, because I wanted to prevent it from using "computer tricks" (i.e. I didn't want it looking farther ahead than a typical human could do with a moderate amount of practice). I tried playing with that number just to see what would happen, but I eventually left it at 4.
By the time I was done with my evolving system, it could evolve in 4 days (using 4 ~2Ghz Intel servers and an island genetic model, for those who know about genetic algorithms) an algorithm which could handily and consistently beat myself and all of my friends.
The interesting thing here is that I didn't even "initialize" it with a basic strategy or any personal training -- it started with randomly-generated strategies (most of which were no better than randomly placing pieces in legal squares). It then played against itself for those 4 days, learning through trial and error (as opposed to training by playing against a human). By the end, it had learned enough without human feedback that it could defeat a group of fairly intelligent (though not very practiced) humans at Reversi.
I never analyzed the generated programs enough to fully understand how they worked, but I did inspect them a little. Each evolved algorithm consisted of no more than 40 lines of C code (which called various global helper functions such as get_opponent_score(), get_self_side_pieces(), etc which I had created). By inspecting algorithms that were able to beat me, I actually learned a thing or two about Reversi strategy.