Slashdot Mirror
The AI Anxiety (washingtonpost.com)
An anonymous reader writes: The Washington Post has an article about current and near-future AI research while managing to keep a level head about it: "The machines are not on the verge of taking over. This is a topic rife with speculation and perhaps a whiff of hysteria." Every so often, we hear seemingly dire warnings from people like Stephen Hawking and Elon Musk about the dangers of unchecked AI research. But actual experts continue to dismiss such worries as premature — and not just slightly premature. The article suggests our concerns might be better focused in a different direction: "Anyone looking for something to worry about in the near future might want to consider the opposite of superintelligence: superstupidity. In our increasingly technological society, we rely on complex systems that are vulnerable to failure in complex and unpredictable ways. Deepwater oil wells can blow out and take months to be resealed. Nuclear power reactors can melt down. Rockets can explode. How might intelligent machines fail — and how catastrophic might those failures be?"
Being free willed does not mean that one cannot or will not necessarily ever make so-called "decisions" that can be predicted from a knowledge of the state, it only means that one is *capable* of it. However, how is the behavior of a system where one is not capable of it any different than one where the state itself is unknowable? You can apply a variation of the halting problem to establish with certainty that either the universe is not deterministic, or else it impossible within its framework to contrive a test that incontrovertibly proves that it is so.
e.g. If I could know what decisions you were making (a notion that is at least theoretically possible, if the universe were truly deterministic), I could analyze it and predict the answer you would give to a particular question, even if I told you truthfully what the answer to that question were going to be.... however, with your so-called illusion of free will, you could utilize the information that I gave you in the present about your alleged future action, and then deliberately contradict it, invalidating the prediction that I made, meaning that my knowledge about the future state was incorrect, which leads one inescapably to the conclusion that even if the universe is deterministic, it is impossible to know it.
And it is noteworthy that by outward present appearances, we appear to have free will with regards to decisions that we make.
So if, by all the standards that can ever be measured, you appear to be a "free-willed" person, then how is your behavior any different than if you actually *were* a free-willed individual? And if your behavior is identical to as if you actually were a so-called "free-willed" being, what purpose does suggesting that you are not free willed even mean?
Plus of course, one cannot advocate the notion that we are not free willed without also suggesting that we abdicate the notion of personal responsibility... but that's a philosophical debate for another time.
File under 'M' for 'Manic ranting'
> making most jobs obsolete.
That's already happened a few times. At one time, most people worked in agriculture- mankind spent most of our time feeding ourselves. An "economic downturn ", therefore, was when a lot of people starved to death.
After machines such as plows and later GPS-guided combines with largely automated food processing plants did most of the work of producing food, inexpensive food was therefore readily available and humans had time to do things not strictly necessary for survival, like education, producing consumer goods, writing and printing books, comfortable clothes, and creating washing machines and dishwashers. With the efficiency of machines, consumer goods including books, and conveniences like washing machines and refrigerators became readily available to the masses.
No longer scrubbing our clothes on a washboard, we then had time to make and play video games.
The replacement of human labor with machines has been a continuous process for over a thousand years, peaking about 200 years ago. In the process, our standard of living has gone from digging for anything we could eat in the winter to stopping by Walmart to select which of the 160 different fruit and vegetable varieties we want to munch on while we enjoy our Netflix movie.
Then so do subatomic particles. You don't need AI if that's all you want. If subatomic particles do not have free will, then neither do humans. This second option allows physics to be Turing Complete and is much more agreeable.
If computers develop sufficient power for intelligence to be an emergent phenomenon, they are sufficiently powerful to be linked by brain interface for the combination to also have intelligence as an emergent phenomenon. The old you would cease to exist, but that's just as true every time a neuron is generated or dies. "You" are a highly transient virtual phenomenon. A sense of continuity exists only because you have memories and yet no frame of reference outside your current self.
(It's why countries with inadequate mental health care have suspiciously low rates of diagnosis. Self-assessment is impossible as you, relative to you, will always fit your concept of normal.)
I'm much less concerned by strong AI than by weak AI. This is the sort used to gamble on the stock markets, analyse signal intelligence, etc. In other words, this is the sort that frequently gets things wrong and adjusts itself to make things worse. Weak AI is cheap, easy, incapable of sanity checking, incapable of detecting fallacies and incapable of distinguishing correlation and causation.
Weather forecasts are not particularly precise or accurate, but they've got a success rate that far outstrips that of Weak AI. This is because weather forecasts involve running hundreds of millions of scenarios that fit known data across vast numbers of differing models, then looking for stuff that's highly resistant to change, that will probably happen no matter what, and what on average happens alongside it. These are then filtered further by human meteorologists (some solutions just aren't going to happen). This is an incredibly processed, analytical, approach. The correctness is adequate, but nobody would bet the bank on high precision.
The automated trading computers have a single model, a single set of data, no human filtering and no scrutiny. Because of the way derivatives trading works, they can gamble far more money than they actually have. In 2007, such computers were gambling an estimated ten times the net worth of the planet by borrowing against predicted future earnings of other bets, many of which themselves were paid for by borrowing against other predicted future earnings.
These are the machines that effectively run the globe and their typical accuracy level is around 30%. Better than many politicians, agreed, but not really adequate if you want a robust, fault-tolerant society. These machines have nearly obliterated global society on at least two occasions and, if given enough attempts, will eventually succeed.
These you should worry about.
The whole brain simulator? Not so much. Humans have advantages over computers, just as computers have advantages over machines. You'll see hybridization and/or format conversion, but you won't see the sci-fi horror of computers seeing people as pets (think that was an Asimov short story), threats counter to programming (Colossus, 2010's interpretation of 2001, or similar) or vermin to be exterminated (The Matrix' Agent Smith).
The modern human brain has less capacity than the Neanderthal brain, overall and in many of the senses in particular. You can physically enlarge parts of your brain, up to about 20%, through highly intensive learning, but there's only so much space and only so much inter-regional bandwidth. This means that no human can ever achieve their potential, only a small portion of it. Even with smart drugs. There are senses that have atrophied to the point that they can never be trained or developed beyond an incredibly primitive level. Even if that could be fixed with genetic engineering, there's still neither space nor bandwidth to support it.
It's a small world and it smells funny; I'd buy another if it wasn't for the money; Take back what I paid (SoM)