Thinking About Thinkism Saturday, May 2 2009
AI and singularity and superintelligence 6:13 pm
Last September, Kevin Kelly posted a critique of a hard takeoff Singularity, based on what he calls “thinkism”:
As an essay called Why Work Toward the Singularity lets slip: “Even humans could probably solve those difficulties given hundreds of years to think about it.” In this approach one only has to think about problems smartly enough to solve them. I call that “thinkism.”
Let’s take curing cancer or prolonging longevity. These are problems that thinking along cannot solve. No amount of thinkism will discover how the cell ages, or how telomeres fall off. No intelligence, no matter how super duper, can figure out how human body works simply by reading all the known scientific literature in the world and then contemplating it. No super AI can simply think about all the current and past nuclear fission experiments and then come up with working nuclear fusion in a day. Between not knowing how things work and knowing how they work is a lot more than thinkism. There are tons of experiments in the real world which yields tons and tons of data that will be required to form the correct working hypothesis. Thinking about the potential data will not yield the correct data. Thinking is only part of science; maybe even a small part. We don’t have enough proper data to come close to solving the death problem. And in the case of living organisms, most of these experiments take calendar time. They take years, or months, or at least days, to get results. Thinkism may be instant for a super AI, but experimental results are not instant.
Interesting argument, and well-phrased.
But what about that recent story of a Cornell researcher Hod Lipson’s AI program that independently derived the laws of motion based on the swings of a pendulum, “a feat that took physicists centuries to complete”? According to Kelly’s thinkism hypothesis, that should be impossible.
And Lipson’s program is just the start of a whole field:
The research is being heralded as a potential breakthrough for science in the Petabyte Age, where computers try to find regularities in massive datasets that are too big and complex for the human mind. (See Wired magazine’s July 2008 cover story on “The End of Science.”)
Surely intelligence can achieve a lot. Solutions which take certain thinkers years to discover are uncovered in a short period of time by gifted experts. It is wrong to place solid limits on what a superior intelligence could do — could chimps predict what humans would be capable of with thinking alone? Of course, I could be wrong. Any intelligence might require extensive experimentation to generate knowledge. But the difference between my and Kelly’s position is that he sets hard limits based on speculating about intelligence fundamentally different than his own, while I acknowledge my basic uncertainty and say that I’m not entirely sure what thought alone is capable of. It could be capable of nearly everything, or it could only be capable of what it has achieved so far (a lot).
The conservative stance would be to assume that a new mind might be able to benefit from thought a great deal, and as such we should take great pains to ensure that all artificial intelligences above a certain level have human-friendly motivations. The “we have nothing to worry about, I guarantee it” stance would be to assume that thought is practically useless without thorough experimentation and to ignore the issue of human-friendly motivations on the pretense of indefinite human superiority and control. Sounds like the premise for yet another science fiction film where AIs get the jump on humans because we were overconfident.
Even if experimentation were required to glean knowledge, why would such experimentation be limited by the anthropocentric designation of “calendar time”? A nanoscale pendulum swinging in a vacuum demonstrate the same laws of motion as a large pendulum, and does so in a fraction of the time.
The human brain operates at about 200 Hz. Imagine hypothetical alien cultures where creatures evolved to have brains operating at 20 Hz or 2,000 Hz. Why would their advancement of science necessarily be limited by the “calendar time” of another intelligent species in an insignificant corner of the Milky Way Galaxy? Why would the limitations on knowledge to be gained from experiments be perfectly aligned with the inherent neural firing rates and “calendar time” of Homo sapiens? This is the Copernican Error of human-centrism — “if we’re a certain way, every other possible mind will have the same limitations, guaranteed”.
There are a variety of ways to boost one’s experimental output and data input. One obvious method is parallelism. Humans can only focus on one thing at a time, but an AI mind could focus on an unlimited number of experiments as long as it has the computing power and hardware. Another method would be miniaturization. The use of microarrays in biology research has made possible far less expensive and far more parallel experimentation than ever before.
I see Kelly’s position as anthropocentric triumphalism — we’re the greatest, no one can be as good as us, we have nothing to fear from a hard takeoff, any AI mind will need to engage in centuries of research to get as far as us. Sure, it might turn out to be true, but why put the human species on the line for this hypothesis?
A sentient that develops theories without experimental feedback to prune the list of possibilities will at best create a fantasy world that bears no resemblance to the physical world.
The Libson AI is not an example of thinkism in action. First it is “rigged” in the sense that it was already given a mathematical formalism that works pretty well at describing the real world. Second, the program was performing experiments because it was making real world observations and then using its mathematical framework to fit the data. Without either of these, the AI would do no better than humans did before the invention of algebra, calculus and the scientific method.
Very good points. I would like to see the author address them as I enjoyed his article very much. Both of you have very good points.
From OB, see Einstein’s Speed and That Alien Message.
There is plenty of data already, which has no good theory to explain it. Too much data almost and to little brains.
For any additional data a new experiment can be designed only by the brain power.
What is more, we already have the science needed for the complete nano, it is Quantum Mechanics. We (or SAI) must only find implementations of it. Thinking is needed.
He’s wrong. Firstly, the vision he’s attacking is a rather extreme one. There is a wide range of possibilities between the hard take off he attacks and a take off that’s so slow that we don’t even notice it. You can’t imply the latter by refuting the former.
That aside, his main argument against the value of thought (ironic!) is very much in contradiction to my experience of science and research. So many times in my research I’ve noticed that if I’d been a bit smarter I would have made a decision that would have saved me a great deal of time. In many cases I could have achieved what I managed in a year in just a month or two if I’d been smart enough to work out a few key things: bugs in code avoided, research paths that were dead ends that I could have seen coming, etc. And this is in relatively pure research. In areas like biology and medicine people are swimming in data now. Even at the level of research papers, the volume of information is getting out of hand. A few years back I saw the proceedings of the top haematology conference. It looked like a phone book, with 3 columns per page, small font and I guess a thousand pages long. And this was just the abstracts for the papers, not the full papers, and only about blood, and only for one year. Even a very gifted human intelligence can only skim parts of this growing ocean of information.
I think our biggest problem is actually the opposite of what Kelly claims: we have so much data flooding in about such rich and complex systems that our biggest challenge is to figure out what all this information means. Intelligence is a major limiting factor in research.
“Why would the limitations on knowledge to be gained from experiments be perfectly aligned with the inherent neural firing rates and “calendar time” of Homo sapiens? This is the Copernican Error of human-centrism — “if we’re a certain way, every other possible mind will have the same limitations, guaranteed”.”
Exactly. Our subjective perception of “time” and its rate of passage is mostly an accident of evolution and local astrophysics.
Regarding a superintelligent AI’s ability to gather its own experimental data, well, we already have Adam, the robotic genetic “scientist” that can formulate and test its own hypotheses. If we are indeed able to create an AGI, we could imagine giving it control over an entire scientific research facility covering a variety of fields, stocking it with robotic equipment, and letting the AI order whatever additional equipment it requires over the Internet. Such an AI would also have access through the Internet to the vast amounts of scientific data us humans have already gathered–data that, like Shane said, is already much too voluminous for even very intelligent humans to grasp all of it.
Such a well-equipped AGI would probably get “up to speed” fairly quickly and begin designing novel experiments of its own creation, including methods for self-improvement. An increase in a sentience’s ability to “think,” I believe, would lead directly to a corresponding increase in the capability and efficiency of gathering/analyzing experimental data. Throw in molecular assemblers and the potential for massively parallel, on-demand experimentation is even greater.
The question at that point would then be friendliness: “did we set its initial conditions properly or is it going to use its fancy new facility to design and implement (whether intentionally or not) a method of destroying the human race?” So I really hope the scientists out there designing the self-experimenting AI “robotic labs in a box” of the future are listening to organizations like SIAI.
A lot of research is done using simulations, and a lot of computational power. Little actual experimentation is involved.
Anyway, why would an artificial intelligence be restricted to thinking alone? Surely it would be able to manipulate the real world in order to acquire the experimental results it needed. In fact, many experimentations are actually conducted by programmed computers and machines, because humans are not capable / fast enough / precise enough to conduct them manually and record their results.
Not if, as has been suggested repeatedly, it were sandboxed.
The real problem with Kevin Kelly’s argument is that he assumes one coherent way of knowing. Your pendulum rebuttal was in my head before I read your lines. Computers will not know the way we know. We may not even know the way each other know. Sensemaking will occur. We prefer this or that notation to show symbolic relationships, but they are not unique, nor are they perfect. They are simply what’s in use.
Further, I see no reason why a combinatorial chemistry robot isn’t a key part of the grid of knowledge. (etc.)
It isn’t that Kelly is wrong. He’s just species bound. Again, your alien argument makes the same point.
It is always good to read a rebuttal by someone else that agrees closely with your own reading…it gives a sense that lines of that do exist even if they are not unique or perfect.
Right on Ryan. The “thinkism” concept is rooted in a subjective human-bound view and its utility is limited to general contemporary situations due to that. It also creates a false split between thought and action. Thought is a form of action too, albeit on a much smaller energy scale. Action/intelligence exists on a continuum, just like info-tech/hard-tech, making Kelly’s framing less useful, especially as we expand and zoom the human ability to convert thought into action.
Just a small point (I agree with your overall conclusion):
“Even if experimentation were required to glean knowledge, why would such experimentation be limited by the anthropocentric designation of “calendar time”?”
Because he was discussing solving issues such as “the death problem” with respect to living organisms – for example, an experiment lasting for a minute seems unlikely to provide you with useful new data about what you can do to prolong the life of a human.
“Between not knowing how things work and knowing how they work is a lot more than thinkism. There are tons of experiments in the real world which yields tons and tons of data that will be required to form the correct working hypothesis.”
It isn’t meaningful to divorce thinking from doing. The sum of all the unconscious mental processing, mental imagery, your stream of consciousness, logic and argument, and including your deliberations, actions, learning, and so on are all necessary parts of any useful thought processes.
He’s arguing against an imaginary and meaningless definition of “thinking”.
“Let’s take curing cancer”
We don’t need artificial intelligence or a singularity or even any ‘thinking’ to cure cancer. It is well known that cancer is primarily caused by the WHO immunisation program established by the elite to cover up the fact that they wanted to introduce many degenerative diseases into society. It’s just not widely published. Nearly all the major ‘problems’ of today like wars, famine, poverty, family breakdown are planned and executed by the global elite. We don’t need more ‘thinking’, we don’t need more technology, we need a ‘global elite endectomy’ to remove these guys and stop them from running the planet like their own little kingdom. Everything else to ‘improve the human condition’ is propaganda and a waste of time. Time to wake up people.
[...] “When Software Goes Mental: Why Artificial Minds Mean Fast Endogenous Growth” (2009) “Thinking About Thinkism” (2009) by Michael Anissimov “Technological Singularity/Superintelligence/Friendly AI [...]
Fascinating topic and comments. I also do agree with the perspective that having a greater ability to think endows an entity with vaster capabilities and expediency. For example, as a medical resident training in a hospital, it took me 3 hours of patient interview, 2 hours of interpreting diagnostic analysis and about 3 more hours reading to make a diagnosis of heart failure due to a leaky valve that my expert cardiologist attending was able to do within seconds of walking in the patient room. One could argue that this is knowledge gained from years of experience that cannot be taught overnight. Actually, this knowledge can be taught in few minutes, which I acquired once it was explained to me. These types of heuristics are very common in many fields, familiar to experts but not to laypersons. Once we are able to create AI with near equivalent human intelligence, they will learn them all and dramatically increase the expansion of intelligence.
howdy, noble blog on blubbery loss. alike helped.
I do trust all of the ideas you have offered for your post. They’re really convincing and can certainly work. Still, the posts are too short for starters. Could you please extend them a little from subsequent time? Thanks for the post.
A lot of thanks for creating so very good blog! i’ve book marked it and will probably be back!