Slashdot Mirror
The Singularity Blinds Sci-Fi
foobsr writes "Popular Science has an article discussing the growing difficulties that Sci-Fi writers encounter when it comes to extrapolating current trends. Doctorow and
Stross , both former computer programmers, are rated to be prototypes of a new breed of guides to a future which due to
Vinge's Singularity might not happen for humanity once a proper super-intelligence - maybe as a Matrioshka Brain - has been created."
I actually ran into all of the talk about the singularity by asking the question: What is the meaning of life? More specifically, I asked Jeeves.
The first result he comes up with (this one) is an FAQ on the meaning of life. Part of the question of the meaning of life is an eventual goal, something to reach towards. Once of the options discussed is the Singularity.
The best place for more info is the Singularity Institute. Their definition of the Singularity is the technogical creation of smarter-than-human intelligence. This is by any possible means, either overclocking the human mind, creating artificial intelligence which is smarter than humans, or some combination thereof (such as uploading human minds to computers to run at a faster rate).
Read the FAQ. It'll clear up your basic questions, and doubtless leave you with many more.
The speed of time is one second per second.
Actually, that sort of thing has been done before. Read Daniel Dennett's Where Am I?, it is a great and though provoking read.
A patriot must always be ready to defend his country against his government. -edward abbey
A gentle but fairly thorough taste can be found in Kurzweil's "The Age of Spiritual Machines". Also check out http://www.kurzweilai.net/meme/frame.html?m=1s ing.html
http://www.ugcs.caltech.edu/~phoenix/vinge/vinge-
http://www.aleph.se/Trans/Global/Singularity/
I am sure interested entities can google more.
The "singularity" is one of the favorite wet dreams of the "transhumanists", a group of spoiled adults who seemingly find it difficult to tell reality and science fiction apart.
Indeed, this can be difficult even for scientists who read the physics literature. Much of what was regarded as science fiction in the 50's is fact today, including some things that were generally considered to be fantasy at one time, like beam weapons. Physicists are carrying out serious experiments on quantum teleportation, and methods of transmitting information (random information, but still information) faster than light.
Now there are multiple lines of serious investigation, any one of which that could lead to massive transformation not merely of human culture (such as happened so recently with the internet, and was predicted by hardly anybody), but also of humanity itself:
-AI
-Genetic modification of human beings
-Direct man/machine interfaces
-Nanotechnology
Perhaps any one of these will not pan out. AI progress has moved fairly slowly of late. On the other hand, neurobiology has been booming along, and there seems little doubt that it will eventually be possible to simulate brain function. I can understand why writers are finding it difficult to extrapolate far into the future; it is simply hard to imagine that all of these will stall out.
Turing showed that no such program exists that can solve the halting problem for all possible input programs.
However, it's a big stretch to go from that to debugging software. Even if you show that the halting problem is equivalent to debugging a program (assuming you can define that formally), you still can get around the proof by designing a program that only debugs some programs. There might be a very large class of programs for which the halting problem can be solved, and that could be enough for practical use.
Anyway, I'm just saying that one needs to be very careful applying things like Godel's Incompleteness Theorem or Turing's proof that the Halting Problem is not solvable. Those theorems are extremely formal and don't necessarily apply to practical situations where partial solutions are good enough.
(Another example is the Traveling Salesman Problem. No one knows a polynomial time algorithm that finds the optimal solution, and it is quite possible no such algorithm exists. However, there are polynomial time algorithms that will get you within a factor of 2 of the optimal solution, and I think there are others that get even closer than that.)
"Enough to know that it's a hard problem, maybe not fully solvable in my lifetime."
How do you know? Before powered flight, how reasonable would the description of a 747 have sounded?
Well, 100 billion neurons or so. Given that these guys are building a system today which emulates 20 billion neurons: http://www.ad.com/ human level consciousness might not be all that far away.
Government of the people, by corporate executives, for corporate profits.
I'm surprised nobody has mentioned John Wright's Golden Age series. The article mentions Stross. I'm in the middle of reading Stross's latest novella, in Asimov's, and as far as I can tell, it's meant to be an outrageous parody of Wright. It's actually pretty funny, and Stross also gets his science right (to the extent that he commits himself to anything very specific), whereas Wright appears to have learned all his science from Star Trek, and seems to take himself entirely too seriously.
Find free books.
You ought to read "Permutation City" by Greg Egan. It's about things like this, and takes them to an extreme conclusion.
I believe posters are recognized by their sig. So I made one.
Not really. In a rough order of magnitude basis, a human brain has a hundred billion (1e11) neurons, each with a thousand synapses capable of firing a hudred times per second. The equivalent capacity in a computer would be 1e11 * 1e3 * 1e2 = 1e16 floating point operations per second. A typical desktop computer today has about ten billion (1e10) operations per second, that is, one millionth of a human brain. If Moore's law continues to be valid, the twenty doublings in capacity needed for a desktop computer to overtake human brains will take 30 years.
Justify that a single neuron firing is equivalent in logical processing power to a floating point calculation, and that all the neurons in the brain can fire continuously, without pause, without brain damage, and that all of them firing continuously would constitute some kind of meaningful process, and that that kind of parallelism would be practical for general purpose computing at the same level of performance that you see on your desktop computer, and you'll have an argument. Otherwise, you've got nothing. Sorry.
And you havn't even touched the memory/storage issue.
But I agree with you that all this means nothing if software cannot be developed. Well, in the next decades, the wide availability of human-equivalent hardware will let us try to develop such software.
We already are developing this software. Compression, speach and face recognition, deductive reasoning tools, and so on, are all on the table. These tools do the kinds of things that people do. It's a just a matter of time, a LOT of time, before we learn to combine and enhance these tools in a way that approaches higher-level intelligence.
Upstairs Dog, Downstairs People.
Do the same thing with a computer, then. (Watchdog timers are already extremely common in embedded devices.)
// returns true if program halts given data as input
// does this function halt? // loop forever
Humans can't solve the Halting Problem, either. Does the function Undecidable below halt? I'll even spot you all the hard work in a library function Halts(), and leave you to decide about a two-line program.
extern bool Halts (char* program, char* data);
bool Undecidable (char* input)
if Halts (input, input) while (TRUE)
else return TRUE;
main()
Undecidable (undecidable);
The function takes itself as input. If Halts() returns true, Undecidable loops forever. But we passed in Undecidable as the argument, which means that Halts must have been wrong when it said the program would halt. On the other hand, assume that Undecidable does not halt. Then Halts() would return FALSE, which would cause Undecidable to halt and return.
There's a contradiction both ways, which means that one of the initial assumptions was wrong. That assumption was that Halts() could determine whether or not a program halts.
That's really all there is to it. Note that you don't know from this theorem that, for example, the only programs that are undecidable are ones with just this peculiar self-denying structure. Perhaps they're just an oddity.
The theorem also depends on an assumption of unbounded program size. It is known, for example, that for all programs of a given length, you can always determine whether or not it halts. In fact, you don't even need a Turing-complete machine to do so; a mere finite state machine will suffice. This point makes application of the mathematical theorem to the real (and finite) world somewhat murky.
AI researcher Eliezer Yudkowsky mentions a number of different ways to reach a singularity:
* Computer software endowed with heuristic algorithms
* Artificial entities generated by evolution within computer systems
* Integration of the human nervous system and computer hardware
* Blending of humans and computers with user interfaces
* Dynamically organizing computer networks
Most of the comments so far have concerned the first method, which basically consists of programming a super-smart AI. However, I think that the third and fourth items listed, dealing with the way humans augment their information-processing capabilities, will have the biggest near-term results.
Fifty years after atomic power, there has been very little progress. We can't make fusion work. Fission is too messy. And there's nothing else in the research pipeline.
Don't think solar or wind will help. Here are the actual figures for California for the last twenty years. Solar power hasn't increased over the last decade, and is stuck around 0.03% of consumption. Wind power is at 0.1% of consumption, and the good sites have already been developed.
T-mobile with a Nokia 3660. I can check my POP e-mail from comcast. Cool stuff. Also have a wireless headset, so I think I'm pretty much adopting technology that will be common place by 2010. Can't imagine what 2020 would be like.