Kaj Sotala: Why Care About Artificial Intelligence? Thursday, Jul 12 2007
AI and singularity 3:02 am
Kaj Sotala, a fellow supporter of both the Lifeboat Foundation and Singularity Institute, has published a new article, “Why care about artificial intelligence?” to follow up on his “Artificial intelligence within our lifetime?” article, which I covered in March.
The main thrust of the article is that AIs could potentially be much, much more powerful than human beings, and therefore we have an important stake in how their motivational systems are constructed. The main talking points are:
- Artificial intelligences can do everything humans can
- Limitations of the human mental architecture
- Limitations of the human hardware
- Comparative human/AI evolution and initial resources
- Considerations and implications of superhuman AI
- Controlling AI: Enabling factors
- Controlling AI: Limiting factors
- Immense risks, immense benefits
- Summary and implications
Also recently published by Kaj on his site are the works, “Transhumanism: Happiness, Equality, Choice”, “Ethics of forced choice and future selves”, and “In defense of transhuman development”. The papers are only a couple of pages each, suitable for a quick and informative read.
Thank you Michael for these wonderful infos, your blog is both informative and insightful, keep up the good work!
Problem is, AGI (friendly or not) won’t happen until a few hundred years from now. It’s not something anyone needs to hope for or worry about.
And no, I’m not saying this because I’m a grumpy, old, diseased guy who’s going to die within a few years. I’m only 18, but I’m convinced that my generation was born far too soon.
Want to tell us the reason for thinking so, since you’re so sure about it?
Yes, here’s why AGI won’t happen anytime soon:
(strongest reasons first, slightly weaker reasons later)
1.) We simply don’t have the hardware for it, and we aren’t making any progress. That’s because clock speeds have stagnated. I have a computer next to me that I bought in 2002, and it’s a Pentium 4 at 3GHz (non-overclocked). Now, 5 years later, I can’t even buy a 4GHz processor that’s not overclocked. Clock speeds used to double regularly in the 1980s and 90′s, but there hasn’t even been a 40% improvement since 2002.
2.) There hasn’t been significant progress in AGI (software side) since the late 1960s, which was the start of an AI winter. Narrow AI has made a strong comeback since the early 1990s, but there hasn’t been any major work or progress on AGI. Not only that, but no program has even come close to beating a good human at Go. That’s just a simple board game that any 5 year old could learn, so it’s a huge jump from winning that game to stuff like finding cures for cancer and doing things that no human can do. (Yes, I do know that computer chess programs are better than humans at chess, but Go is closer to AGI, while chess is closer to narrow AI.)
3.) There simply aren’t enough people working on AGI, and that number is declining, not increasing. The number of computer science majors has declined each year since the early 2000s dot com bust, and out of those who major in computer science, only a small number of them are working on AI. Even among that small group, almost all of their focus is on narrow AI, not AGI. In fact, one of my professors mentioned that the number of people in the world working on AGI is less than the number of students at a typical college.
4.) Even if the hardware were available and people were making significant progress towards developing AGI, governments (not only the U.S) would pass laws that prohibit the development of AGI. If stem cell research and gay rights are such a big issue now, there is no way that they’ll allow the development of something that’s much more capable than they are. It’s simply a basic fact. People in government want to have power, so they won’t allow other things (like AGI) that will threaten that power.
And this isn’t something you can do illegally. Unlike building bombs or counterfeiting money, AGI requires more than a few people and a few months to develop. There is simply no way that a few thousand people can work on something illegal and remain undetected for enough time until their mission succeeds.
5.) I know that I’m acting on a hunch and that this is unscientific, but I’m reminded of the saying “If it’s too good to be true, it probably is”. I’m agonistic, and a lot of the implications of AGI makes it sound closer to a religious cult than an actual reality. I may be wrong, but reasons 1-4 are enough to make me think I’m right.
It’d be much more beneficial if people started focusing on issues that matter within this century. Things like preventing bad stuff (runaway global warming, oil scarcity, asteroid impact, etc), and introducing good stuff (space colonization, molecular manufacturing, gene therapy, etc).
1) The processing power of computers are following Moore’s law. The clock speed is insignificant, what is important is the processing power
2) Process has been made in Go and the game will be beaten by computers in the future. There are currently many serious AGI projects (atleast 10), and each year we get more information about one AGI architecture (the human brain)
3) More people are working in AGI now than 5 years ago
4) Governments are always years late in legislation. They will notice AGI projects probably just before we get the first human level AGI, and even then you can just move to another country (it is easy to transport AGI code), or code it underground.
5) Many inventions we take for granted today were seen as “too good to be true” before they were invented.
Thousands and thousands of people are working on the issues of global warming, asteroid impact etc, and only handful people are working on AGI. The benefits of AGI are so great (nearly infinite), that it is rational to use some resources in trying to build it. Another reason is, that the first AGI should be Friendly to humanity, and not many people are taking this risk seriously.
“We simply don’t have the hardware for it, and we aren’t making any progress. That’s because clock speeds have stagnated.”
Try running any modern video game on that computer. Or Windows Vista. Vista requires 1 GB of RAM minimum. As for processing power, IBM’s BlueGene/P came online two weeks ago, giving us our first petaflop (10^15 FLOPS) supercomputer, and a $700 PS3 can be re-engineered to give you 20 GFLOPS.
“2.) There hasn’t been significant progress in AGI (software side) since the late 1960s, which was the start of an AI winter.”
The classical field called “Artificial Intelligence” hasn’t advanced much, but the theory of how to build an AGI has advanced tremendously. On SIAI’s reading list (www.singinst.org/reading), which is quite extensive and covers a huge variety of topics, there is only one item written prior to 1975. Most of it dates from 2000 or later.
“Not only that, but no program has even come close to beating a good human at Go.”
Every time we actually achieve something, people simply move the goalposts and declare “no score”. I have a copy of Ray Kurzweil’s “The Age of Spiritual Machines”, which was published in 1999. It predicts a huge number of things for 2009 which have already been achieved. Among them:
“Cables are disappearing. Communication between components, such as pointing devices, microphones, displays, printers, and the occasional keyboard, uses short-distance wireless technology.”
“Computers routinely use wireless technology to plug into the ever-present worldwide network (the Internet)”.
“Computer displays have all the display qualities of paper- high resolution, high contrast, large viewing angle, and no flicker.”
“While schools are still not on the cutting edge, the profound importance of the computer as a knowledge tool is widely recognized.”
“The majority of reading is done on displays, although the “installed base” of paper documents is still formidable.”
“Learning at a distance (for example, lectures and seminars in which the participants are geographically scattered) is commonplace.”
“Users can instantly download books, magazines, newspapers, television, radio, movies, and other forms of software to their highly portable personal communication devices.”
“The online chat rooms of the late 1990s have been replaced with virtual environments where you can meet people with full visual realism.”
Strange how people tend to count the misses and forget the hits.
“In fact, one of my professors mentioned that the number of people in the world working on AGI is less than the number of students at a typical college.”
The number of full-time researchers working on the Manhattan Project in 1943 was probably less than the number of students at a typical college, and that was a huge government-financed multi-billion dollar effort.
“governments (not only the U.S) would pass laws that prohibit the development of AGI.”
AGI is not on the government’s radar. No congressman, senator, cabinet member, or other government official has even acknowledged transhumanism’s existence (as far as I know).
“There is simply no way that a few thousand people can work on something illegal and remain undetected for enough time until their mission succeeds.”
Criminal organizations all over the world have existed for decades operating in such an environment.
“If it’s too good to be true, it probably is”.
See http://www.singinst.org/blog/2007/07/10/the-power-of-intelligence/.
THANKS yet again, Tom!!
Tom Wrote:
Politically, Fukuyama has in the past been considered neoconservative. He was active in the Project for the New American Century think tank starting in 1997, and signed the organization’s letter recommending that President Bill Clinton overthrow the then-President of Iraq, Saddam Hussein [1].
Dick Cheney is a signatory of the PNAC’s Statement of Principles. There’s a connection there. While not addressing your statement, it’s relevant to it. I don’t think it valid to say that government officials are unaware of transhumanism. But there’s a difference between acknowledgement and awareness.