Hijacking Nanotechnology Terminology Again?
In the early 80s, and the great scientist and engineer Eric Drexler came up with the term “nanotechnology†to describe a manufacturing technology that builds products from the atoms up. Around the turn of the century, the term was hijacked to mean anything involving nanometer-scale features, like modern computer chips. Technically, this means you could use the word “nanotechnology†to mean anything, because practically everything has nanoscale features that play a role in its overall properties. The result is that the original meaning of the word “nanotechnology†went kaput, and nanotech enthusiasts had to start saying “molecular nanotechnology†or “molecular manufacturing†to refer to what they were talking about.
Around 2001 or so, the Center for Responsible Nanotechnology started using the term “nanofactories†to describe desktop molecular manufacturing units. Now it seems like a group of researchers is attempting to hijack this word too, even though I’m sure they well know that the word already has an established meaning. From ScienceDaily:
The list of side effects on your prescription bottle may one day be a
lot shorter, according to researchers at the University of Maryland’s
A. James Clark School of Engineering.That’s because instead of taking a conventional medication, you may
swallow tiny “nanofactories,†biochemical machines that act like
cells, first conceived of at the Clark School.For example, these ingested nanofactories, using magnetism, could
detect a bacterial infection, produce a medication using the body’s
own materials, and deliver a dose directly to the bacteria. The drug
would do its work only at the infection site, and thus not cause th
side effects that may arise when an antibiotic travels throughout the
body in search of infections.William Bentley, professor and chair of the Fischell Department of
Bioengineering at the Clark School, and several graduate students
including Rohan Fernandes, have developed this “magnetic nanofactoryâ€
concept and published their research in Metabolic Engineering in
December of last year. Colleagues around the country voiced their
support for the technology in Nature Nanotechnology last month.
Artificial cells are not nanofactories! A “nanofactory†is a desktop manufacturing system! Why does the mainstream constantly steal cutting-edge terminology and water it down? My guess is that the word “nanofactory†is being used here instead of “artificial cell†or “nanobot†because a “factory†sounds more benign and neutral. People might not want to think of the idea of autonomous little robots in their bloodstream, so “nanofactories†sounds better. But they’re stepping all over the prior use of the term! Researchers know how to use Google, and I'm sure they saw the term on other websites, but they just didn't really care.
Nifty Nuclear Blast Maps
That's what the radius of destruction would look like if a 10 kT nuke were detonated on top of my house! Put in your own zip code, and see how bad it would be for you.
I found this page by following a link from NTI, the global security organization founded by Ted Turner. Warren Buffet is another billionaire who supports NTI and encourages his shareholders to read books and watch films about the threat of nuclear terrorism.
You can order a free DVD of Last Best Chance, a film warning against nuclear terrorism, by visiting here.
Another blast calculator can be found at this URL.
Toroidal Colony
The pictured colony is certainly a big one. Kalpana One is currently my favorite space colony design, in terms of relative feasibility and usefulness. One might ask, "what's the point of spending tons of money on building a space colony when Friendly AI could build us one for free, and when unFriendly AI could easily take down such a colony?" The reasons are, 1) governments will spend money on space colonization whether we want them to or not, so we might as well keep an eye on the field, 2) space colonies are an insurance policy against pre-AI disasters, 3) the prospect is inspiring in general, and even if such colonies are never produced en masse in the real world, they'll still be featured in the fictional worlds we choose to inhabit.
The pictured colony looks really, really huge. Probably would weigh trillions of tons. Seems to be about 50km across at the torus, maybe 1000km across total.
1,000,000 Views
Earlier today, this blog had its 1,000,000th view. This is out of approximately 300,000 unique visitors since it was started on November 9th, 2005. Unsurprisingly, the top referrers have been StumbleUpon, Reddit, and Digg. I really appreciate everyone who has taken the time to see what I have to say. I will keep blogging for as long as I have a computer and and Internet connection!
The Psychology of Security
Although this may be old hat to some of you already familiar with heuristics and biases, an excellent article by Bruce Schneier on the psychology of security is available here. It starts as follows:
Security is both a feeling and a reality. And they're not the same.
The reality of security is mathematical, based on the probability of different risks and the effectiveness of different countermeasures. We can calculate how secure your home is from burglary, based on such factors as the crime rate in the neighborhood you live in and your door-locking habits. We can calculate how likely it is for you to be murdered, either on the streets by a stranger or in your home by a family member. Or how likely you are to be the victim of identity theft. Given a large enough set of statistics on criminal acts, it's not even hard; insurance companies do it all the time.
We can also calculate how much more secure a burglar alarm will make your home, or how well a credit freeze will protect you from identity theft. Again, given enough data, it's easy.
But security is also a feeling, based not on probabilities and mathematical calculations, but on your psychological reactions to both risks and countermeasures. You might feel terribly afraid of terrorism, or you might feel like it's not something worth worrying about. You might feel safer when you see people taking their shoes off at airport metal detectors, or you might not. You might feel that you're at high risk of burglary, medium risk of murder, and low risk of identity theft. And your neighbor, in the exact same situation, might feel that he's at high risk of identity theft, medium risk of burglary, and low risk of murder.
The difference between the feeling of security and true security, and the difference between pursuing one thing or the other, is central to the Lifeboat Foundation's mission. For example, planetwide risks like synthetic life or unfriendly AI should be analyzed more thoroughly and given more effort than prevention of nuclear proliferation, even if we consider the near-term probability of the former scenarios to be less, simply because their scope is so much larger. For more on this topic, see Cognitive biases affecting judgement of existential risks.
Alien predators are more dangerous than native predators
This is according to a recent University of New South Wales study. From the Eurekalert article:
Introduced predators such as foxes and cats are twice as deadly as native predators to Australia’s unique native animals, a new study has found.
The new finding, published this week in the prestigious journal Proceedings of the Royal Society B, is the first confirmation of what has been a long-held hypothesis among scientists.
It also highlights the heavy continuing impact of these predators long after their introduction and that Australia's fauna has been among the hardest hit in the world.
Experts believe that introduced "alien" predators are more dangerous than native predators because their prey, such as numbats and bettongs, are naïve to the hunting tactics of alien predators.
This is relevant to discussions of AI, intelligence-augmented humans, and synthetic life. Human beings are used to human modes of attack. We think we've seen it all. But we haven't. If novel life forms devise attack strategies we can't neutralize, then we could be in big trouble. The only way to win this game is not to play it at all: by creating a human-friendly AI that can counteract the risk from other potential nonhuman aggressors.
Bite-Size Molecular Nanotech Glossary
Programmable positional assembly: The ability to place individual atoms precisely and in a reprogrammable way is the gateway to super low-cost and high performance manufacturing. This method of manufacturing is frequently referred to as "bottom-up", because it would build products from the bottom up, as opposed to "top-down", the traditional way of manufacturing products.
Molecular assembler: This is the machine that does the positional assembly. Think robot arm, but made out of somewhere between 3,000 and 4,000,000 atoms. Very few engineers or scientists have looked into molecular assembler designs. The most famous are Eric Drexler, Ralph Merkle, and Robert Freitas.
Parallel assembly: Because assemblers would be extremely small, you'd need billions and billions of them to build human-sized products in a reasonable timeframe. In parallel assembly, numerous assemblers would cooperate with one another to build useful products. They would all need to be programmed to work together in an organized way.
Diamondoid mechanosynthesis: Abbreviated as DMS, diamondoid mechanosynthesis refers to the chemical synthesis of diamondoid nanostructures based on positional assembly. One of the challenges of DMS would be removing the numerous hydrogen atoms that cover the surface of diamond structures, depositing carbon atoms in a designated pattern, then restoring that layer of hydrogen. Research is underway on the technology of DMS. You can read more on DMS at this web page.
Nanofactory: Sometimes fancifully referred to as a matter compiler, a nanofactory would be a manufacturing unit, possibly around the size of a microwave, that takes in simple hydrocarbons and electricity, converts them into feedstock, and uses many trillions of molecular assemblers to manufacture human-scale products. Most of the moving parts in nanofactories would be built out of diamond. Because nanofactories could have tremendous technological capabilities, it is important that they be regulated carefully. The regulation of nanocomputers is a particularly significant concern, due to the threat from AI, among other reasons.
Superproducts: Because nanofactories would build diamondoid products from the atoms up, they could arrange them for superior performance. Tiny structural flaws would be non-existent, making missiles and body armor extremely strong. If superproducts go entirely unregulated, anyone with a nanofactory will be able to build computers billions of times faster than today's, engines thousands of times more powerful, and materials 10-100 times stronger.
Toy Goal Systems for Friendly AI
Say that some group of researchers truly understands the processes necessary for intelligence and is able to use their theory to build a seed AI in the next month. Let us say further that the researchers have cracked the basic requirements of Friendly AI, that is, as the AI gets smarter, it maintains an abstract invariant in its goal system that causes it to have a reliable philosophical thrust (but not monomaniacal tendencies) in its actions. I use "philosophical thrust" here in extremely loose terms - I'm not talking about anthropocentric human philosophical belief systems, but simply a set of differential desirabilities that are generated and updated in a nonrandom way. We can further say that the programmers building the AI have solved most of the problems of "structural Friendliness", that is, trivial subgoal stomp scenarios, like the AI trying to convert the Earth into computronium, are no longer the prime threat.
The challenge at this point is "what to wish for" - the Friendliness content instead of the Friendliness structure. And just as the saying warns - "be careful what you wish for" - we have to be especially careful in what we wish for with seed AI. Because seed AI is specifically designed to make itself smarter, and would not be subject to human-style cognitive handicaps like fixed neural mass or single-digit item limits for working memory, an AI of this type can be expected to become very intelligent very quickly after passing a certain intelligence threshold, possibly a level roughly analogous to human intelligence. A more detailed argument in favor of this hypothesis can be found here. Instead of arguing in detail why expecting a glass ceiling in AI intelligence slightly below that of the species H. sapiens is far less probable than a relatively quick transition from roughly human-similar intelligence to transhuman superintelligence, I'm just going to assume that the first seed AI will substantially surpass the intelligence and ability of human beings shortly after matching it. This idea, of going from prehuman or human-level intelligence to vastly superhuman intelligence in a matter of weeks, days, or even hours, has been referred to both as a "hard takeoff" or the "singularity hypothesis". It is generally assumed that after a hard takeoff, the result is a superintelligent AI with sophisticated molecular manufacturing at the very least. A global presence and human-superior technology in all areas are also implied.
In advanced AI, it will pay to be "conservative", that is, not assume that you can pull the plug whenever you want. It's more conservative, and prudent, to assume that highly intelligent AIs will become capable of autonomous movement and self-improvement and pass beyond our direct control. In this view, it is the AI researcher that is responsible for assuring that, even if the AI becomes superintelligent and starts manufacturing advanced robotics to advance its goals, it won't become a threat to the human race or the biosphere.
An AI moving beyond our direct control need not be an inevitable death knell for the human race. In an effort to get a better idea of the problem and potential solutions, let me propose a few seed AI goal systems that are constructive for brainstorming. I stress that these goal systems are for brainstorming only - I don't think that any of them could actually be implemented without extensive work and theoretical progress, and even if they could, they might very well lead to suboptimal scenarios. They are just examples, so please don't take them too seriously.
1) The "Benevolent Mimic" goal system. In this variant, the AI is programmed to copy directly the philosophical belief system and goal structure of the most trustworthy and benevolent known human on Earth. If substantial agreement cannot be found as to who exactly this person is, then a person is located that is widely regarded as a very moral and compassionate individual, and upon meeting them in person, you would be strongly compelled to agree with the assessment. Since the AI could reprogram itself at will, among its benevolent behaviors would be the honing and improvement of its own benevolence, through simulated self-tests and cognitive rewrites. The advantage of the mimic approach is that it would be guaranteed to produce a superintelligence at least as benevolent as that which could be provided by an idealized human upload.
2) The "Benevolent Genie" goal system. For this goal system, we have an AI that strives to act like a genie and fulfill human wishes. It fulfills any genuine requests as long as they do not threaten others or the AI's continued ability to grant wishes. The genie can distinguish between the letter and spirit of wishes, and grants the spirit of all wishes while applying superintelligent common sense and knowledge of nuance to their implementation. The watershed of material abundance from manufacturing technology available after the creation of true AI causes scarcity-based political squabbling and traditional economics to evaporate except to the extent that humans cannot live without them. The genie is widely regarded as a human-superior moral philosopher that consistently finds solutions that please the widest possible number.
3) The "Democracy Engine" goal system. This goal system does not engage in independent moral philosophy, but merely implements the democratic consensus of human beings on a nation-by-nation basis. Under this goal system all the countries in the world would become de facto democracies, if the AI is able to extend its influence globally, which is what we are assuming in this hard takeoff model. Standard "tyranny of the majority" and compounding of voter irrationality concerns apply. There is also the threat of hardwiring some form of 21st century democracy that looks primitive in retrospect, but continues to be enforced for all eternity by the superintelligence. (A risk whenever you build a seed AI that can't reflect on or improve intelligently on its own goal system.) Presumably nations could vote to merge with each other and perhaps evolve into some global system whereby a majority of people on the planet are necessary to implement major decisions.
4) The "Careful Conservationist" goal system. For an AI based on this goal system, extreme care is taken to preserve the "natural human flow of history" and global biodiversity present at the time of its creation. This AI changes the "background rules" just enough to eliminate the most obvious of civilization's banes - for example malnutrition, poverty, crippling diseases, violence, et cetera. Other than that, it is extremely conservative about intervening too far into human affairs. Human intelligence enhancement is released through a slow, gradual process of incrementally better data sharing and neurological therapies and implants. This enhancement is invented by humans, tested by humans, and explained by humans to other humans. The AI intervenes if such implants or therapies cause mortal danger to human beings. Otherwise, it butts out. This is the goal system that might be most favored by people who worry that the deployment of autonomous superintelligent AI would screw up the human way of life.
5) The "Flock of Angels" goal system. For this class of goal system, the superintelligent AI creates billions of "Angels", autonomous subroutines that commune with the people of Earth and try to assess what they "really want". Once this is determined, the appropriate changes are enacted, including the removal or preservation of the Angels as deemed desirable by the populace. Perhaps the Angels return every fixed time interval or whenever there is a crossroads in global development. The postulation of this goal system is designed to counter the objection that true desires can only be conveyed through hands-on social interaction. For an AI that is smart and computationally wealthy enough, direct interaction with human beings is probably not necessary to evaluate desires with a legimately high degree of confidence. However, there may be inherent advantages to the introduction of nonhuman de novo persons designed to resolve conflicts and improve society. Such persons would probably be developed even in the absence of AI, through cybernetics or some other means instead, although this depends on your perspective.
Some theoretically implementable goal systems may be classified as hybrids of these. In considering your possible response to this post, please refrain from pointing out potential failure modes of the toy systems - anyone can easily imagine them. Instead, suggest your own ideas for toy models, or perhaps triangulations between the models which might allow us to see the full spectrum of possibilities more clearly. Assume that the AI in question undergoes a hard takeoff and has fine-grained control of physical reality, even if you believe the gap between human-equivalent AI and superintelligence is fairly large.
Free Money Anyone?
A site I've worked with over the years, WiseGEEK.com, is having a huge writing contest with cash prizes to celebrate the submission of our 10,000th article. Multiple submissions are allowed, and the Grand Prize is $7,500, with 10 runner-up articles receiving $250. We are talking 400-600 word articles that answer simple questions in a clear and concise way.
Ever since I read Bryan Caplan point to the extreme value of entering essay contests, I've always kept an eye out for them. About university essay contests, he writes:
The truth about essay contests is that the number of submissions is usually absurdly low considering the size of the prizes and the opportunity cost of students' time.
Unfortunately, all past and current employees of Conjecture Corporation (the company that runs the site), and relatives/household members are ineligible to enter, so I can't participate. If that doesn't apply to you, then I strongly recommend entering!
Specific Power Graph
This is a graph I just drew. What does it mean? I'm not exactly sure. But it may be important. The topic under consideration is specific power. The dots indicate a possible wall of maximum cooling capacity for stable products.
The cosmic objects and physics experiments aren't really products, they're just there for reference. Note that as size increases linearly, the volume - and therefore total possible power - increases cubically.
I know the graph is not perfectly accurate because I just made it up as a thought experiment. But I'd like to make it more accurate with your feedback.
Developer Platform from Numenta
Numenta has released their NuPIC platform for developers. Jeff Hawkins' release note starts as follows:
"To our developer partners:
Although you may be anxious to download Numenta's software and give it a try, you should read this brief note first. I will give an introduction to what this release is for and what you should expect as you use it.
Why did we create Numenta?
Numenta was formed to develop and promote a technology called Hierarchical Temporal Memory, or HTM. We are confident that the principles underlying HTM are the same principles that govern much of the operation of the human neocortex; thus, HTM enables the creation of machines that have some of the capabilities of the human brain. We see HTM as a fundamental new computing methodology able to solve longstanding problems in artificial intelligence and machine learning. Be sure to read the white paper on our web site that describes the basics of HTM and what it can do."
The white papers are here.
I read On Intelligence and thought that while Hawkins' ideas were interesting, his expository style was underwhelming. His HTM ideas simply remind me of how cortical hierarchial systems in the brain work. Because On Intelligence was a popular book, he kept inserting anecdotes about his thought process leading up to his "discovery" of the concept, and other personal references that got in the way of his explaning his theory. He kept talking about why his ideas were sure to be revolutionary and widely adopted by everyone in the future, which I found to be a turn-off.
I think it's great to build AIs inspired by the human neocortex. But I don't think that human-equivalent intelligence necessarily follows from such efforts. The brain has other components aside from the neocortex. There are probably processing subtleties necessary for implementing true intelligence that we don't yet understand. I'm not saying that we need to describe the operation of every single neuron in the brain to develop a theoretical framework for artificial intelligence that works, just that in the absence of persuasive empirical evidence, we should stop jumping to conclusions about the scalability or real-world effectiveness of a system implemented based on our model of intelligence.
I do plan to read the white paper and other material on the Numenta site, though, and will be watching closely to see what developers have to say about the NuPIC platform. The fact that there are emerging companies pursuing artificial general intelligence shows that we should take the possible consequences of their success seriously.