|
||||||
|
Eric Drexler is apparently at the Renaissance Weekend with the intent to speak to the assembled interesting people about how “advanced nanotechnology can address the climate change problem providing low-cost solar energy and by removing accumluated CO2 from the atmosphere.” In the same spirit, for the rest of us, here’s how I think we should go about using advanced nanotechnology to address the problem of climate change:
Robert A. Freitas Jr., author of the Nanomedicine series of books, has just published a major new theory paper on aspects of medical nanorobot control, providing an early glimpse of future discussions of this topic that are planned to appear in Chapter 12 (Nanorobot Control) of Nanomedicine, Vol. IIB: Systems and Operations, the third volume of the series (still in preparation). The paper is part of an edited book collection, available for purchase from Amazon, on bio-inspired nanoscale computing that was published about a week ago by Wiley. Freitas’ contribution to the book is the following chapter:
The chapter is about 5.2 MB in size and a draft preprint version may be downloaded from Freitas’ nanomedicine website. From the abstract:
The nanomedicine books remain freely available online here, with links to MNT-based medical nanorobot designs here. There’s a paper on roboethics by Yueh-Hsuan Weng of Taiwan’s Conscription Agency in the International Journal of Social Robotics that has gotten a write-up on Physorg (h/t to Accelerating Future). Here’s the abstract: There’s a post on Technology Review’s blog about a paper on arXiv about a theoretical result in magnetic memories. Current-day magnetic memory is already “nanotechnology” under the loose definition, involving 5-nanometer particles of cobalt (having about 50,000 atoms). The authors have shown that a single molecule consisting of a cobalt dimer sitting on top of a benzene ring would have a high enough magnetic anisotropy to store a bit magnetically.
(surprisingly enough, the cobalts prefer to stack up rather than so lie down flat on the carbon ring.) Don’t expect this in your computer any time soon; the authors write:
But they then go on to show that the bit could be written (reading is relatively easy) by a scanning-probe like tip which briefly ionized the upper cobalt. The mechanisms to do that would of course still have to be designed and built; but this is exploratory engineering in atomically-precise mechanisms, and we’d like to see more of it. The nominations for Foresight’s 2009 Feynman Prize will be closing soon, so if you know someone who has done outstanding work to advance the goal of molecular nanotechnology, please visit the Instructions Page
One thing I was at some pains during my recent visit to Willow Garage was the likely impact of Moore’s Law on the course of robotics development in the next few years. This is of great interest to a futurist because if computation is a bottleneck, it will be loosened in a well-understood way over the next decade or so, and we will have robots of rapidly-improving capabilities to look forward to over the period. After all, the skill, ingenuity, and technological base was available in 1900 to build steam-powered robots of human size, range of motion, and other physical characteristics (think of watchmakers and the rapidly-burgeoning capability of industrial machinery of the day). What was lacking was sensing and control. I got mixed signals at WG. On the one hand, it was clear that the real bottleneck today is software: “We don’t sit down to have discussions about whether we should hire another person or put a bigger computer on the robot,” one person told me. The value added is clearly with the increased talent at analyzing, programming, or whatever. On the other hand, I also heard a description of a vision/ranging module where implementations on (the current) standard processor versus a GPGPU version were compared: 13 seconds per frame versus 60 frames per second. The fast version wasn’t better in the sense of getting more detail or recognizing more objects — but it was faster, and in a regime that bumped it from much worse than, to somewhat better than, human real-time performance. My personal take on this is that in many fields, the advances due to algorithms have matched those due to raw processing power, and that robotics is in a position to take advantage of both. WG’s open-source strategy is great for leveraging their resources in this area while benefitting the field as a whole. Over the course of the 2010s, as robots get better able to cope with domestic environments and are more widely used there, the pressure for them to be more robust and adaptive, to learn and exhibit common sense, will increase. The big breakthrough in robust 2-D navigation came from Hans Moravec’s Bayesian grids, which changed the whole style of navigation from efficient symbolic — but brittle — algorithms to robust but computationally brute force ones. My intuition is that a similar revolution awaits in virtually everything the robot does and thinks about, and Moore’s Law will make it feasible. Over at the Moral Machines blog, there’s a pointer to an AP story about the recent DC train crash:
The post is just a news clipping, and offers no interpretive comment, but I think some is perhaps appropriate. Train crashes have been happening regularly for over a century. They are not something new that has anything at all to do with AI or machine ethics or any similar concern. They are, however, a reminder that there is something very important that is often overlooked in the popular concern about the increase in technological impact on our lives. And that is that technology already has a huge impact on our lives, and has done since the industrial revolution — and the first, most important concern we must have is to make sure that the technology we have works properly, as intended. Unless I am completely mistaken and deluded, there was and is nobody associated with the DC train system who wanted the crash to happen. It’s not a question of morality at the level of bad intentions, either of people or machines. It was, in simple terms, a case of incompetence. It may have been of design, or of management, or of implementation, or maintenance. It may have been software or hardware. Most likely it was some combination. But the bottom line is simple: things didn’t work the way they were supposed to. The modern world is full of movements that are overly concerned with motivations, and it is passe to worry about whether whatever cause you’re espousing will actually accomplish the grand goals that are claimed for it. Bluntly put, people are too concerned with other peoples’ wishes, which are none of their business, and not enough concerned with other peoples’ competence, which is very much a legitimate concern. You’ll find that for things that really matter — like not having train wrecks — people pretty much all want the right thing already. After hearing an excellent talk by Willow Garage president Steven Cousins at PARC last Thursday, I wangled a visit to the company Monday and talked to a few more people. Willow Garage is a research robotics company in Silicon Valley which has a unique mission for a start-up. They are oriented to making an impact on the field of robotics rather than making an immediate profit. Cousins explained it in these terms: the average robotics PhD student spends 90% of his time building a robot and the remaining 10% extending the state of the art. If Willow Garage succeeds, those numbers will be reversed. Thus the WG design is very general and very robust, designed to be very hard to break and also fairly safe in the hands of an experimental, buggy, program. It’s a gorgeous piece of hardware. In a move that resonates strongly with Foresight, their software is open source. Wearing my futurist hat, I asked several researchers at WG how strongly near-term improvements in processing power, a la Moore’s Law, would affect the performance of their robot. By and large they didn’t seem to think it was critical. The present bottleneck appears to be software — ideas, algorithms, integration, and experience. The OSS community may be able to make a significant contribution here. And because it’s open source, anything you add would redound to the world at large and not just the company. Rosie the robot maid by 2020? I wouldn’t bet against it. Suppose there were a class of technologies called millitech: science and engineering that could be measured in millimeters, from say about a tenth of a millimeter to 100 millimeters — in any dimension. That includes hairs, paper, pebbles, marbles, anything you can hold in the palm of your hand, anything less than 4 inches thick no matter how long or wide it is. This would be, frankly, an insane classification on which to base regulations of whatever technology you had in mind. We regulate hair from appearing in our food in restaurants; we regulate cell phones from operating on the wrong frequency; we don’t legally regulate golf balls but the USGA does. Regulating hair and 2-inch steel plate the same because they were both “millitechnology” would be nuts. Similarly, the ridiculously broad term “nanotechnology” is a monumentally silly basis for regulating things. Nano-powders and particulates are substances — regulate them according to their toxicity, residence time in the atmosphere, biodegradability. Something real. An article this past weekend on Nanowerk reports on a study about attitudes toward regulation of nanotechnology among nanoscientists and the general public:
Confounding the study, of course, or any study like this, would be the fact that what the researchers are calling “nanotechnology” and what the public thinks it is are two different things. And of course anyone writing in the Journal of Nanoparticle Research is likely to be about as far from a notion of nanomachines, even nanoelectronics with no moving parts, as anyone in the field. To my mind, this is just another piece of evidence that the word nanotechnology has broadened to the point where it is more a hindrance than a help in understanding what’s really going on and how the future of technology may develop. This is essentially a follow-on to yesterday’s post about increasing intelligence (you might want to go back and read the comment by Michael A.). The main idea behind that essay was that intelligence consists of a varied lot of skills, which we’re building one at a time (or at least in separate efforts). When we build a formal, mechanical version of a given skill, we don’t save it to be part of a single huge AI system as if we were building the Forbin Project, but deploy it directly in the form of a software app or machine controller or accounting practice or whatever is appropriate. It gets hooked into the existing huge network of information processing and feedback/control function that forms civilization. A century ago, that network consisted almost entirely of human brains and ink-on-paper records and messages. The telephone and telegraph were decades old, and just becoming integrated into society. Today, not only are most messages and memories run by machine (and see how many more of them they are), but there are hugely many more of them than there were a century ago. But the important point is that at no time was any single person (or machine) doing a significant part of the total cognitive work. Even Edison’s biggest invention was the research lab — where he put hundreds of people to work inventing. A single human is not really an effective thinking machine. A feral child who manages to survive in the absence of language-speaking elders winds up not being able to learn language at all. Of all the ideas, concepts, thoughts, and so forth we use individually, only a tiny fraction are original. Almost everything we know, everything we are, is absorbed from the culture around us. In very strong sense, our minds are not our physical brains, but the cultural software running on them. Although there is an element of the personal in memories and aptitudes, by and large the same software would run as well on someone else’s physical brain. Or, when we figure out a bit more about it, on a computer. Building an AI is hard (mental) work. We have been at it for the better part of a century (including cybernetics and the development of computers themselves). I’ve personally worked on it for 35 years. We’re getting tantalizingly close to being able to build a machine which can do the kind of thinking that an individual human can. But when we do build such a machine, it will account for one more human’s effort in the acceleration of the progress of AI. Only when we have as many such machines as we have AI researchers will they significantly increase the rate of progress. Interestingly enough, most of the advantages a machine mind might have in being a better AI researcher than a human would be had in even greater degree by a program that wasn’t structured to emulate a human mind at all — things like not being distracted by personal or ego concerns, and in particular not having to decide what it wanted to do next. It would just be the problem-solving parts, and it would get all of its motivations from outside. It would be plugged into the network of civilization directly, primarily at the call of human researchers. In the meantime, other parts of the network currently done by humans will continue to be transferred to machines. Take driving. There’s no need to make a chauffeur robot with a physical body wearing a uniform and a cap: just build it into the car (and take out the controls, for a more comfortable seat). Does it need an ego, a social life, a wife and kids and cat and dog and mortgage? No. What it does need is telepathic contact with other chauffeur robots, courtesy of wireless internet or whatever, in its area and a semi-autonomous, semi-collective algorithm that steers all the cars knowing what all the other cars are going to do. Likewise most of the other AIs we build. They’ll be built for a purpose, and that purpose will almost always be better served by plugging them into the overall information-providing and goal-setting fabric of our increasingly interconnected civilization. And by he time there are enough AIs that their total thinking and inventive capacity is such as to rival that of our civilization (e.g. to accelerate AI research enough to improve themselves), they’ll be our civilization. Jamais Cascio has an article in the current Atlantic about how humans are getting smarter. This is the best article on the subject I’ve seen in the mainstream press, and better than most in the transhumanist corner of the web. Cascio’s main point is that, as we’ve always done, we build our technology to make ourselves smarter. It doesn’t matter if it’s drugs or PDAs or Google, the technology (and the ideas it embodies) makes us effectively smarter. This has been true since the invention of writing, or longer. Cascio has one point that transhumanists such as Michael Anissimov disagree with. Cascio writes:
and Anissimov replies at length.
Happy to oblige. The bottom line is that AI simply isn’t going to appear all at once in a single, anthropomorphic system. Intelligence is a huge, complicated mass of knowledge that is being invented one little piece at a time. A lot of the pieces are the very kind of thing that Cascio talks about in the article, which we are applying piecemeal to improve our own intelligence. Chances are that in a decade or so we’ll have enough of the pieces worked out that it will be possible to put them all together in system which evaluates, controls, selects, and otherwise manages them so as to act like an integrated, human-style intelligence. But that won’t be an easy task — and at the same time we humans will be using the pieces ourselves, and doing just the parts of the managing puzzle that we’re best at. Human-level intelligence is a moving target. The key part of AI yet uninvented is the fluid, intuitive, estimating, connection-making, higher-level manager that controls the formal, boiled-down, automatable skills that form most current AI. But that’s exactly the form Cascio claims the human mind is moving toward (since we’re putting the rest on silicon). Machines have a huge advantage over human brains when it comes to hard, symbolic, calculation. Now I’m not among those who believe that the nebulous intuitive stuff can’t be done by computer, but I do think that to do it, you’re going to have to use some pretty brute-force methods and the machine-to-neuron advantage will shrink considerably. Will pure machine intelligence pass us, individually? Almost certainly yes, because a wild-type human has a fixed amount of processing power, and the machines won’t have such limits. Will the machines surpass us as a civilization? No — because they’re part of civilization. We build machine intelligence specifically to make ourselves, collectively, smarter. As you can see, we’re in the process of upgrading nanodot (and the main website will follow). Please bear with us while we work the kinks out. This Physorg story gives the details, hat tip to Sander Olsen…
Last week I posted a story of strange behavior in the simulation of molecular machines. One commenter asked if this was due to something unusual in the starting configuration of the atoms. This was the first thing we investigated, and didn’t seem to be the case. There was a small amount to strain energy in the assembly, which promptly thermalized, but this was a minor, one-time, and very brief warm-up, whereas the puzzling one was much slower, but accelerated over time (i.e. over a nanosecond, which is a very long time in the world of molecular mechanics). What finally seemed to be going on was this: I had built a model that was physics-like as far as entropy was concerned, i.e. it conserved information and was reversible; but not as far as energy was concerned, so there were pathways from low to high energy states and vice versa. Now a real physical system will seek states of higher entropy, but because it can’t just take energy from nowhere, the only high-entropy states available to it are ones characterized by more disorder. But in my system, there were pathways to hotter states, which are much higher entropy than colder ones. So the system evolved into the hotter ones by the second law of thermodynamics, blithely ignoring the first law. So what’s the moral of the story? Is it that you can’t trust computer models? No, there are some trustworthy computer models, but there are some out there that are trusted and shouldn’t be (pre-crash financial risk models spring to mind). The point is that a model is like a scientific theory: it has to be tested by controlled experiments, or else it’s just a conjecture. Even though you think you understand the microscopic dynamics in the standard reductionist way, things you think will average out often don’t, giving you a system with radically different macroscopic behavior than the real world. One of the constraints laid down by DARPA at the recent Physical I have long held something of a nuanced view on this point. On the On the other hand, the need to interact with a robot body and cope Even a simulated world can turn out to have assumptions built in which The point of molecular dynamics is to simulate the atoms as if they The point of the story is that when you take a group of atoms with no The problem is, of course, that the numerical simulation doesn’t quite We were trying to simulate molecular machines, however, and one So I tried to write a simulator which conserved energy at the very So what happened? It turned out that you could take any random But I had an ace up my sleeve: since I had made the system reversible, I finally figured out what was going on, but I’ll let readers chew About a month ago, the web was all agog over the announcement of DARPA’s Physical Intelligence program — Wired wrote:
I’m currently at the Proposer’s Workshop for the program, and it turns out that what they’re actually talking about is a lot more like cybernetics. The “thermodynamics” they are talking about is a bit more like the entropy in information theory (Shannon, you will remember, was a student of Wiener, founder of cybernetics). The term cybernetics itself isn’t much used anymore but the reason is more historical than anything else — there was a strange soap opera that broke up the intellectual cadre of cybernetics in the 50s for personal reasons, and computers and symbolic AI stepped into the vacuum, but the core discoveries are still valid. There’s a chapter about cybernetics in Beyond AI, including the soap opera. Brian Wang over at Next Big Future has an article about improving the properties of aluminum as a structural material by filling with buckytubes, the way plastics are made stronger by filling by fiberglass. This isn’t particularly new: what’s new is that Bayer appears to be able to make nanotubes in enough quantity to make this an affordable structural material instead of a laboratory curiosity. The Nanomanufacturing Summit, held in Boston recently, was largely what you would have expected — near-term bulk-tech approaches to nanostructured materials, some interesting research aimed at new electronics, and so forth. Notable, however, was a plenary talk by M. C. Roco, who appears to have changed his tune to the extent of predicting nanorobotics and “molecular nanosystems” (see slide 5) and “Hierarchical nanomanufacturing with atomic / molecular precision” (slide 29). Is the “nanopants winter” nearly over, then? Don’t hold your breath, but we do see signs of a thaw… Hat tip to Howard Lovy. Preserving Security and Civil Liberties in the Sensor Age Palo Alto, CA — June 8, 2009 — A new open source-style project to promote Open Source Sensing has been started, with the goal of bringing the benefits of a bottom-up, decentralized approach to sensing for security and environmental purposes. “The intent of the project is to take advantage of advances in sensing to improve both security and the environment, while preserving — even strengthening — privacy, freedom, and civil liberties,” said Christine Peterson, coiner of the term ‘open source software.’ “We have a unique opportunity to steer today’s emerging sensing/surveillance technologies in positive directions, before they become widespread.” “Cheap, ubiquitous sensing has the potential to turn the worlds of privacy and civil rights upside-down,” said Brad Templeton, a futurist and civil rights activist who chairs the Electronic Frontier Foundation. “No easy solution stands out, but the quest for an answer to these problems — by learning from the bottom-up approaches of the open source community — may provide some water in the desert.” Participation is welcome from individuals and organizations, both non-profit and for-profit. The project is coordinated by Foresight Institute, a non-profit 501(c)3 organization focused on transformative technologies. Link to website: Contact: |
||||||
|
Copyright © 2009 the Foresight Institute - All Rights Reserved |
||||||
