A belief in the power and imminence of the Drexlerian vision of radical nanotechnology is part of the belief-package of adherents of the view that an acceleration of technology, linked particularly with the development of a recursively self-improving, super-human, artificial intelligence, will shortly lead to a moment of ineffably rapid technological and societal change – the Singularity. So it’s not surprising that my article in the IEEE Spectrum special issue on the Singularity – “Rupturing the Nanotech Rapture” – has generated some reaction amongst the singularitarians. The longest response has come from Michael Anissimov, whose blog Accelerating Future offers an articulate statement of the singularitarian case. Here are my thoughts on some of the issues he raises.
One feature of his response is his dissociation from some of the stronger claims of his fellow singularitarians. For example, he responds to the suggestion that MNT will allow any material or artefact – “a Stradivarius or a steak” – to be made in abundance, by suggesting that no-one thinks this anymore, and this is a “red herring” that has arisen from “journalists inaccurately summarizing the ideas of scientists”. On the contrary, this claim has been at the heart of the rhetoric surrounding MNT from the earliest writings of Drexler, who wrote in “Engines of Creation” “Because assemblers will let us place atoms in almost any reasonable arrangement , they will let us build almost anything that the laws of nature allow to exist.” Elsewhere, Anissimov distances himself from Kurzweil, who he includes in a group of futurists who “justifiably attract ridicule”.
This raises the question of who speaks for the singularitarians. As an author writing here for a publication with a fairly large circulation, it seems to me obvious that the authors whose arguments I need to address first are those whose books themselves command the largest circulation, because that’s where the readers are mostly going to have got their ideas about the singularity from. So, the first thing I did when I got this assignment was to read Kurzweil’s bestseller “The Singularity is Near”; after all, it’s Kurzweil who is able to command articles in major newspapers and is about to release a film. More specific to MNT, Drexler’s “Engines of Creation” obviously has to be a major point of reference, together with more recent books like Josh Hall’s “Nanofuture”. For the technical details of MNT, Drexler’s “Nanosystems” is the key text. It may well be that Michael and his associates have more sophisticated ideas about MNT and the singularity, but while these ideas remain confined to discussions on singularitarian blogs and email lists, they aren’t realistically going to attract the attention that people like Kurzweil do, and its appropriate that the publicly prominent faces of singularitarianism should attract the efforts of those arguing against the notion.
A second theme of Michael’s response is the contention that the research that will lead to MNT is happening anyway. It’s certainly true that there are many exciting developments going on in nanotechnology laboratories around the world. What’s at issue, though, is what direction these developments are taking us. GIven the tendencies of singularitarians towards technological determinism, it’s a natural tendency to assume that all these exciting developments are all milestones on the way to a nano-assembler, and that progress towards the singularity can be measured by the weight of press releases flowing from the press offices of universities and research labs around the world. The crucial point, though, is that there’s no force driving technology towards MNT. Yes, technology is moving forward, but the road it’s taking is not the one anticipated by MNT proponents. It’s not clear to me that Michael has understood my central argument – it’s true that biology offers an existence proof for advanced nanotechnological devices of one kind or another – as Michael says, “Obviously, a huge number of biological entities, from molecule-sized to cell-sized, regularly traverse the body and perform a wide variety of essential functions, so we know such a thing is possible in principle.” But, this doesn’t allow us conclude that nanorobots built on the mechanical engineering principles of MNT will be possible, because the biological machines work on entirely different principles. The difficulties I outline for MNT that arise as a result of the different physics of the nanoscale are not difficulties for biological nanotechnology, because its very different operating principles exploit this different physics rather than trying to engineer round it.
What’s measured by all these press releases, then, is progress towards a whole variety of technological goals, many very different from the goals envisaged for MNT, and each of whose feasibility at the present time we simply don’t know. I’ve given my arguments as to why MNT actually looks less likely now than it did ten years ago, and Michael isn’t able to counter these arguments other than by saying that “Of course, all of these challenges were taken into account in the first serious study of the feasibility of nanoscale robotic systems, titled Nanosystems…. We’ll need to build nanomachines using nanomechanical principles, not naive reapplications of macroscale engineering principles.” But Nanosystems is all about applying macroscale engineering principles – right at the outset it states that “molecular manufacturing applies the principles of mechanical engineering to chemistry.” Instead of work directed towards MNT, we’re now seeing other goals being pursued – goals like quantum computing, DNA based nanomachines, a path from plastic electronics to ultracheap computing and molecular electronics, breakthroughs in nanomedicine, optical metamaterials. Far from being incremental updates, many of these research directions hadn’t even been conceived when Drexler wrote “Engines of Creation”, and, unlike the mechanical engineering paradigm, these all really do exploit the different and unfamiliar physics of the nanoscale. All these are being actively researched now, but not all of them will pan out and other entirely unforeseen technologies will be discovered and get people excited anew.
Ultimately, Michael’s arguments boil down to a concatenation of ever-hopeful “ifs” and “ands”. In answer to my suggestion that, if MNT like processes could only be got to work at low temperatures and ultra-high vacua, Michael says “If the machines used to maintain high vacuum and extreme refrigeration could be manufactured for the cost of raw materials, and energy can be obtained in great abundance from nano-manufactured, durable, self-cleaning solar panels, I am skeptical that this would be as substantial of a barrier as it is to similar high-requirement processes today.” I think there’s a misunderstanding of economics here. Things can only be manufactured for the cost of their raw materials if the capital cost of the manufacturing machinery is very small. But this capital cost itself mostly reflects the amortization of the research and development costs of developing the necessary plant and equipment. What we’ve learnt from the semiconductor industry is that as technology progresses these capital costs become larger and larger and more and more dominating for the economics of the industry. It’s difficult to see what can reverse this trend without invoking a deus ex machina. Ultimately, it’s just such an invocation that arguments for the singularity seem in the end to reduce to.
Thanks for the response, Richard.
The word “Singularitarian”, as defined in the Singularitarian Principles (2000), basically just means someone that encourages the pursuit of smarter-than-human intelligence. Nothing less, nothing more. It’s worth pondering on how innocuous this is.
With his 2005 book, Kurzweil hijacked the term “singularitarian” and has tried to apply it to his highly complex, occasionally doubtful claims. I reject this redefinition, and identify with the older, innocuous definition.
Besides semantic differences, there are actually distinct groups associated with each. I identify with the former group (which is relatively small, only consisting of maybe a thousand people) and don’t identify with the latter group (which may consist of a large percentage of people who read Kurzweil’s book).
You may criticize Kurzweil if you’d like, but the point I’m making is that there are many “singularitarians” (in the 2000 definition sense) that never bought into all of his claims.
There’s some reference that addresses the assembler point, I’ll look for it.
No tendency towards technological determinism here. It seems like you’re not consistently responding to my response here, preferring to switch over to other people that are easier to attack — those that believe every advance in nanotechnology is necessarily a step towards inevitable assemblers. I don’t believe that they are, though certain nanotech developments suggest the eventual creation of assemblers more than others.
Yes, biological nanotechnology does not necessarily mean that MNT based on mechanical principles will work. I totally understand this 100%. But obviously, considering that I feel that MNT based on mechanical principles is probably possible, it makes sense that I consider biological MNT better evidence for mechanical MNT than you do.
It seems that many of your arguments against MNT could have been applied 10 years ago anyway. I’ll go back and look at them, but I think many of these points have been addressed in debates with the CRN people. “Addressed” not as in one side definitively won, but that arguments on both sides have been presented and people are free to make their own choice.
You seem to be implying a probability of near-zero that mechanical MNT is possible, which is a much stronger claim than mine, which is that both types of MNT may be possible, and we lack the information to make any definitive judgements one way or the other.
When Drexler wrote that MNT was “mechanical engineering principles applied to the nanoscale”, he also implied that these principles would need to be significantly modified to address the physical differences at the nanoscale. Any physicist approaching the problem of engineering nanoscale systems recognizes this a priori, and it’s odd to imply that Drexler didn’t.
MNT and runaway intelligence enhancement are two entirely distinct things. Even if some others blur these together in their minds, I certainly don’t. You’re attacking a strawman if you think that I am invoking superintelligence to solve the problems of MNT. In your response here, you seem to repeatedly attack Kurzweil (who is easier to argue against) rather than people like myself and Drexler (more difficult). Sure, Kurzweil is better-known, but if you’re bothering to write a post to respond to my arguments, then addressing Kurzweilian arguments is out-of-bounds.
The capital cost of machines that can self-replicate using sunlight and readily available materials would be extremely small, yes. So even if the cooling and vacuum requirements are great, these requirements could potentially be met economically. Capital costs in the semiconductor industry are very high because the equipment cannot self-replicate from raw materials.
Michael, perhaps I’m having trouble criticising your arguments because I’m not sure what they are. You say “I feel that MNT based on mechanical principles is probably possible” – why is what you feel an argument? 15 years ago, much less was known about how biological nanomachines actually work, so given that state of knowledge the existence of biological nanomachines was significant evidence for the possibility in principle of mechanical MNT. Since then, there’s been a huge advance in our knowledge of how cell biology works, with structures obtained for many of the most impressive biological nanomachines, single molecule biophysics studies analysing their operation (see, for example, my recent post about the ribosome), and theory and simulation, all of which make clear that these machines don’t work on mechanical principles at all. It’s in the light of this new knowledge, rather than what you feel about the feasibility of mechanical MNT, that we need to reconsider the degree of support that the existence of biological MNT offers to the feasibility of mechanical MNT.
As to how you need to judge specific technical criticisms of MNT, it’s not a matter of “arguments on both sides have been presented and people are free to make their own choice”. If people really want to make a judgement they need to equip themselves with an appropriate level of knowledge in the relevant disciplines, particularly surface chemistry and solid state physics, and look in detail at how the claims and estimates in Nanosystems stack up against what’s now known about nanoscale physics and chemistry. And, once again, a huge amount more is known about this than in 1992, on issues like surface reconstruction in diamondoid clusters and nanoscale friction, to give just two examples.
If/when Intel starts making chips out of diamond, the pace of diamond nanoscale research (not necessarily surface site-specific) will be quickened. It would be nice if there were some industry bigger than SPMs (revenues of maybe $100 million/yr?) to sustain the kind of research that would be required to make actuators suitable for manufacturing, say, R.Freitas’s X-bar tool-tip and handle. Existing research grants aren’t big enough to support such a project, even assuming it was doable and we knew how to do it.
Diamond electronics is a niche interest at the moment; there are some applications like radiation detectors where the material’s radiation hardness provides an advantage that outweighs other disadvantages. And there’s an interesting line of research developing that uses controlled point defects in diamond as the sites of qubits for quantum computing. But the type of carbon I think Intel is going to be interested in now is graphene.
I agree graphene. But diamond might still hold its own in niche applications like spacecraft computer chips or if you’d want chips as sensors in an extreme environment like drill-heads or something, where its thermal and wear properties are superior to flimsy graphene/CNTs. Maybe alternate piezo-actuator applications are a better segway to an SPM surrogate. A researcher last year came up with a desktop mini wind-turbine that uses a piezo-actuator (I think a polymer magnet too) instead of “standard” wind turbine designs. Dont’ know if the morphology of the piezo-actuator is anything like an SPM piezo-tube stack. Maybe next-gen hearing aids? I can see diamond products being valuable enough to be MNT-able assuming MNT, but never MNT-ed piezo-electric goods over conventional manufacturing techniques.
When I do a literature search of nanotech research around the world, what I notice is the vast amount of work going into developing nanotech thats based on solution-phase chemistry (synthetic biology being one example of this). Such “wet” nanotech is being developed in both university and private labs all over the world – Asia, North America, and Europe. I see almost no work going into “vacuum-phase” chemistry and the development of “dry” nanotech.
This suggests to me that “wet” nanotech will make its appearance in the next few decades. “Dry” nanotech will occur later, if it is even possible. It is much more likely that many of the benefits we want (life extension, better and cheaper manufacturing) will be realized by “wet” nanotech and the “dry” stuff will be regarded as a historical footnote like the Babbage machines of the late 1800’s. This is what I think will happen.
Brian Wang had a thoughtful approach, which was to identity what we want, then to go about getting it using whatever technology that works, rather than to obsess over the development of “dry” nanotech and A.I. I like his idea of singularityLite, which uses more prosaic technologies such as biotech and robotic construction technologies.
As of yet, I am not convinced of the possibility of dry nanotech. Even if it is possible, I think “wet” nanotech will be more profound in that it works with a far greater number of the elements in the periodic table than the “dry” stuff is supposed to do.
I think A.I. is a bit further off than most people think and will be much less significant when it gets developed. The reason why I believe that is because of the way our brains work as well as the expected future development of semiconductor technology.
First, our brains. Brains do not work like semiconductor chips. The dendritic connections that make us us are dynamically reconfigurable. That is, they delete and regrow themselves all the time. This is how learning occurs. Also, human memory is not electronic. It is chemical. The dendrites are not only dynamically reconfigurable. They vary by chemical type. This, no doubt, has implications for how human memory occurs. No computer system or even proposed computer system is designed on the basis of these characteristics.
Second, semiconductor technology. Currently, semiconductor chips are made by the deposition, patterning, and etching of layers of thin film materials. This technology will reach its limits in the next 10-15 years (somewhere between 22nm down to the 10nm design rule). The next fabrication technology will be self-assembly chemistry. In other words, computer chips will be grown, rather than carved as they are now. This is more like how biology works. Already, some of the thin-film layers are being processed using self-assembly chemistry. As the technology progresses, more and more of the vacuum process technology will be replaced by self-assembly chemistry. Eventually (around 2030), the entire electronic system will be grown, rather than built. The biological processes that govern the design and functionality of “wet” nanotech will also be the same processes to build our computer systems. In other words, the scenario is the reverse of what the singularity people expect. Instead of our nano fabrication technology becoming more “machine-like”, our machines and computers will become more “biology-like”.
At this point, the chemically-mechanical dynamic characteristics of dendrites will be incorporated into our computer designs. This is when we are likely to create sentient A.I., not before. Is it not likely, then, that any such sentient A.I. will be more like our biological brains, in design and functionality, than current semiconductors? One would think so.
One thing I notice about transhumanism and especially Singulatarianism is that neither groups respond well to criticism, unlike the greater scientific body at large, who generally welcome it.
I’ve just been told by E.Yudkowsky and R.Hanson not to post on their blog at Overcoming bias, affiliated with Oxford University.
Err, when I have a chance to use someone else’s internet or public/private, I tend to cram in posts that would otherwise be spread out. This seems like pretty common behaviour from what I can tell, some people post a lot on weekends.
I’ve been banned from the CRN blog (I assume), sl4 and now a blog affiliated with a British institution that has been ranked as the world’s leading centre of Philosophy. Nothing personal, but many Singulatarians and Transhumanists have already made up their minds about many issues using facts derived from 1970’s/1980’s guesses about minds, policy and chemistry, that have been proven wrong. I’d urge those in these communities to actively seek debate and analysis with mainstream AI, health policy and nanotech communities. And to modify their beliefs accordingly, just as scientific hypothesees would be were the dialectic to poke holes in existing scientific thoughts.
Probably the idea that abacuses are conscious, health-care should be all private and diamonds don’t reconstruct, made some sense decades ago. But there is no method for updating beliefs in these communities. I gave the deprogramming my best shot for the sake of impressionable minds. I’m mildly disappointed in myself more than anything as unless religion, there is no candid admission a leap of faith is needed in these belief systems.