Effective Altruism Book Review: Radical Abundance (Nanotechnology)

post by Isnasene · 2018-10-14T23:57:36.099Z · score: 48 (12 votes) · LW · GW · 5 comments


  Book Review: Radical Abundance
  I. Introduction
  II. How APM Works
  A. Advanced APM in  a Nutshell
  B. Mechanical Devices in Nanotechnology
  C. APM as Macro-Scale Manufacturing
  III. APM Will Change Everything
  A. Drexler's Partial Notes on APM Benefits
  B. Drexler's Notes On How APM May Go Wrong
  IV. Pathways to Nanotechnology
  V. Conclusion

Book Review: Radical Abundance

I. Introduction

As a materials engineering major with, roughly speaking, a year of full-time experience with molecular dynamics simulations, I have a special place in my heart for high impact materials both literal and figurative. As a lurker in effective altruism, I was delighted to see something potentially relevant to my experience show up. I made a post in the "Nanotechnolgy in EA" Facebook group expressing/re-iterating the following thoughts:

  1. Nanotechnology will be chaotic and hard-to-predict, requiring massive advancements in computation before effective machines can be designed.
  2. Our technology in terms of actually producing APM seems plausible but very uncertain at this point

From my post, I got some interesting comments suggesting an alternative to 1. along with a recommendation to read K. Eric Drexler's new book on atomically precise manufacturing (APM), Radical Abundance: How a Revolution in Nanotechnology Will Change Civilization. This, in part with some googling, gave me the impression that I was also wrong about 2.

The goals of this book report from least to most significance are: providing a decent summary of Radical Abundance, expressing my own thoughts on what I find to be the most salient aspects of nanotechnology, and clarifying possible common misconceptions that contribute to the frequent confusion about nanotechnology's place in effective altruism.

Anyway, Radical Abundance lightly discusses three inter-woven concepts: how APM would work, how APM could impact the world, and how APM can be achieved. Let's do this.

II. How APM Works

A. Advanced APM in a Nutshell

The description and understanding of APM can be best characterized by Drexler's own words:

APM-based materials processing technology will employ nanoscale mechanical devices that operate at high frequencies and produce patterns of atoms...think of an APM system as a kind of printer that builds objects out of atoms just as a printer builds images out of patterns of ink, constrained by a limited gamut, not of colors, but of output materials.

This clarifies two distinct points that I, as someone working in molecular dynamics and , would miss. First, despite my priors from experience with soft matter molecular dynamics, APM systems will be based on mechanically-inspired rigid mechanisms rather than biologically-inspired soft matter machines. Second, at its heart, most APM will focus on the fabrication of macro-scale materials with nano-scale optimization (i.e. really good materials) rather than nanomachines (i.e. tiny robots), though these will be necessary to some extent in creating APM in the first place and have massive peripheral uses and implications.

B. Mechanical Devices in Nanotechnology

First, while existing examples of APM (i.e. ribosomes) are biological in nature and demonstrate the potential for leveraging nano-scale devices in complex ways, Drexler's APM uses down-scaled mechanical devices like those in modern factories to achieve its ends instead of the chaotic complexity of biology. In order to successfully imitate the behavior of mechanical machines, nanomechanical machines will typically be made of "stable covalent structures that consist of fused rings; among hydrocarbons, small-scale examples include the adamantanes and the somewhat more flexible aromatic molecules." Thinking along these lines simultaneously opens up new possibilities and grounds others. When I think about biological APM, I think of spectacularly complex inter-plays of atoms and molecules superbly optimized in parallel with their environments over millions of years to provide the sufficient but incredibly improbable conditions for life. When I think about mechanical APM, I think about the Saturn V Rocket.

To illustrate the difference, consider the following situation where someone asks me about the future of leveraging nanotechnology to achieve goal a goal in a given environment:

Me thinking about biological nanotechnology: "Of course! We'll use incredibly complex computational tools coupled with automated laboratories to build a black-box machine that does the job you need... At least, we will if we ever figure out how to make computers and laboratories that effective. Once we do though, the sky's the limit!"

Me thinking about mechanical nanotechnology: "Hmm... It looks like achieving your goal also requires some flexible components or at least a fair number of automated computerized systems so I'll need to see if we can get materials to work well with that at the nano-scale... And your environment is a bit different so I might need to select different molecules to make sure they're stable and rigid unless you want to incorporate homeostasis which would add another order of complexity... Give me five years, a few millions dollars and the atomically precise tech I need to build something and I'll let you know."

The difference between these biological and mechanical characterizations of nanotechnology is simultaneously complicated and subtle. Both seem a long ways off in terms of technology but the former demands a clear conceptual leap (much better computers) while the latter requires an extensive but plausible level of technological progress (Improved knowledge of chemistry and existing chemistry techniques). This means that, while the former might never be achieved, the latter will probably be achieved eventually.

A similar point can also be made about potential impact. Biological interpretations of nanotechnology very quickly start hinting at massive revolutions in nano-machines that constantly respond and improve all aspects of human life and engineering, aggregating together to form macro-level machines of incredible complexity and computational capacity. In contrast, mechanical interpretations of nanotechnology are grounded in limited types of materials in limited environments with limited capabilities.

For the most part, these limitations are the same for both macro-scale mechanical devices and nano-machines since both can be described in "familiar, mechanical terms." However important differences do exist. Drexler notes that "a machine can't work well if its parts can't move smoothly and interactions between atomic-scale bumps on surfaces might seem to make smooth movement impossible." This can be resolved through designing bump patterns to produce superlubricity but comes with its own constraints. Drexler also notes the presence of drag and thermal motion but these forces are not very significant in the context of rigid mechanical structures. I would also like to add an additional constraint: Because "stretching space and time in equal proportion scales properties like mass, force, and velocity in exactly the right way to make mechanical motion the same," one quickly realizes that gravity, producing an acceleration of 9.8 meters/second down regardless of size, vanishes in importance. In this sense, nano-scale factories will not be like the factories seen on Earth but rather more like the ones designed in space. Additionally, they will be driven purely by motors and mechanical devices with no incorporation of electrical wiring (classical electrical engineering breaks down at the nano-scale).

Initially, these limitations made me skeptical about whether APM could feasibly create complex designer materials or extend to advanced nano-machines outside of a manufacturing context. However, this video led me to realize that the sort of target materials to be constructed could rely on basic patterns of picking up and putting down different atoms and molecules. Additionally, because the nano-scale is so small, relatively large nano-machines can still be build at very low size-scales (though this nano-machines are still what I am most skeptical about).

For the most part though, the outputs of my biologically-inspired expectations of nanotechnology and Drexler's mechanical vision seem similar. As Drexler says:

A SCIENTIST WROTE an article about the nanomachines of the general sot I've described, but he suggested that they couldn't be used in a biological environment because biomolecules would gum up gears and other moving parts. The answer, of course, is to keep gears in a gearbox, and to place all critical moving parts inside a sealed shell.

The moral here is that most mechanical nano-machines can be designed to avoid many of the problems that they might face relative to their magical biologically-inspired cousins. At the same time, this indicates a more general problem: to build a functional nano-machine, there must be some input and some output which facilitates action on the environment. This means that some of the mechanisms in a given nano-machine must be designed for the environment. However, the severity of this limitation also depends on the nano-machine's functionality: For power, mechanical mechanisms can be designed to minimize complex parts exterior to a core gearbox. For expelling objects, a nano-machine might incorporate a simple airlock-like design. However, for taking in and processing new molecules in an uncontrolled environment, nano-machines would require both filters and safe-guards to prevent clogging.

C. APM as Macro-Scale Manufacturing

While these limitations do not preclude the existence of nano-machines (and especially those that manufacture on the fly) in uncontrolled environments, they do raise questions. However, this specific technology is not the end goal that Drexler focuses on in Radical Abundance. Though Drexler does allude to "fast, thorough data collection and the means for rapid deployment of nano-scale devices" in the context of biological interventions, the lion-share of speculation about the future is focused on machines like this one:

Picture yourself standing outside the final assembly chamber of a large-product APM system and looking in through a window to view the machines at work in a space the size of a one-car garage....

To the right, you see an exit door for products ready for delivery. To the left, you see what look like wall-to-wall, floor-to-ceiling shelves, with each shelf partitioned to make a row of box-shaped chambers. In the middle of the garage-sized chamber in front of you is a movable lift surrounded by a set of machines.

The machines look uncommonly sleek, yet very familiar. They resemble machines in an automated factory, with robotic arms programmed to swing around, pick up components, and swing back to snap the components together. The machines look like this because they are, in fact, machines in an automated factory and because machines that perform similar motions often have similar shapes and similar moving parts. Because they are made of materials better than steel, however, they can be faster, lighter, and more efficient.

Looking back at the wall on the left, you can get a clear view into several chambers that happen to be at eye level and near the window. Each smaller chamber contains machines with swinging arms, and the overall setup inside looks like a scale model of the larger chamber, complete with a rear wall with wall-to-wall, top-to-bottom rows of yet smaller chambers. It’s hard to see in detail what these small chambers-within-chambers contain, but they seem to hold a tiny yet familiar set of machines mounted in front of a rear wall with rows of yet smaller chambers.

With the press of a button, the machinery kicks into gear. At first nothing seems to happen, but in less than a minute the large machines in front of you start to pick up parts as they pop out of the chambers in the wall at the left, moving these parts to the platform in the center where the first parts are clamped, and the rest snap together. As the machines put the parts together, a familiar product takes shape, an automobile, different in almost every detail from those built today, yet having a form that reveals the same function. chambers.

With the press of a button, the machinery kicks into gear.

At first nothing seems to happen, but in less than a minute the large machines in front of you start to pick up parts as they pop out of the chambers in the wall at the left, moving these parts to the platform in the center where the first parts are clamped, and the rest snap together. As the machines put the parts together, a familiar product takes shape, an automobile, different in almost every detail from those built today, yet having a form that reveals the same function.

Each part takes several seconds to put into place and new parts slide out of the chambers at a corresponding rate, each chamber delivering a component every few seconds. To the left, inside the closest chamber, you can see the machines working inside. These miniature machines seem to be performing similar tasks, but at a rate of several cycles per second, their motions are almost too quick to follow. It’s easy to guess what’s happening in the yet-smaller chambers farther back, yet the motions there are no more than a blur.

In the main chamber the work is complete in less than a minute. The door to the right then unseals and opens, and a car moves out into a receiving area, sealed in what looks like a plastic sleeve. A moment after the door reseals, the sleeve is pulled back for recycling and the process is done. (This exit maneuver is part of a cycle that prevents contaminants from entering when the product exits.

This description is in line with the video linked earlier and leads me to infer that, though many other uses also exist for nanotechnology, the main use-case emphasized in Radical Abundance is extremely efficient factories that build extremely high quality materials by exploiting "lightweight, carbon-based materials" and interesting already-discovered patterns that may produce exotic electronic properties on the macro-scale. I mention this because APM as advanced manufacturing and APM nano-machine technology itself reflect different risk profiles that ought to be discussed in the impact section. As Drexler says, "where the physical nature of APM technologies is concerned, the relevant questions pertain to the physics and engineering of compact, highly capable factories-- not vague dreams, exotic products or nanobugs."

III. APM Will Change Everything

A. Drexler's Partial Notes on APM Benefits

Because APM will use cheap, readily available materials like carbon to produce high quality materials in minutes from a compact machine, it has the ingredients to revolutionize the existing means of production:

"Nanoscale size enables extreme productivity as a consequence of mechanical scaling laws. In addition, small-scale, versatile, highly productive machinery can collapse globe-spanning industrial supply chains to just a few links."

This massive simplification of supply chains coupled with APM's use of abundant molecules implies a massive economic shift. Locations impoverished by lack of capital, business connections and resources could, in principle, use nanotechnology to sustainably and locally produce high quality infrastructure and reliable food sources.

Other benefits of APM include transforming information technologies through more advanced materials, improving infrastructure through better construction and transportation technology, improving agriculture through efficient recycling and water-processing/desalination, and resolving global warming carbon dioxide capture. The running theme themes in each of these solutions are APM’s blend of efficiency and ability to cheaply construct many of the advanced nanomaterial solutions which currently exist but are too costly to implement industrially.

The benefits of APM in improving agriculture also struck me as particularly significant. In particular, Drexler notes that enclosed agriculture (i.e. large-scale greenhouses) offers "higher yield per hectare, better food quality, freedom from pesticides, extended growing seasons, freedom from constraints of soil quality and available water, and protection from drought" while also reducing "water demand and contamination." These latter benefits—freedom from soil, water and weather constraints—are already achieved achieved in modern greehouses and act as the corner stone for alleviating starvation in impoverished regions. Nevertheless, the modern world, let alone impoverished regions, cannot meet the efficiency and infrastructural requirements that would make implementing greenhouses economical. APM’s capacity for rapid and cheap infrastructure along with highly efficient energy systems (i.e. solar power) will change that.

Overall, I see only two reasons to doubt these benefits. The first reason is economic. Historically, international trade has been a necessity for technologically advanced countries which has allowed the flow of money to less developed countries. However, APM will finally allow technologically advanced countries to become self-sufficient to the extent that they do not need to import goods from less advanced countries. If these countries cannot amass the wealth to gain nanotechnology, the absence of economic ties with technologically advanced autarkies may lock them into poverty. However, the presence of these self-sufficient countries also makes this situation unlikely. Because APM-driven countries will be so wealthy and likely retain some altruistic leaning, the only reason for them not to provide APM to less advanced countries would be some sort of non-economic cost or risk—that is, a military risk. Fortunately, because APM factories will likely be both technologically complex and optimized for use in fabricating specific materials rather than every material, the military risk of providing factories designed with food and infrastructure in mind is relatively low. This is especially true when noting how military APM will empower the most advanced countries relative to others. To be fair though—in the era of semi-costly/only partially effective nanotechnology—economic concerns could be very serious even if they are only short-term considerations.

The second reason for doubt is more significant and relates to an important criticism of Radical Abundance: Drexler effectively discusses the capabilities of nanotechnology but does not seriously discuss the details of the problems he claims that it will solve. In practical terms, the most important altruistic contribution of nanotechnology would be poverty reduction. Explicitly, nanotechnology should provide sustainable food sources in places where food cannot currently grow, water in places where water is currently inaccessible, infrastructure in places where infrastructure cannot currently be established, and energy in places with limited energy resources. These problems are almost all geographical constraints and, while Drexler claims that "the most useful elements--including carbon nitrogen, oxygen, and silicon--are not all that scarce," nitrogen deficiency is one of the reasons for agricultural difficulties in Africa. Moreover, many impoverished countries suffer from lack of water/reliance on contaminated ground water and, while nanotechnology may provide cheap purification methods, I am not convinced that it will either offer purification methods or water transportation methods better than the ongoing science in those fields. Finally, many impoverished countries suffer from rocky, mountainous and swampy terrain which inhibits movement and the feasibility of taking advantage of large-scale infrastructure. Enclosed agriculture on forty-five degree rocky inclines does not seem very promising, especially when compared to just using longer supply chains which APM would hopefully avoid. In short, while nanotechnology might allow the collapsing of globe-spanning supply chains to a few efficient links, Drexler does not discuss how the locations that are most harmed by the need for these supply structures would surmount their existing geographical constraints using nanotechnology. If this is not addressed, it creates a massive problem for APM. After all, if geographical issues limit the efficacy of altruism through nanotechnology, then—for impoverished countries—access to nanotechnology may not outweigh the economic loss of being unable to trade with wealthy countries.

Beyond the potential for providing the basic staples of human life, an adjacent significant contribution that highly advanced nanotechnology brings is control of the ecosystems that make these staples. As someone concerned about wild animal suffering, this development could massively reduce human harm to insects and potentially offer strategies for systemic wildlife interventions. While restraint and hesitation should generally be applied to actions like this as they currently have dramatic and unpredictable impacts on wild ecosystems, the extensive surveillance technologies provided by APM (discussed later as a drawback) along with advancing science may provide the exact kind of ecological knowledge needed to make these adjustments wisely.

While I have previously suspected that limitations on chemicals and manufacturing methods may render APM less useful than it seems, I think that these views fail to appreciate simply how much can be accomplished through the cheap production of high quality infrastructure and electronics alone. With this in mind, I think that the only scenarios where APM is less-than-revolutionary are scenarios where some other easier-to-develop technology focused on particularly important use-cases succeeds faster by virtue of a more direct path to implementation. Off the top of my head, these might look like some of the following:

This is my greatest and most explicit uncertainty with respect to APM's benefits. I agree that APM would provide these benefits but I am not sure whether APM will be the first to provide them. Heuristically, I usually expect that technology specifically designed to address a specific issue will advance much faster than a more generic abstract approach. This leads me to doubt the comparative benefits of pursuing APM as opposed to other more explicitly oriented research endeavors. On the other hand, APM is a relatively concrete idea and, even if it is likely more difficult to pursue than a more focused technology, may be worth it due to its large potential range of impact.

B. Drexler's Notes On How APM May Go Wrong

One of the dangerous aspects of APM that Drexler notes is surveillance.

Small-scale components and systems open new possibilities. For perspective, consider that a one-gram platform built with advanced technologies could produce teraflops of computational power (and much more, in bursts), together with a million-terabyte data storage capacity and better-than-human sensors, all with a power demand comparable to that of a cell phone on standby. Now consider what could be done if devices of this class cost about $1.00 per kilogram and could be delivered by small drone aircraft. $100 billion would buy roughly one device of this sort per square meter of land area, worldwide.

Because most of this technology will not need teraFLOPs of computing power, this raises the possibility of globally abundant third-party surveillance as easy to bring to work as it is to track dirt into a house. Coupled with the ability to relay information, these sensing devices would be able to trivially transport massive amounts of data to a centralized system for fast (>petaFLOP) processing. This could quickly lead to a race-to-the-bottom favoring those who are willing to care less. Combining these same devices with chemical payloads may also enable groups to engage in the soft psychological manipulation or undetectable assassination of their enemies. Combining these "secretive surveillance regimes" with a world that replaces wealth with individual virtue as a status symbol in absence of poverty may also lead to the micro-regulation of human behavior. Additionally, because surveillance does confer advantages onto the groups that use it and many countries vary significantly in their attitudes toward individual privacy, establishing a common pro-privacy agreement to avoid a race-to-the-bottom will be unlikely.

Overall, I am not concerned about this risk mainly because this surveillance technology is oriented around active nano-machines in complex and unpredictable environmental situations. Unlike the benefits of nanotechnology which are produced in controlled manufacturing contexts, this technology must function in a less manageable space and I expect physical and design limitations to seriously come into play here. Undoubtedly, nanotechnology will cheapen surveillance but I doubt that it will be sufficiently cheap to subvert personal privacy more than it would be anyway. Furthermore, while a race-to-the-bottom may happen, it may be partially mitigated by the radically different economic circumstances which more simple nanotechnology will have already provided the world, making it "hard to find an external resource that would continue to be a vital national interest."

These are also the same reasons that I am not seriously concerned with nanotechnology causing massive environmental damage through self-replicating robots: the problem is uniquely challenging and strong incentives for doing so appear to be absent. In fact, Drexler notes that nations have "strong, shared interests in constraining non-state actors" from the use of unrestrained APM. A weaker variant of this issue is the potential for nanotechnology to create nanoparticle pollutants on a massive scale, leading to massive ecosystemic damage. However, countries capable of producing nanotechnology at this level will also likely have technology capable of rapidly identifying (via surveillance) and addressing pollutants anywhere on the globe. This means that mass pollution is only a risk if done by a country that would not fear retaliation. Because this type of country can already threaten the world in many more ways than just pollution, I consider this to be a minor concern.

The second main risk that Drexler discusses is a two-pronged change in military strategy. First, because APM will be used to design sequential generations of APM, "it is therefore easy to envision scenarios in which a modest degree of asynchrony, measured in months or less, could swiftly lead to radical military asymmetries even with relatively shallow exploitation of the overall potential of APM-level technologies." Second, APM would provide "abundant, affordable, non-lethal, remotely operated weapons" that reduce risk of harm and encourage more war-like policies. While these shifts may encourage technologically advanced countries to pursue aggressive expansionist policies, expansionist actions will offer little in terms of resource gain and cost a level of societal stability due to cultural conflicts. In fact, because of the absence of resource gain, expansionism will primarily be driven by a desire to establish different societal values--a goal that governments will more likely pursue through subtler methods than outright conflict.

Lastly, Drexler also notes that one of the commonly discussed risks--the idea that self-reproducing nano-machines will consume vast swathes of the world--is unlikely given a manufacturing model of nanotechnology instead of a biological model. I agree that this is unlikely for the same reason that I saw surveillance as unlikely but multiplied several-fold. Building environmentally robust reproducing nano-machines using abundant natural supplies and establishing enough stable complexity for those nano-machines to reconstruct themselves both require an enormous amount of effort far beyond the scope of the primary benefits of nanotechnology which Drexler discusses.

One danger that I believe merits concern but goes unmentioned by Drexler pertains to advanced general intelligence. If Drexler is correct in his claim that nanotechnology will offer teraFLOP/gram computers, then an average laptop computer weighing five pounds would have about a petaFLOP of computational power and a thousand of such computers would start pushing on the capabilities of a human brain (according to Kurzweil) without any additional deliberate optimization beyond emulating nature. This is more than sufficient to produce intelligence far more competent than any human and intertwines nanotechnology deeply with its relationship to artificial intelligence. If humanity is not confident in its ability to align the goals of advanced general intelligence (which it is not), then nanotechnology in this regard ought only be pursued if does not increase the risk of misaligned advanced general intelligence. This will be the case if either advanced general intelligence is expected to emerge prior to advanced nanotechnology (in which case AI Safety research is a much more pressing concern than nanotechnology) or Drexler is incorrect about teraFLOP/gram computers.

Overall, my opinion is that Drexler's benefits of nanotechnology outweigh Drexler's risks but, when I consider nanotechnology's relationship with artificial intelligence, I begin to suspect that it is far more dangerous than it initially appears. That being said, this only applies directly to the development of nanotechnology itself. Development of nanotechnology policy on the other hand may have significant value in the case where humanity weathers advanced general intelligence. However, the benefits of this work may be limited by the likelihood that advanced general intelligence will transform the political landscape to such an extent that any current policy design would be inapplicable.

IV. Pathways to Nanotechnology

When wondering about the likelihood of achieving something like APM, a useful question is "Why has this not been achieved already?" This is an especially salient question for APM both because of the extensive funding providing in the field of nanotechnology and the relative age of the idea itself. Drexler answers this question with two main explanations. The first is that extensive funding may have been focused on "nanotechnology" but minimal funding has been focused on APM.

"The great promise of nanotechnology is atomically precise manufacturing, and the US Congress established a program directed toward this objective, but the program instead did something entirely different. The program's leaders redefined "nanotechnology and supported only nanoscale materials and devices, technologies as different from APM as cloth, cement, and wires are from a programmeable digital computer. Most research advertised as "nanotechnology" has therefore been irrelevant to what had been widely expected, and while atomically precise fabrication has flourished in the molecular sciences, people looking for progress toward APM-level technologies have been led to look in the wrong direction."

The second is a rather philosophical discussion about how the basic science that mostly comprises nanotechnology research is ineffective at producing the system-engineering coordination required to make an actual technology. As Drexler says, "no matter how research-intensive a project may be, work coordinated around concrete engineering objectives will eventually be required to produce concrete engineering results." So far, work at this level of coordination has not been pursued.

Both of these explanations demonstrate that the existing limited state of APM is due to political and organizational constraints that do not reflect negatively on the feasibility of APM itself. As for that feasibility, Drexler notes that designing the mechanisms of APM devices is already relatively feasible: "Machine components based on rigid covalent structures cannot yet be implemented, yet are already easy to design and model using standard computational chemistry software." Indeed, I suspect that one way to motivate APM-directed research would be to use the tools we have to preemptively design an advanced APM machine and then to use that device as a motivator for nations to funnel research into its ultimate construction.

The main current method for developing APM will involve advancing stereotactic chemical reactions. While "conventional, solution-phase chemical reactions are enabled by local structural features of molecules," more complex synthetic targets with many more structural features would be "difficult or impossible to direct reactions with sufficient specificity." Stereotactic chemical reactions address this issue by having "linking structures direct reactions by constraining encounters among potentially reactive groups, separating some pairs while increasing encounter rates between others." By definition, stereotactic chemical reactions reflect the atomically precise manufacturing of complex structures through sequentially bonding molecules together. This description applies to biological proteins as well but unlike proteins, APM will develop stereotactic reactions through mechanical designs rather than thermally driven ones.

Drexler notes that stereotactic reactions in advanced APM must satisfy a number of constraints: structures must be stable and rigid; structure motion must be well controlled; reactions must happen reliably; reactions must be irreversible; and reactions must yield only a single product. Many of these requirements (motion control, reliable processes, irreversibility, single product) are consistent with most fabrication procedures. However, rigidity holds special importance for APM because it allows fabricated materials to occupy meta-stable states. This means that, while flexible biological materials can quickly restructure themselves into lower energy/higher entropy states by just moving, APM materials move in a slow manner due to rigidity that prevents them from finding those states as quickly. This ensures that while APM materials, like diamond, maintain their properties for a long time even though they might not be naturally stable.

Current technology cannot meet all these requirements however the assembly of "polymeric building blocks, cross-linked via conventional reactions, with all stereotactic operations performed in aqueous environments" may readily be achieved through existing methods like DNA origami. This offers a bottom-up APM starting point wherein "large self-aligning building blocks, loose positional tolerance margins, low stiffness materials, simple machines, and simple motion constraints" are iterated upon to produce the more complex, more precise, and higher quality nano-devices needed to reach advanced APM. In other words, simply iterating on existing nanotechnology related to stereotactic control will facilitate the improvement of stereotactic control.

Drexler also addresses another aspect of nanotechnology that had previously led me to be more suspicious about its success: On the macro-scale level, a factory may rely on machines produced by other factories but those factories themselves are assembled by humans (though often humans piloting machines themselves). On the nano-scale though, human assembly is impossible so factories would need to produce factories--which seems hard. In actuality, the solution is simple and practical: Stereotactic control will lead to "improvements that facilitate the design and fabrication of complementary surfaces... Stereotactic synthesis can enable advances in component level self-assembly." Once humanity can produce nano-machines, humanity will be able to produce nano-machines attached to surfaces with binding affinities for complementary surfaces. At this point, near current-day technology would be able to produce large-scale complementary surface patterned sheets that bind only the right nano-machines in only the right places. In other words, success will not initially be achieved through a closed system of nano-factories making nano-factories but rather through a blend of manufacturing nano-machines to more easily interact with higher level self-assembly methods.

V. Conclusion

Throughout Radical Abundance, Drexler emphasizes two common misrepresentations of nanotechnology: the popular representation of nanotechnology as almost magical nanobot swarms and the academic representation of nanotechnology as pertaining to science at a given size-scale rather than technology at a given size-scale. As someone both aware of popular culture and closely involved with conventional academic research in nanotechnology (i.e. self-assembling nano-particles), I blended these ideas together into an improbably influential technology mediated by an improbably hard-to-control science. In reality, APM is a comprehensibly impactful technology mediated by under-researched but high-potential iterative design. Finishing the book, I feel reasonably convinced that APM is achievable based on existing technological progress, partially convinced that APM will have the benefits that Drexler claims it has, and unconvinced that the benefits outweigh the risks if we limit them to those that Drexler has described.

Nevertheless, while understandable in the context of nanotechnology's previous over-hyped history in pop-culture, I think Radical Abundance's discussion of risks is too conservative in estimating their magnitude. This is because, in a significant number of circumstances where human-developed nanotechnology becomes important, I envision scenarios that accelerate the emergence of a misaligned general artificial intelligence. These scenarios may be absolutely low probability but should be accounted for in nanotechnology policy given the existential nature of the threat. Beyond this, I think that Drexler's goal of addressing misconceptions about out-of-control nano-machines may also fail to recognize risks from out-of-control nano-machines which are outside the purview of comparatively limited APM. I have no reasoning for this beyond the sense that any theoretically possible existential risk ought to merit closer consideration and that many people have given closer consideration to it.

So in summary, APM is a lot more likely than I initially expected and I also suspect it could be a lot more dangerous as well. Let me know your thoughts!


Comments sorted by top scores.

comment by PeterMcCluskey · 2018-10-15T16:46:52.293Z · score: 5 (4 votes) · LW(p) · GW(p)

nitrogen deficiency is one of the reasons for agricultural difficulties in Africa.

There's no shortage of nitrogen in the air. Nitrogen-fixing bacteria exist. Getting them to do the right job does not seem particularly hard compared to the rest of what Drexler discusses.

These problems are almost all geographical constraints

With sufficiently advanced technology, that claim seems obviously wrong.

comment by Isnasene · 2018-10-16T20:50:23.585Z · score: 1 (1 votes) · LW(p) · GW(p)

I recently had a conversation with Christine Peterson and she pointed out the same thing. Imagine! A materials engineer who forgot that there was nitrogen in the air!

After some thinking, I also believe that nitrogen deficiency in Africa is not the concern that I indicated it was in the paper. I also agree that with sufficiently advanced technology (even when that technology is limited by physical laws), one should be able to overcome any geographical constraint in principle. This is Lesswrong and, thus, I must say that I was wrong about this.

Given that we can construct APM factories anywhere--something we likely can do once APM tech is achieved (though I suspect international interests would want to keep the APM tech used to create APM factories relatively inaccessible)--and that they can be maintained (again, something that APM can also ensure happens), I don't imagine geography being an issue. Thus, I don't really expect geography to be an issue.

I'm now much more confident that APM will lead to dramatic economic improvements in the places where we most care about them. There are still some practical considerations (i.e. ensuring the accessibility of people/machines capable of trouble-shooting the factories are everywhere the factories are) but these considerations are readily achievable.

comment by ryan_b · 2018-10-16T13:49:12.963Z · score: 2 (1 votes) · LW(p) · GW(p)
Fortunately, because APM factories will likely be both technologically complex and optimized for use in fabricating specific materials rather than every material, the military risk of providing factories designed with food and infrastructure in mind is relatively low.

While I agree with this as written, I expect the difference is smaller than one would expect.

  • Given the choice between an opponent with APM armaments and a standard infrastructure, or an opponent with standard armaments and an APM infrastructure, the latter is a greater military threat.
  • It is trivially easy to militarize any given civilian tech: most military vehicles are armed and up-armored civilian ones anyway; armed drones initially were literally and exactly unarmed drones with missiles bolted to them; the opponents history's foremost military power have been fighting these last 17 years have been doing it with old gear, gear provided to them by outsiders, and armed civilian gear.
  • While APM-hybrid equipment may be inferior to APM-optimized equipment, it will probably still be superior to standard equipment.
  • In the current security environment it is common to provide more advanced armaments to less advanced belligerents, and the nature of APM seems like it would make reverse engineering more difficult which suggests to me providing such armaments is an even safer bet than usual.

The role of logistics (which is to say, infrastructure) in modern military affairs is widely underappreciated, and I have no reason to suspect Drexler of any particular expertise in this area. In sum, it looks to me like the security impacts have been underrated in their complexity. Although I note this effectively pushes more strongly in the direction of your sense of being uncertain if the risks are worth the gains.

comment by Isnasene · 2018-10-16T22:04:39.562Z · score: 3 (2 votes) · LW(p) · GW(p)

Given the choice between an opponent with APM armaments and a standard infrastructure, or an opponent with standard armaments and an APM infrastructure, the latter is a greater military threat.

I'm not knowledgeable enough about modern military affairs to be certain of anything but I currently agree with this overall--though it depends on how advanced the APM tech is. The main advantages that APM provides generally fall into the categories of increasing the resources we can use and increasing the precision of our equipment. For the most part, I have difficult envisioning increasingly high quality/precise versions of any APM-based conventional weapon making a bigger war-time difference than APM's potential ability to make large numbers of lower quality weapons while providing cheap vehicles for troops.

Some caveats exist to this though in unconventional weapons that do rely highly on precise design. In particular, the ability to make efficient, cheaply made autonomous drones and transport vehicles for them is a massive logistical advantage. Between reducing the risk of human life and allowing local recuperation of energy***, these sorts of weapons should quickly overwhelm weaponized civilian weapons without those advantages. That being said, if APM factories for certain civilian uses like power generation are included in APM infrastructure, I can envision a less tech-advanced opponent putting up a difficult enough fight to make war unwinnable. This is one of the points against military risk being low.

Still, APM infrastructure and APM weapons aren't an either/or situation. More frequently, the strong nations will have both. This is one of the points in favor of military risk being low.

While APM-hybrid equipment may be inferior to APM-optimized equipment, it will probably still be superior to standard equipment.

I agree with this but suspect that the weaponry possibilities that APM opens up reach far beyond the purview of upgraded civilian tech to the extent that these improvements may not matter so much. This point also applies to the question of providing APM tech to belligerents--it depends on what the tech is.

The role of logistics (which is to say, infrastructure) in modern military affairs is widely underappreciated, and I have no reason to suspect Drexler of any particular expertise in this area.

Overall, I think the crux of the impact of APM lies in how accessible APM tech will be in the future. Because geographical constraints likely won't be a limiting factor (see PeterMcCluskey's response), this comes down to how good APM nations feel about giving APM to others. This means that the military risks of APM is likely one of the key factors in determining its effectiveness. In Drexler's defense, it's hard to predict the details of a massive change like APM and logistics often come down to small details that get very confused in high variance situations.

***Unlike nuclear weapons, abandoned or broken APM weapons have a relatively high risk of being salvaged for parts. Even if it provides a logistic advantage, I'm not sure that an APM nation would be willing to send very efficient motors/engines/energy harvesters into foreign territory for risk that the opponent takes those parts and hooks them into a weapon since those simple sorts of improvements can be a nightmare to combat. This argument also applies to providing other countries with APM energy sources and motor fabrication plants too.

comment by avturchin · 2018-10-15T17:03:01.433Z · score: 1 (3 votes) · LW(p) · GW(p)