ciphergoth comments on David Matthewman on the Whole Brain Emulation roadmap on David Matthewman on the Whole Brain Emulation roadmap by Doug 2010-02-23T16:24:35+01:00<p>No, feel free to bump it up if helpful &mdash; and happy for you to do it i.e. I haven&#8217;t time to&nbsp;:-) </p> <p>Not really on topic for my work blog and my LiveJournal isn&#8217;t (supposed to be) search engine-able so better for your purposes to put it here and I&#8217;m happy to have it done that&nbsp;way.</p> <p>For the avoidance of any doubt, have a <span class="caps">CC0</span>/public domain release: To the extent possible under law, I hereby waive all copyright and related or neighboring rights to the blog comment I made above, which I made from the United&nbsp;Kingdom.</p> Comment on David Matthewman on the Whole Brain Emulation roadmap by Paul Crowley 2010-02-23T14:43:00+01:00<p>Wow, thank you very much. Do you mind if I turn this in to a <a rel="nofollow" href="">post in its own right</a>? I believe I can simply mark your account as having posting privileges so that it can directly be in your name. Or of course if you prefer to post this to your own blog that would also be&nbsp;wonderful!</p> Comment on David Matthewman on the Whole Brain Emulation roadmap by Doug 2010-02-23T14:24:02+01:00<p>I too am short of time, but have given this paper a quick run through. Here are some unstructured and unedited quick notes I made while I was at it. Apologies for brevity and errors &mdash; I almost certainly missed some of their points and have misrepresented parts of their&nbsp;case.</p> <p>It does seem to be a serious and reasonably well-informed piece of work on speculative science and technology. Emphasis on the speculative, though &mdash; which they&nbsp;acknowledge.</p> <p>The distinction between emulating a brain generically (which I reckon is probably feasible, eventually) and emulating a specific person&#8217;s brain (which I reckon is a lot harder), and emulating a specific dead person&#8217;s brain (which I reckon is probably not possible), is a crucial one. They do make this point and spell it out in Table 1 on p11, and rightly say it’s very&nbsp;hard.</p> <p>p8 &#8220;An important hypothesis for <span class="caps">WBE</span> is that in order to emulate the brain we do not need to understand the whole system, but rather we just need a database containing all necessary low‐level information about the brain and knowledge of the local update rules that change brain states from moment to&nbsp;moment.&#8221;</p> <p>I agree entirely. Without this the ambitious bit of the enterprise fails. (They make the case, correctly, that progress down these lines is useful even if it turns out the big project can’t be done.) I suspect that this hypothesis may be true, but we certainly need to know a lot more about how the whole system works in order to work out what the necessary low-level information and update rules are. And in fact we’ll make interesting scientific progress – as suggested here – by running emulations of bits of the brain we think we might understand and seeing if that produces emergent properties that look like what the brain does. Actually they say this on p15 &#8220;<span class="caps">WBE</span> appears to be a way of testing many of these assumptions experimentally” – I’d be a bit stronger than&nbsp;that.</p> <p>Table 2 on levels of emulation makes sense. My gut instinct (note evidence base) is that we will need at least level 8 (states of protein complexes – i.e. what shape conformations the (important) proteins are in) to do <span class="caps">WBE</span>, and quite possibly higher ones (though I doubt the quantum level, 11, is needed but Roger Penrose would disagree). Proteins are the actually-existing nanobots that make our cells work. The 3D shape of proteins is critical to their role. Many proteins change shape – and hence what they do or don’t do – in to a smallish fixed number of conformations, and we already know that this can be hugely important to brain function at the gross level. (E.g. transmissible spongiform encaphalopathies – mad cow and all that – are essentially caused by prion proteins in the brain switching from the ordinary shape to the disease-causing&nbsp;one.)</p> <p>The whole approach is based on scanning an existing brain, in sufficient detail that you can then implement an emulation. I think that’s possibly useful, but I think a more likely successful route to a simulated (!) intelligence will be to grow it, rather than to bring it in to existence fully-formed. By growing, I mean some process akin to the developmental process by which humans come to consciousness: an interaction between an environment and a substrate that can develop in the light of feedback from that environment. But based on their approach, their analysis of technological capabilities needed seems&nbsp;plausible.</p> <p>The one that leaps out as really, really hard (to the point of impossibility in my mind) is the scanning component. There is the unknown of whether the thing is doable at all (what they call scale separation), which is a biggy, but falsifiable by trying out experiments in this&nbsp;direction.</p> <p>They talk about electron microscopy as being the only technology which offers sufficient resolution. They say that the neuronal/synaptic level would only require trivial increases in microscopy resolution. That’s missing the point. You just can’t scan enough of a brain with the sorts of microscopy that work at that&nbsp;resolution. </p> <p>(This is leaving aside the entirely non-trivial question David addressed of whether it&#8217;s possible or not to extrapolate from existing technologies to ones that would have sufficient resolution <em>for tiny prepared samples</em>.)</p> <p>Almost all forms of microscopy that could conceivably come close to being useful here give you an image of an exposed surface. You’re going to need to chop the brain up in to fragments that are thin enough to expose every single synapse, at a minimum. That’s not feasible without destroying at least half of what you’re trying to analyse. When you slice something, you basically smash up a thin column of stuff in the path of the knife (or laser beam, or whatever). And even if you invented some magical way of preparing the samples without that mechanical damage, you’d still have to pull the network of synapses apart in order to expose the surfaces to microscopy&nbsp;.</p> <p>Sure, you can do it for very small, thin organisms (they mention C. elegans) – because they’re not much more than a couple of neurons thick. But human brains are a lot thicker than that. And anyway, C. elegans was (I strongly suspect) done by scanning loads of individuals and aggregating the date. That’s automatically shut you out of the big-ticket replicating-a-person&nbsp;bit.</p> <p>Oh, and it’s worth noting that all of this, to be remotely possible, is really quite spectacularly destructive. Your brain is not going to be doing any thinking once this process is done with. Which is another reason why I think growing a simulated/emulated brain is a better research&nbsp;plan.</p> <p>There are some imaging techniques which don’t require the slice-and-dice bit: <span class="caps">MRI</span> is a better possibility, at least superficially. And this is probably something I could dig in to more later, since I do know quite a lot about the physics/chemistry behind the technique. (Part of my PhD was developing a teaching simulation of an <span class="caps">NMR</span> spectrometer, which is a simpler thing than an <span class="caps">MRI</span> machine.) But off the top of my head it really doesn&#8217;t seem likely for all sorts of reasons &mdash; resolution, in multiple senses (you&#8217;re going to need a radio sensor with finer resolution than is theoretically possible). If you want, nudge me and I&#8217;ll try to find time to spell this out more and may even dredge up enough of the maths to do sums on&nbsp;it.</p> <p>Their ideas on nanodisassembly seem like nonsense. You can’t build Drexler-type nanobots: the physics/chemistry just doesn’t work like that at that scale. Think proteins, not a teeny version of Robosapien. They say (p51) “Given that no detailed proposal for a nanodisassembler has been <br /> made it is hard to evaluate the chances of nanodisassembly”. I don’t think it’s hard to evaluate: the chances are negligible to&nbsp;nil.</p> <p>Chemical analysis – this is really not going to happen. I’d just been thinking about neuronal connections. Blimey, they are really stretching the bounds of what’s even theoretically possible here. There’s several techniques they mention that I don’t know about, but I do know about some and <span class="caps">SFAICT</span> they all suffer from the need-to-break-the-brain-up problem only worse. They mention dyes – but dyes are generally pretty large in chemical terms and will almost certainly destroy information if you perfuse a brain with&nbsp;them. </p> <p>I’m not paying any attention to the information processing stuff. I could do, and the challenge is large, but (a) my seat-of-the-pants feeling is that Moore’s Law can do more than enough here, and (b) lots of people – ciphergoth included! – are at least as capable as me of doing the detailed scrutiny&nbsp;here.</p> <p>Likewise the image processing and scan interpretation bit, and the neural simulation component, and so on – that doesn’t strike me as the really theoretically hard part. It might turn out to be practically impossible, of course, but it doesn’t ping my bogosity meter the way the scanning part&nbsp;does.</p> <p>Ah, actually, one thing on the implementation side that might scale up to be infeasible is if you need to get serious about shape variability of proteins. Calculating possible conformations of proteins is (currently) a classic application that requires Grid computing type resources: it’s at the edge of what we can do with current computing resources. And that’s simply twisting a single smallish protein around. If we have to do that for a large proportion of the proteins in a brain, it’s easily above what’s going to be computationally feasible this side of the singularity. But I suspect we’ll be able to get by with a lot less detail, though not nothing &mdash; e.g. for <span class="caps">TSE</span> prion proteins, we might not have to do the whole calculation of possible conformations for each protein, and how that’s affected by its neighbours: we might well be able to just model it as a location in space and a one-bit state variable indicating whether it’s in normal or <span class="caps">TSE</span> conformation. My guess is that you’d need a bit more than that but not a lot. But that really is a guess that wants empirical&nbsp;testing.</p> Comment on David Matthewman on the Whole Brain Emulation roadmap by Paul Crowley 2010-02-22T08:51:10+01:00<blockquote> <p>Just to be clear on the scope of this: no matter how many times you ask me for ‘anti-cryonics’ writing, you’re not going to get it from me, because I’m not&nbsp;‘anti-cryonics’.</p> </blockquote> <p>Understood, and I don&#8217;t mean to mis-characterize you. Is there a better word for what I&#8217;m looking for? Roughly, it&#8217;s authors sounding a note of caution and who aren&#8217;t already in the cryonics&nbsp;&#8220;camp&#8221;&#8230;</p> Comment on David Matthewman on the Whole Brain Emulation roadmap by David Matthewman 2010-02-22T06:53:41+01:00<p>Just to be clear on the scope of this: no matter how many times you ask me for &#8216;anti-cryonics&#8217; writing, you&#8217;re not going to get it from me, because I&#8217;m not&nbsp;&#8216;anti-cryonics&#8217;.</p> <p>This rely was a specific response to your assertion that: &#8220;&#8230;it&#8217;s plausible to imagine that it might be as simple as serially slicing the brain open and scanning the surface with an <span class="caps">SEM</span> and various technologies to find out about the chemistry at the exposed surface, then doing a whole-brain-emulation on the result.&#8221; You then linked to this paper as evidence that it might be that simple. I think that paper is strong evidence that it won&#8217;t be that simple, and those are my&nbsp;reasons. </p> <p>Criticising me for (<span class="caps">AFAICT</span>) correctly reading the paper, especially as you continue to cite the paper as a valid &#8216;roadmap&#8217; seems odd. As I&#8217;ve said elsewhere, if it is a roadmap at all, it is a roadmap of our current position in the swamp, with big &#8216;X&#8217;s marking all the surrounding alligators. I&#8217;m not saying such a map isn&#8217;t useful in avoiding getting killed, but until someone invents a completely new scientific technique that allows us to build a bridge, it won&#8217;t help anyone get&nbsp;home.</p> Comment on David Matthewman on the Whole Brain Emulation roadmap by Paul Crowley 2010-02-20T14:40:28+01:00<p>Having described this as the best objection, on re-reading here I&#8217;m not sure I see its force. The paper is trying to be a fairly comprehensive survey of the foreseeable possibilities, so it covers a variety of scanning technologies, but points out where their shortcomings may leave them unsuitable for the task; the discussion of <span class="caps">KESM</span> makes it clear that it could not be a complete solution because of the wavelength on which it operates, just as you discuss. And I find over six pages of discussion of <span class="caps">EM</span>-based scanning techniques, most of which <span class="caps">AFAICT</span> seem to be forms of&nbsp;<span class="caps">SEM</span>.</p> <p>However, I defer to your expertise on electron microscopy &mdash; any gaps in their discussion will be much clearer to you than to&nbsp;me.</p>