Japanese BCI Research Enters the Skull Wednesday, Apr 16 2008
cybernetics 11:22 pm
Researchers at Osaka University are stepping up efforts to develop robotic body parts controlled by thought, by placing electrode sheets directly on the surface of the brain. Led by Osaka University Medical School neurosurgery professor Toshiki Yoshimine, the research marks Japan’s first foray into invasive (i.e. requiring open-skull surgery) brain-machine interface research on human test subjects. The aim of the research is to develop real-time mind-controlled robotic limbs for the disabled, according to an announcement made at an April 16 symposium in Aichi prefecture.
(Via Pink Tentacle.)
For Brain-Computer Interfaces to get anywhere, they’ll need to interface with the brain directly. Hot brain-on-computer action. Anything less may perform interesting tricks, but will never make machine intelligence truly available to human thinkers. The key is to increase electrode density as much as possible, while figuring out what all the neurons do. Simple to say, difficult to achieve.
To me, Brain-Computer Interfacing seems like such a difficult path to greater-than-human intelligence that Artificial General Intelligence will reach the milestone far earlier. Unfortunately, this won’t stop many from ignoring AGI and focusing on BCI, because the BCI path allows you imagine yourself getting the enhancement, while AGI is a foreign other. I see this as a simplistic and xenophobic way of thinking about the situation. (Not to say that BCI is always a silly idea — just that it is if your main reason for supporting it is carbon chauvinism.) If AGI is brought up carefully, like a precious flower in a garden, it will not only help make intelligence enhancement available to all humans, but will introduce us to a world of thoughtful minds with an entirely different perspective than our own. This prospect is not something to be feared, but cautiously embraced.
Although I agree brain interfacing could be farther off, I don’t agree that it’s just xenophobic. I’m definitely going to be xenophobic about an intelligence that only needs one hiccup of a bug to kill us all. On the other hand, a enhanced human may become psychotic partway during his “upgrading”, or just as suddenly change opinions on what his/her idea of the human race should be/do. It’s easy to sit on the fence about this.
I don’t know, perhaps AGI will be easier to acheive, but I always thought that BCI would be easier, since you already have the power of the human brain to work with, all the functions, neurons etc., are already right there.
With AGI, it is going to take a lot of effort just to get the computer or machine up to an equivalent human level intelligence, nevermind a greater than human intelligence. There are some people out on the Net, ( I’m certainly NOT one of them) who feel that even getting to an equivalent human intelligence in a machine is absoultely impossible. If they happen to be right, then BCI might be the thing.
I want for AGI to happen as much as the next guy, but if we want increased intelligence, we may have to keep our minds open for other ideas in case the naysayers turn out to be correct.
James Frye and friends have already performed accurate, real-time cortical simulations on mice. For human-level simulation, some researchers are talking about a 2020 timeline. Stanford has developed a tool that scans molecules down to
the nanometer. The European Commission is funding research into simulating the entire human body at the molecular level. fMRI has further opened avenues of research into the inner workings of the human brain. Emotiv Systems is commercializing BCIs that help control and manipulate dreams. The list goes on… The objections that we don’t understand what is going in the brain are falling before new advances.
Once yer at this point, the rest is just engineering.
Many objections to AGI state that we need ten to twenty petaflops to even begin to match human level processing speeds. Brian Wang had post at NextBigFuture in February that addressed the near-future advent of petaflop personal computing by 2013 – five years from now – with 100 plus petaflop super-computing in about the same time frame.
The technical objections to BCI and to AGI are starting to disappear under current factually based evidences.
Warren, you forgot the objection, “all this stuff scares me, so I want to ignore it”.
But how do we know for sure that 10 to 20 petaflops is the equivalent of human intelligence? Can we really say this with confidence? What if it’s much higher, like 10^30- 10^40, which I have heard is the classical limit to “cold” computing for a kilogram of matter. Even at that point, if you put a bit on every atom in a kilogram, each atom would have to do around 10^15 bit flips/second, which many people doubt is possible.
As I understand, there are also some who say that the brain acts as some kind of quantum computer (although support for this seems to be very limited at best) If that’s the case, then we may be a VERY long way off ( maybe millennia or longer) from getting human level intelligence in a machine.
I’m sorry if I sound like a pessimist. Actually I really hope that the technologies discussed here come to pass. I am quite a newbie in terms of AI and the Singularity, so forgive me if my comments sound a bit entry level. I’m just pointing out some of the things I’ve heard from others on the Net (mostly on the MindX forums) who feel that these things are are silly to hope for.
Michael, as always I have to (this time quite) respectfully disagree with your conclusions regarding AGI v. BCI. But for once, it will be for a slightly different avenue.
I think that your argument that AGI will come sooner than BCI really only applies to the ‘core elements’ of biological enhancement v. synthetic ‘enhancement’. That is to say, you’re looking at BCI being used to make people fundamentally smarter in the way a seed AGI could recursively self-improve.
It is //my// contention that this is wholly unnecessary to create a radically, fundamentally superior human. Further, it is //also// my contention that any research that advances one goal will also advance the other. So we cannot say that one will be “far sooner” than the other. Cognitive science is necessary for both — but here’s my real thinking:
It is already theoretically possible to augment a human mind; all we are left with is the engineering difficulty of introducing the silicon-substrate neocortical models to the ‘natural’ ones in a functional manner. Do that, and presto! Instant smarter-than-human human. That is to say, we only need a partial operating model for general intelligence in order to enhance an already-functioning human. The same cannot be said of AGI.
Ignoring all this, however, we are left with //another// avenue of approach that would leave a human being significantly better off than we are today: “data-in” BCI. By data, I do not mean sensory input, but rather memory input. If becoming a polymath with near-instantaneous ability to retrieve new information was as easy as undergoing a surgical procedure no riskier than breast augmentation — well, I’d be more than willing to go 10k in the whole for that. Hell, that’s cheaper than a college education.
Creating a community of 10,000+ polymaths would radically and significantly alter human events. Make even 10,000,000 — and then you’re talking about singularity-level undertakings: a point beyond which prediction is simply not possible.
I see the possibility of “digital memory interfacing” as being a far less complicated task than generating an AGI that can actually outperform even an average human being.
By the way — I see you’re using Gravatars now. Nifty. You might think about picking up one yourself.
Ah, I see the answer to my previous dumb question is here. I shall investigate “gravatars”!
Question for Michael: I tried to sign up to accelerating future using my gmail email, but it said that there was already an account set up using this email. I didn’t do it, so I assume that something has gone wrong… Perhaps WordPress saw that I had already posted using that email.
Gravatars are “globally recognized avatars”. The link is to the main page; basically, you sign up there with a specific e-mail address and the image link will call up whatever image you uploaded as coincides with the e-mail address you use when making your comment.
Different blog commenting systems allow for it — and the website even has a walkthrough for that. Haloscan, which I used to replace Blogger’s commenting “engine”, has Gravatar support as one of their features you can turn on or off.
It certainly adds a nice touch to the site, I think — and since it’s calling via another link, doesn’t suck up bandwidth for Michael, which is even better given the etymological commonality between ‘hostage’ and ‘hosting’. >;- )
BCI being used to make people fundamentally smarter in the way a seed AGI could recursively self-improve.
Correct.
It is //my// contention that this is wholly unnecessary to create a radically, fundamentally superior human.
If cautious transhumanists don’t, someone else will, and do we want that person/group to be egoistic/undemocratic?
I think I prefer the cautious transhumanists, ones that open source the technology. Guide it to open and safe ends.
Further, it is //also// my contention that any research that advances one goal will also advance the other.
Not if BCI doesn’t progress to the point of producing seriously significant upgrades. Otherwise you’re just using it for stuff like manipulating a mouse and turning your light bulbs on and off. Wooo.
Cognitive science is necessary for both
Agreed… there is overlap. AGI designers have to know about the brain, just as BCI researchers obviously do.
It is already theoretically possible to augment a human mind; all we are left with is the engineering difficulty of introducing the silicon-substrate neocortical models to the ‘natural’ ones in a functional manner. Do that, and presto! Instant smarter-than-human human. That is to say, we only need a partial operating model for general intelligence in order to enhance an already-functioning human. The same cannot be said of AGI.
You keep saying this. If it’s so easy, should I expect it in the news in the next five years? Will we have a Singularity in time for 2012?
but rather memory input. If becoming a polymath with near-instantaneous ability to retrieve new information was as easy as undergoing a surgical procedure no riskier than breast augmentation — well, I’d be more than willing to go 10k in the whole for that. Hell, that’s cheaper than a college education.
Ah, yes, Matrix style. Unfortunately, memory may be intricately tangled up in complex networks of millions of itsy bitsy neurons distributed across the brain. The brain may say, “use nanobots to enhance me or GTFO”.
Creating a community of 10,000+ polymaths would radically and significantly alter human events. Make even 10,000,000 — and then you’re talking about singularity-level undertakings: a point beyond which prediction is simply not possible.
Even creating one would be a big deal. But I doubt memory input would be so easy. The format might not be the same across brains. Each new idea is related to every single idea the person had before. Simply manipulating/stimulating the neurons a certain way using crude BCI might make the knowledge oddly disconnected or even useless. We are talking about a complex web of 100 billion little nodes, each 10 microns across. Ultimately, the brain is a computer, but it’s an extremely complex one.
“But how do we know for sure that 10 to 20 petaflops is the equivalent of human intelligence? Can we really say this with confidence? What if it’s much higher, like 10^30- 10^40, which I have heard is the classical limit to “cold” computing for a kilogram of matter.”
First, we only need as much computing power as the brain if what we want to do is run a model of a brain. There are probably more elegant ways to do AGI that would require less computing power.
As for the computing power of the brain, we can estimate it by knowing how much energy it uses, how fast neurons fire (100-200hz), how many neurons there are, etc.. I’m not expert, but I’m fairly certain that it’s far from the theoretical limit of computing power for that amount of matter. Evolution builds robust stuff that works well enough, not the most efficient and high-performance version possible.
That’s true, but perhaps maybe not so relevant. We needn’t duplicate the memory structure itself, but rather introduce information through the means by which the brain processes memory. I seem to recall something about the hippocampus being involved in that. Rather than reproduce the source, we could reproduce the //signal//. :)
I said it would be easier; not that the work is being done — at least, not consciously. If I would say that the bet primarily rests on Ted Berger’s implantable artificial neurons (meant, of course, for the medicinal purpose of restoring memory functions to those with Alzheimer’s Syndrome at the moment. You have to admit, Michael, that the fact that someone is working on the engineering aspects of something that requires no theoretical development is a sight closer to ‘real-world implementation than is something that we don’t even have a sufficient theory to even model, let alone work out the engineering problems of.
Just noticed that I hadn’t addressed this point of Michael’s. It’s worth addressing.
I remain unconvinced that such a technological development will rest solely in the hands of one group or person — though initially speaking it will most likely be in the hands of DARPA, if we can prognosticate effectively from modern trends.
But yes, Michael — I would far rather such technology be open-sourced. With what I envision being possible for early BCI, however, I’m not at all concerned that such an entity would block progress towards a Friendly posthuman intelligence (or plurality of such intelligences). This is for the same reason that superior genetic mutations often disappear within a population despite the common idea of evolution being survival of the fittest: the principle of reversion to the mean — even the smartest man in the world will be beaten by a sufficient number of dumb opponents. And, at 6 Billion people and growing, I suspect we’ll have enough ‘dumb opponents’ to counteract anything but a truly hard takeoff recursively improving intelligence for quite some time to come.
<>
This is rather annoying and, because of its plausibility, a bit depressing. We should strongly encourage global cultural memetic evolution to counteract this lamentably likely (or at least not-especially-unlikely) neophobic upshot. ‘Course I’m preaching to my fellow choir members, here, as it were… ;)
And, as far as which development—AGI or BCI/IA—to “prefer” or “root for”, as I’ve opined more than once in the recent past both here and on various other blogs, they’re respective R&D trajectories may (at least partially) converge/co-evolve from here on out, since it may very well prove fruitful (at least in the minimal sense of being marginally more convenient/helpful/productive), e.g., for Ben (Goertzel, that is) to directly interface, within the Novamente VR simulacrum, with his proto-AGI “child”-entity (i.e., “directly” “inhabit” that entity’s “expereinced”/”perceived” environment *as a “tangible” entity in his [Ben's] own right*). Indeed, I suspect that, as/when such capability comes to be in the offing, such tech would be *very* fruitful indeed, quite possibly in further-accelerating both AGI’s **and** BCI/IA’s development. It will be interesting (duh! ;) ) to watch this possible (and, in my [admittedly NON-cutting-edge] opinion, farily highly-likely) convergence over the next 3-12 yrs…[likely, btw, imo, because much of BCI---such as the Aussie push---is geared toward ultimately enabling, among many other things/goals of course, a human to completely "inhabit" a VR game-world---more-or-less like "Lawnmower Man" meets "Wild Palms"---and obviously such tech would allow for precisely the Novamente scenario I posit... ;) ]
Let’s keep our eyes on the prize(s), boys & girls…and keep those persuasive, well-articulated transhumnist, liberal, technoprogessive memes a-comin’!!
Ciao…
“reversion to the mean — even the smartest man in the world will be beaten by a sufficient number of dumb opponents. And, at 6 Billion people and growing, I suspect we’ll have enough ‘dumb opponents’ to counteract anything but a truly hard takeoff recursively improving intelligence for quite some time to come.”
The above is what was supposed to appear between the two “”. The system unexpectedly omitted it! WTF?!!?!! ;)
MCP2012 —
Mike’s blog eats incorrectly formatted html tags. Using ” < ” and ” > ” to surround text is a bad idea unless you absolutely have to — in which case, I suggest replacing ‘em with the following characters: ” < “= ” < ” | ” > ” = ” > “
It seems to me that AGI is the easier, and probably safer path to superintelligence.
Many people have spoken about how BCI will be easy because the brain is already there: as if nature has done most of the work for you, all you have to do is plug an extant brain into an extant computer, maybe tweak it a bit, and hey presto!
But look at it this way: would you try to build an aeroplane by grafting your latest pair of wings to the arms of a human? It was tried: it didn’t work. Why not? Well, one can look at the first successful attempts at flight using a human/machine hybrid (which required a much more advanced machine than the first machine used in machine-only flight). It turned out that the existing human body was simply not optimized for doing the job required of it (poor power to weight ratio).
The human brain contains about 10^12 neurons, which we can think of as dots on a graph, and 10^15 axons, which are connections between them. Imagine a graph with 10^12 dots, and roughly 1000 edges per dot… It’s not a pretty thought for a scientist. It’ll be a complete mess: evolved systems always are. Most people on this blog have probably heard about how your esophagus crosses your windpipe unnecessarily, leading to the danger of choking because evolution cannot look ahead, and mistakes get frozen in. Imagine that, multiplied by 10^15.
Now it may be possible to interface this mess with a computer, but because of the messyness on the human brain side, things WILL go wrong. Things like, for example, psychological disorders in the human concerned, or perhaps irreversible brain damage. (not to mention purely implementational risks like the risk of infection, which was a big worry for Kevin Warwick when he did some interfacing)
We worry about friendly AI, but how much more difficult will friendly BCI be? You interface the first 10 people with a computer, they all go mad. Person no.11 seems fine, until you realize too late that she has been acting essentially like an uFAI.
Also: I’ve cross-posted my comment on my blog. Thanks for sorting my account out, Michael :-0
Roko wrote:
False comparison. The people who did that never had the lift ratios necessary — because they didn’t even know to do so.
Apples and oranges here.
Roko Wrote:
Not quite:
Roko wrote:
It is precisely that messiness which might actually lend to the facility of incorporation of new substrates within the existing neural framework.
From Scientific American:
From MIT’s Technology Review:
The brain is essentially a highly redundant, highly adaptable piece of machinery that fails to differentiate between hardware and software.
AGI, on the other hand, requires us to successfully deduce a digital process which can reproduce the fuzziness already extant. It is for //this// reason that I feel that augmenting human mental capacities is likely going to be a ‘by-blow’ of reaching human-level AGI.
Roko, I wrote more on this here. I would ask you to look over it and see how it impacts on your views on this issue. :)
BCI is way better than AGI.
Firstly, everyone will be scared of AGIs, and won’t want to build them. So, research will be halted, or severely slowed. BCIs, on the other hand, will continue to be developed for prosthetic limbs or just for gaming purposes.
Secondly, the electrode density is reaching a limit, not unlike that for transistor density. We are due for a paradigm shift, perhaps using something other than electrical signals to get information into the brain.
On the other hand, perhaps we will need a crude AGI – on par with a mouse brain – in order to be able to correctly interface with the human brain. It could be used to work out what each neuron does.
Just musing here…and besides, BCIs are more my field (being a neuroscience student), so call me biased.
What the devil else could be used that would have a //higher// information-to-volume ratio than electricity? I find this honestly baffling.
I’ve seen reports that indicate we might move over to carbon nanotubes for signal transception.
“Carbon chauvinism”? Is such name-calling necessary? One could make equally unhelpful remarks about the highly optimistic notion that AGI “if brought up properly” will enhance all of us and we will live happily thereafter. We don’t know how to AGI can be or if it can be “brought up properly”. We have no means of confidently stating what AGI will or will not do with respect to humans.
Besides we all are humans here. Of course we are going to find solutions appealing that enhance creatures like ourselves directly. Of course we are going to find such solutions more “like us” and thus perhaps much easier to predict. A lot of our sense of safety is dependent on a workable theory of mind, a model of the other we are reasonably confident off. I would not call that “chauvinistic” in the least.
It also should be pointed out that enhancing ourselves as much as possible does make it much more likely we will solve our manifold problems including creating Friendly AI.
Firstly, allow me to say that i think this a great blog. i’ve been lurking for a whie but this is my first post.
it seems that procedural memory is perhaps the best candidate for ‘digital memory interfacing’, although some very basic forms of declarative knowledge may also be in the running. it seems more likely that the way we represent more complex forms of knowledge is deeply entangled with experience, which is deeply entangled with personal identity. if so, we may not be able to simply acquire these kinds of knowledge via BCI without a) experiencing a significant shift in personal identity b) just fucking the whole consistency of the system up (think here of the challenge of importing part of an image from one hologram into a different hologram – you need to examine the totality of the first system and change the totality of the second system for the process to work without loss of resolution (note that by the use of this analogy I’m not trying to imply that any aspect of the brain is necessarily holographic, instead just that similar principles may apply). This doesn’t mean though that the whole process nonetheless might not be possible, or that we shouldn’t be optimistic; just that we should be cautious about the plausibility of simply ‘dumping’ new information into our brain that isn’t tied to very specific modular domains, such perhaps as motor skills or recognition of basic facts.
Of course a more modest version of BCI would simply give us something like direct cognitive access to electronic information; a neural portal to wikipedia perhaps. But I imagine that this is far from what most would regard as digital memory interfacing, where it is expected that complex understanding can be implanted at a deep level.
It is quite true that it is basically impossible to separate experience from knowledge in the ordinary circumstances for humans. But if the baud rate of communication for ‘fetching’ of information from the “neural wikipedia” is high enough, the functional difference could be effectively moot. Especially if the process is an active narrow AI rather than a passive ‘query’ system.
Certainly, however, your comment touches on the incredible dangers to personal identity such technology could face. That’s one reason why using the approach of ‘active memory’ — that is, making the information available to the cognition process (say, via the hippocampus) rather than incorporating them as actual memories — would actually be much safer: It would be, to invoke a computer metaphor, like the difference between running your system as root on a Unix OS without a firewall and running as a standard user with a firewall.
Ian, don’t be so sure that an intelligence-enhanced human couldn’t overpower 6 billion humans. Random genetic mutations give people a teeny tiny intelligence advantage over their peers — bitflip mutations simply can’t improve the hardware in big steps. However, technology is a different beast entirely. With intelligently directed, specially targeted interventions, it may be possible to create a being smarter than humans in the way that we’re smarter than Homo erectus was. And it’s entirely possible it could take over the world. There all sorts of tricks that we’re too dumb to think of.
Roko, I agree with your comment entirely. Biology is a messy thing. Human “normality”, both in terms of intelligence and morals, are a tiny portion of the state space. Perturb it slightly, and you have a maniac on your hands. A superintelligent maniac would likely have no trouble appearing absolutely sane to its human-level victims, until it’s too late. Even more annoyingly, superintelligent humans might prefer to dominate humanity rather than killing it outright. This could be unpleasant enough that we would prefer to commit suicide rather than dealing with it.
Joshua, currently there is far more serious effort going into AGI, more people support it, and it’s further along. A reasonably-sized group of wealthy/intelligent people care about Friendly AI. Regarding BCI: the brain is fundamentally electrochemical, it’s the language you have to talk to it in. Yes, you may be biased. Hopefully I can persuade you to at least think about AI a little more.
Snape, good points, but many people are fundamentally xenophobic and anthropocentric about this whole Singularity thing. I feel like encouraging people to drop the attitude, while being as polite as I can. “Carbon chauvinism” is not as insulting of a term as you’re making it out to be. It just refers to someone who only things in terms of organics and has trouble imagining synthetic intelligence that has the same cognitive and moral capabilities as humans. Not a sin, but short-sighted in my opinion.
Even with all those caveats about theories of mind, I still think AGI can be made safer. This is because AGIs can be built without the trappings of Darwinian construction. See “Creating Friendly AI” for the long version.
That’s true, but I think you’re viewing the timeline of Homo erectus‘s death as being a bit too antagonistic with us. Even if we just view the process as competition, it’s worth noting that the populations of Homo erectus and Homo sapiens were never that dissimilar. If there were, say, a million of ‘my’ ‘biomech augments’ (term chosen because I’ve always wanted to use those words in a sentence legitimately! :) ), then sure — there’d be a problem assuming they were all looking to be unFriendly. But at some point we must recognize that no matter how smart you are, a big enough rock will still kill you when it hits you. Six or seven or even eight billion people is a pretty damned big rock.
Besides — what if we were to also take these fancy mirror neurons and augment them as well? Turn all these super-geniuses into super-empaths as well. I seem to recall having discussed that possibility in the past as well.
Now, sure — it wouldn’t be a 100% effective thing: not every person would become sufficiently Friendly to ‘offset’ the risk of unFriendliness in others. But a sufficiently great number of individuals more than likely //would// — in my non-scientific estimation, anyhow.
This, I think, is an area we will have to simply agree to disagree on. The challenges of defining a goalset that would manufacture Friendliness within an AGI haven’t even really be defined yet, let alone how we’ll build an AGI. I //DO// differ on ‘mainstream’ science’s opinion of the possibility of constructing an AGI — to agree with you, Michael — but I remain unconvinced in the manner you have been.
Penultimately, however, I’m really not that worried about it: the two are mutually non-exclusive to each other as approaches, and both have viable avenues of creating strongly superior minds that have humanity’s best interests in mind. So it’s a “po-tay-toe” / “pa-tah-tah” thing, insofar as I’m concerned.
Main reason I truly support BCI as opposed to AGI is the personal experience implications. I.e.; I am selfish at heart. (Though I //do// know enough to separate personal selfishness from philosophical implications!)
Six or seven billion people won’t be hitting you though; there’s a problem with the concentration of power, right? By the time everyone is aware of said unfriendly Augment, she is likely extremely powerful already.
A million people armed only with rocks can’t stop a single M1A1 Abrams; a very intelligent being could arrange that kind of power gap as well, compared to the populace.
If you are smart enough, the big rock never hits you.
A) Not true about the Abrams. They certainly could do so — if only by throwing themselves in its tracks and then mass-swarming the access points. Hundreds would die — perhaps thousands — but it could be done if they were dedicated to it.
B ) I do believe that this description more accurately fits the comparative differences between a seed-IA; which has little to do with what I’m talking about here. I highly doubt that the //early// individuals will have anywhere near the overbalancing intellect necessary to make that sort of a gap.
Just depends on the size of the rock, now doesn’t it? :) There are, last time I checked, some //awfully// damned big rocks out there. And a human-based superintelligence would have to have been around an awfully long time, I imagine, to become immune to nuclear weaponry.
You know what one viable solution to a really high-tech lock is? Blow it up.
People seem to be split on this.
IConrad likes the idea of the brain being “fuzzy”, I say it’s a “mess”. I suppose two people can look at the same system and see it in different lights.
I don’t worry very much about the feasibility (or otherwise) of various BCI techniques. If they work, they work, and I don’t think I have enough information right now to know one way or another.
What worries me is the probability of things going nasty: unfriendly BCI. Now, unfriendly AGI is a distinct possibility too, so in order to argue against BCI, I have to argue that on balance, the failure modes of BCI are more serious than those for AGI. This is, given our state of ignorance today, a tall order.
Arguments about “throwing big rocks” etc seem to apply equally to BCI and AGI as far as I can see. It seems likely that the same defense works in both cases: the AGI/BCI-person is cleverer than you are, so they’ve thought of a counter strategy that is simply beyond your intellectual reach. Saying that you can kill a BCI enabled supervillain with a nuke is a bit like saying you can defeat Gary Kasparov at chess if you could just get his king on its own with your queen and your rook…
Here’s what I can think of:
1. if you’re doing BCI, you have to choose one person who will get augmented (or perhaps a small group who will get augmented). This means that you have a strong incentive for people to fight and squabble over who gets to be augmented before you’ve even started.
2. BCI is likely to be dangerous for the human concerned. This means that, on average, cautious people will not want to go first; the first generation of BCI patients will self-select for recklessness.
3. The messyness of the human brain means that you will not be able to have a provably friendly BCI. You can’t do software verification on someone’s brain.
4. Similarly, you cannot do open-source BCI: well, I presume there will be no way to make someone’s mind open source (although if one could do that, then I’d be very interested…)
This comment is mostly for IConrad, but Roko has made good points about lack of information. (My lack of information, at least! I shouldn’t presume.)
His first three points are dead on, and that first point is a strike against BCI in favour of AGI, I think.
To continue, the Abrams hatches are sealed, so I doubt swarming the access points would work. When the tank ran out of fuel, the drivers would be in trouble anyway, so I concede your point, and yes, the description fits farther in the progression.
That might be the point of concern though; UnFriendlys can hide their early progress for some time. You need a fair number of smart people watching them, as you say; but with many convergent subgoals, distinguishing the Friendly and UnFriendly is difficult.
I wish I knew enough about the potential rate of intellect progress to say whether that is sufficient.
This also seems to be a problem for open-source BCI; you just gave the technology to Roko’s self-selected reckless people, and you can’t watch them anymore. At least in a single lab there’s some direct control, and it is possible to find the person if you are worried.
As for open sourcing BCI, releasing the hardware designs and software used would seem to count as open source to me. Open sourcing each individual would be rather difficult, yes.
I’ve been reading your Functionalism blog, and have been impressed with your writing. Are there any technical books you’d particularly recommend, on biotechnology, say, or one of your other interests?
(An aside: Nuclear weaponry immunity already exists for small targets. Most armoured vehicles are hardened against EMP; give them launch warning and paranoia, then go in a line for an hour, the blast wave doesn’t hit them. Faster delivery than missiles is required. See Michael’s Technological Transcension for a near future version of this.)
This pre-assumes that they themselves aren’t the target… or that someone hasn’t carried the weapon up to them in a suitcase. There is no force known to humanity that could protect a person from being at ground zero of a nuclear detonation.
@kraryal: His first three points are dead on, and that first point is a strike against BCI in favour of AGI, I think.
I meant them to all be disadvantages of a BCI based attempt at superintelligence
Sorry Roko, didn’t mean to put words in your mouth. I was viewing BCI and AGI as a dichotomy, since they seem to be the only two methods available. (BCI would include uploading, it seems to me.)
I did assume the vehicle was the target; once the missile is launched and the vehicle knows, it has time to get out of the way. I specified paranoia, so that the driver always assumed a launch meant they were the target. Are there actually suitcase nukes? They would count under my ‘faster delivery’.
That right there, sir, is why I think you and I are not seeing eye-to-eye. You’re thinking of the mind as a software problem rather than as a hardware problem. One simple way to ensure Friendliness in a human is to ensure that empathy is augmented as much as — or more so than — intelligence. And as there is a neurological basis for empathy, that’s a fairly safe approach.
Meanwhile, we have no way of knowing how to produce Friendliness with a set of code in a digital AGI.
@IConrad, a major issues with increasing implant electrode density is the electrical interference between electrodes. So, you’re either going to have to insulate the electrodes perfectly, or use mechanical or light-based means to stimulate and record from neurons.
@Michael, the only reason AI research is going ahead so strongly now is that it currently can’t even beat a honeybee’s brain. Once it starts getting up around the canine level, I think we will start to see people panicking and banning research.
There’s a couple of folks here who have a *lot* of reading to do. Then they need to learn about the proper uses of metaphor and allegory. (You can really, *really* confuse yourselves and others if ya don’t have that part right.)
…besides, as a Veteran Sergeant of Marines, I can think of several ways to use a million rocks to easily stop a modern tank… Especially if the tank’s been isolated.
Under the right circumstances, it only takes one man to stop a tank.
All he has to do is to stand there.
Change Blindness, folks. Learn about it. Appply the lessons to yourselves, first.
It is not that all problems are
irresolvable. It is how we think about these problems that render them unsolvable…even after the factual evidence reveals that they
have already been solved by others.
1. Insomnia is a pain.
2. D’oh! I totally missed photonics.
3. I’d say they’re working on new approaches to the problem of signal interference; such as insulated carbon nanotubes. Given the significant reductions in the power consumption v. signal production ration inherent to using a material that, as a single molecule, is as conductive as a standard copper house-wire, even an equivalent capacity for insulation / passivation would yield strongly superior results. :)
That right there, sir, is why I think you and I are not seeing eye-to-eye. You’re thinking of the mind as a software problem rather than as a hardware problem. One simple way to ensure Friendliness in a human is to ensure that empathy is augmented as much as — or more so than — intelligence. And as there is a neurological basis for empathy, that’s a fairly safe approach.
hmmm. I’m not entirely convinced ;-0!
But presumably you’re saying something like this: it’s easy to make a human behave well: just add more empathy neurons/mirror neurons! I suspect that there is probably truth to this. But, on the other hand, you’re tinkering with an evolved, messy system. How do you know what will actually happen if you put, say, more mirror neurons into someone’s brain, past the maximum that any previous human has had? What happens when you do this at the same time as interfacing them with a computer? I suspect that the answer to the question:
“what will happen to the patient if we do neuro-intervention X”
is likely to be split evenly between
“the patient will go mad”
and
“nothing happens”
with perhaps some very tiny probability of something pleasant happening. This is typical of what happens when you interfere with a very complicated system that you don’t understand. Of course, neuroscientists will accrue more and more knowledge about the human brain over time (at what rate? who knows!), so as time goes on, our understanding of the system will go up, and the probability of a nice outcome will increase.
Now I ought to add, for balance, that AGI is not without problems. AGIs may become very powerful before they really understand what we humans consider valuable in life. I’ve discussed the pros and cons of making a putative superintelligent system more or less like a human mind here.
I think that a good compromise solution might be to try to understand the human mind using brain scanning techniques, and then to implement a human mind fully in software, from scratch, with no copying from any one particular human involved. This would effectively be an “open source person”, and would have several benefits IMO: the open-source person wouldn’t have any hidden personality traits that went in unnoticed, the open-source person’s values and personality could be mutually agreed upon by all interested parties [perhaps even by an international treaty]. This would encourage people to co-operate and negotiate, rather than to sneakily stab everyone else in the back. Also, since there is no evolved system, no neurons, no brain involved, you don’t have a massive complexity/mess issue: one can cream off the finer points of the human mind without all of the messy implementational errors that evolution will have introduced.
I’ve cross-posted my comment on my blog, with some minor additions and corrections.
I kinda push and prod at both ‘sides’ once in awhile, but I wanted you all to know how much I appreciate all of the communications that are shared here at Accelerating Future.
It’s very interesting, very educational, and I’m always left
with something new to think about.
Thank you all.
…and thank you Michael. This is an excellent blog.
Roko, I see your avatar…
Warren, you’re welcome! I just hope it leads to something substantial, like more funding for H+ orgs, or the launch of significant new projects. If you ever have any interesting ideas that might be worth blogging about, send me an email.
Roko, I will leave off at this; I strongly suspect that you have an emotional bias against biological systems. Where you see irreducible complexity, I see a robust, redundant system.
You say that we do not know what will happen; that it is too unpredictable and dangerous. I say that it is at this point //more// predictable than something that has never before existed that we cannot even begin to guess at the correct goal architecture for.
I mean, taken to another extreme; it is well known that the larger the relative portion of a given brain’s structure is, the more significant that portion is. Animals with larger areas of the brain dedicated to scent have stronger senses of smell, for example. Now, we know that mirror neurons fire during the process of empathy. It is also known (low confidence threshold at this point, but still reliable) that autists typically have weaker mirror neuronal activity. This means that we can guarantee that with a sufficiently higher mirror neuron count, even if our superintelligence is not more empathic, it will certainly be more social. Simply wanting to have other people around is one way to guarantee it won’t eliminate them.
Finally, unlike an AGI, an implant wouldn’t be likely to be one-of-a-kind at implementation nor would it be likely to be even remotely capable of a “hard” takeoff.
But, I see Roko’s & Michael’s positions and they //are// rational. I simply disagree.
Where you see irreducible complexity, I see a robust, redundant system.
You say that we do not know what will happen; that it is too unpredictable and dangerous. I say that it is at this point //more// predictable than something that has never before existed that we cannot even begin to guess at the correct goal architecture for.
yeah, I can kind of see where you’re coming from. I can envisage it being the case that the human brain is both robust and predictable, even when you perform a really major intervention on it, like interfacing it with a massive electrode array connected to large knowledge bases, external memory and perhaps even external processing power. It is possible that straightforward extrapolations like
(more mirror neurons) = (friendlier behavior)
will still hold. [wouldn't that make friendly superintelligence easy! - too easy IMO. but still, it might hold to a good approximation ... ] There’s a lot of uncertainty, because as far as I know, no-one has ever done these kinds of things to a human brain.
Some people are expressing worries about BCI leading to unfriendly superintelligences. I agree that this is a risk, given what we know about human psychology (people given absolute power rarely act too friendly). However, I don’t think it’s really a serious risk, for the very reasons of slowness and messiness that have been brought up. Unlike AI where you can have a hard takeoff scenario, human BCI intelligence improvement is likely to be a gradual process, giving time for many people to adopt the first stage of the technology before the second stage of technology will be developed. Ideally, this will lead to a society of beings of approximately equal intelligence, each watching each other for signs of unacceptable behavior – just like society is now.
IMO, the main safety issue in BCI is that enhancing human intelligence by any means makes it easier to build AGI… were it not for that fact, it would probably far more safer.
Regarding Roko’s open-source person proposal:
The main issue I find with it is that it’s unlikely that we’d ever get that far without having AGI already.
For instance, you write:
The open-source person wouldn’t have any hidden personality traits that went in unnoticed, the open-source person’s values and personality could be mutually agreed upon by all interested parties [perhaps even by an international treaty].
This would require us to have a complete understanding of the brain: good enough to know exactly what causes what kinds of personality traits, and an ability to analyze any suggested changes in order to make sure that nobody’s getting in minor changes that would have massive consequences later on (after all, we’re talking about a complex system, here) – either on purpose or by accident. By the time we knew all that, wouldn’t people have created AGI already?
I think that we can add a neuron dendrite connections and that the interface will be possible to move up to the level of communication like speech or better to share a thought! I will volunteer my mind!
I would like to see some sort of simple neural-implant like a CPU port which can be easily upgraded from that site with minimal risk…like a chip port on your scandisk…that way you don’t get obsolete chips embedded directly into the brain which require later upgrades with a nasty surgery.
Also…I read somewhere the chips are kinda brutal and the metal spikes actually eventually ‘burn’ out or destroy delicate nerves which interface with it until the chip stops working at that site.
Better to get some nanotech in there which links brain/nerves to ‘material’ which can both manage the signal and preserve tissue.
Bonus: a ‘port’ can be put almost anywhere…maybe even allow for wi-fi connection to a signal receiver, smart-phone or whatever. We may not even need trodes or direct port-links to external interface arrays.
56. Hey very nice website!! Man .. Beautiful .. Amazing .. I will bookmark your site and take the feeds also…I am happy to find a lot of useful information here in the post, we need work out more strategies in this regard, thanks for sharing. . . . . .
Hello my family member! I want to say that this post is awesome, nice written and include approximately all significant infos. I’d like to see more posts like this .