Facing Power Asymmetry Monday, Jun 18 2007
transhumanism 3:47 am
Transhumanism is about embracing the prospect of using technology to modify our bodies and minds. Even using the Internet is a weak form of transhumanism, because it takes advantage of technologically magnified communications to expose us to viewpoints and cultures which we would have never been exposed to otherwise. If it weren’t for the Internet, would you even have a clue who I am, or what I have to say? Probably not.
An inevitability associated with technological magnification of human capacities is some degree of asymmetry. Some people have Internet access and some don’t. Many hope for a day when Internet access to available to all who want it, and that day is rapidly approaching. Australia recently announced a national broadband plan, for instance, which aims to bring fast and affordable Internet access to 99% of their population in two years time. For a country as diffuse as Australia, with only 20 million people in a space the size of the contiguous United States, this will be a big accomplishment.
When the first true transhuman enhancements become available, such as implants that upgrade sight and hearing, or integrate information from the World Wide Web into augmented reality overlays, some people will have a distinct advantage. The advantage enjoyed by these augmentees may very well exceed the current advantages enjoyed by people with access to the Internet or modern medical technologies. This is an inevitable byproduct of technological advancement and morphological self-determination.
When more radical modification technologies are invented, such as implants that increase human IQ, or allow close cooperation between humans and near-human AIs, the power asymmetry will increase further. To avoid classist fragmentations in this transhuman society, egalitarian and democratic philosophies will need to be disseminated as widely as possible. The haves will need to care about the have nots. Technology can facilitate this, if used wisely. For instance, a child slavery ring was recently uncovered in China partially thanks to an online petition which 300 parents with missing children signed. Satellite surveillance has uncovered the brute reality of hard labor camps in North Korea, leading to greater international pressure on the government.
In the end, it is the most powerful entities at any given time that must care the most. If they don’t care, then an attitude of indifference spreads down to everyone. Acknowledging that the most powerful entities must be the most egalitarian to make a difference is not worshipping power for its own sake - only accepting the pragmatic reality that only a privileged few truly can manipulate the levers that change the world on a wide scale.
This insight is particularly important with regards to future advances which topple human beings from our historical role as the most powerful and intelligent species on Earth. Instead of ignoring the inevitability that transhumans will eventually be in de facto control, or letting the chips fall where they may, we can take an active role in the pre-transhuman era to increase the probability that these new beings will be connected to unaugmented humans empathically. Friendly AI is an important initiative in this direction, probably the most important, due to the speed at which AI technology will be able to improve once it passes the human equivalence threshold.
June 18th, 2007 at 8:59 am
Impressive thinking.
June 18th, 2007 at 10:07 am
I angrily demand (1) fiberoptics and decentralized speed internet for as many people as possible; (2) portable computers and display devices; (3) cheap, safe software for even trivial things and (4) freedom to use any of these without being manipulated or coerced by bastards, such as spammers, data-miners or meddlesome government.
I can’t imagine (yet) what I will be with some extravagant nanoid in my blood that will allow me to feel invigorated every second of my waking day, or IA or real mind-machine interfaces.
But I sure as hell can extrapolate towards a world where computing isn’t the buzzing victorian bulk it is now.
If only for the games I hope that will come ! …. but there is plenty of fun to go around in such a world.
I wait impatiently, almost angrily.
June 18th, 2007 at 10:17 am
Asymmetry has always been around, but it is getting much narrower than in the past. Notice how cell phones have penetrated deep into the third world, bringing access to communication, and its benefits. It was a very short period, perhaps just a little over 20 years between the time cell phones were only a tool for the extremely wealthy, to complete worldwide ubiquity.
My guess is that AGI will have an astoundingly short period of asymmetry, compared to past technologies. So, perhaps, in the near future, asymmetry will only be there as a human perceived bias, but mostly irrelevant.
June 18th, 2007 at 11:45 am
Perhaps.
It really depends on where, who and how the AGI is developed. Those in “control” of new technology will likely determine its speed of dissemination and thus its general accessiblity for adoption. As Michael said, “…the most powerful entities must be the most egalitarian…”
June 18th, 2007 at 5:36 pm
Welcome back to the present.
It’s a good thing to recognize (as I have been over the past year or so) that the problems with emerging tech in the future will be pretty much the same as we have today, because it gives us the idea that we can ALL do something about it NOW.
Most humans already live in a world ruled over by more powerful minorities. Despite the novelty of formulating these minorities as fanciful AIs or uploads or cyborgs, the general fact of subordination to more powerful forces is well known and understood.
Same story, different commercials.
Not only that, but the story as it’s told now is itself a retelling of the story that started 200-300 years ago (and more), as the enlightenment led people to demand more egalitarian and democratic societies from the monarchies and dictatorships that we’d been pushed around by since the dawn of recorded history.
Good to hear.
June 18th, 2007 at 7:48 pm
Well said. All tools can be used positively or negatively depending on the intention and the application, and the destructive uses of AI seem to preoccupy our policy-making skills. But pointing out ways in which the tools themselves will, with some awareness and coaching, help generate egalitarian and democratic philosophies, provides an optimistic yet realistic antinote to the shadowy overtones of existential threats. We can be proactive in our search for ways to resolve asymmetry and promote our core values as technological advance accelerates. Thanks for helping to light the way.
June 18th, 2007 at 8:23 pm
Nato,
This is not the same story… with potentially non-human-like psychologies, these new agents are not “the same old kings”. “Subordination” is not necessarily a given, in this situation there could be entities far more powerful than us without the least interest in subjugating us. That subjugation effect comes from the fact that humans developed in a highly competitive evolutionary environment and have innate genetic tendencies to undermine one another. These new agents will be designed artificially and will not necessarily retain the cognitive or behavioral tendencies of their predecessors.
The potential asymmetry in this case is also much greater. Rather than bothering about oppression or whatever, a truly superior being indifferent to humans could probably wipe us out in a matter of days or even hours using fast, self-replicating actuators. There would be no social or military struggle to speak of. The threat could be entirely disconnected from us empathically - not particularly spiteful of us but just purely indifferent.
It’s tempting to call this the same story as before… the statement reminds me something my dad said about what he thinks that if AI “rebelled”, it would be “no different than human workers rebelling”. There’s a word for when people assume a situation filled with nonhuman agents would be highly analogous to a historical situation with human agents - anthropomorphism!
June 19th, 2007 at 2:24 am
There’s a good reason why we should consider it as being the same story as before.
Humans have a tendency to project their image onto everything as you yourself point out. This is far beyond just “thinking” about it in human terms, it defines everything that humans have achieved and developed up until this point in time. We cannot do anything else but this, since we are humans.
When we analyse the world around us, we do so on a human level, and we shape the world around us according to human needs, wants and desires. All technological inventions have, up until this point, been extensions and amplifications of ourselves.
It is surely logical, then, that advances in AI are geared around creating not something “new”, but a replica of a human being, a replica of human consciousness. It may be stronger, faster, more clever, but I personally cannot see how it can be anything more than a Human2.0
Any human creation will surely be defined in human terms because that is all we know, so it simply does not stand to reason that we are going to create a new being with traits far removed from our own.
Also, is it not the case that powerful, vested interests will not likely attempt to spawn something that will usurp their dominance?
Governments and corporations the world over want tools that give them power, or will help make money, and since these institutions are the primary backers in the development of technology, it is extremely unlikely that they would ever give that up.
June 19th, 2007 at 8:30 am
Not to be contrarian here, but that really seems to depend, to me, on what the goal of the people making the AI happens to be. While hardly likely, it seems entirely possible (if less than plausible), that some form of AGI might arise as an almost accidental product from some “broadly narrow” AI project.
I think it’s a question of the route that is taken to get there.
June 19th, 2007 at 10:25 am
Are you aware just how many features of human cognition are unique idiosyncrasies introduced by evolution, which would not be present in an AI unless explicitly built in, painstakingly, one by one? Michael has posted on this topic a few times before. I find it much easier to believe that before creating humanoid AGI, the designers would go for the much easier project of creating a more general AGI, which would be utterly alien.
The danger is that someone will design it without realizing how likely it is to, well, kill us all.
June 19th, 2007 at 11:01 am
Thanks for the responses. Interesting! Forgive me if I don’t come across well, I don’t know all the right terminology for some of this.
“but that really seems to depend, to me, on what the goal of the people making the AI happens to be”
That’s really part of what I was trying to get at: “the goal of the people making the AI” i.e. human goals/constraints are being used to define an AI, so shouldn’t therefore the AI in question possess human characteristics, even if they are limited, or only a small proportion of the whole complexity of what a human is?
I understand it’s going to be vastly easier to only concentrate on one or to aspects of human cognition, but I guess I have difficult in understanding how you can take XYZ components of human cognition and create something that is wholly alien to what XYZ does/means to a human.
June 19th, 2007 at 12:08 pm
The co-development/evolution of AGI and IA, if properly done (there’s the rub—but I have every confidence in Ben, Eli, and others to navigate that terrain…) should help both “(tans)humanize” the AGI mind, and ultimately produce a Christ-like/Buddha-like “personality” (or, at least, *intentionality*). Can we develop a quasi-human synthetic supermind, while filtering-out (or just explicitly *omitting*) the hominid dominance/territory/powerlusting/red-in-tooth-&-claw nastiness? (Quasi-human, that is, at least to the extent that it will be able to *relate* to humans, and capable of meaningful interaction with [trans]humans…) Well, that’s what Ben & Eli are already doing their nascent R&D for. And, fortunately for the world, they & their colleagues/peers are pretty bright fellas (much brighter than this ol’ yours-truly, that’s for damn sure!
), and so there is reasonable cause for hope, if not, indeed, optimism…
IA comes into the (meta)scenario because it, too, is being driven by current global human cultural needs and ambitions, it so will start to be ever better-funded (you heard it here first…). So a convergence (or at least significant overlap) of these two tech R&D trends will surely emerge over the next 5-15 yrs. And, again fortunately, SIAI (The Singularity Institute for Artificial Intelligence) is well-positioned & poised to be the leader (or at least one of the leaders) in all this…
And a synthetic FAGI need not be utterly NON-human-like, but it will certainly be super- or hyper-intelligent, and thus just as certainly NON-*typical-human*-like. If it can be fundamentally infused with a proper (from a transhumanist perspective) psycho-(applied-)axiological “core” or “constitution” (to use Moravec’s term…), then Irving J. Good’s pronouncement will indeed be true: It will be the last invention humans need come-up with.
See also Bill Hibbard’s **Super-Intelligent Machines** (Kluwer Academic/Plenum, 2002)
June 19th, 2007 at 1:23 pm
I think something’s getting lost in the medium. Let me rephrase: your argument, here, only seems to hold up to snuff if the goal of those designing the AGI is, well, to design an AGI.
Now, that being said: I still fail to find the argument, “Mental faculty “X” was evolved, and therefore wouldn’t be present at all in any AGI to be built” somewhat specious: What proof do we have that this evolutionarily achieved GI-feature would not be a mandatory feature to exist? For example; political/social awareness — what proof do we have that this would not be present in an non-human intelligence?
This might seem like pointless semanticism — and I must admit it could be — but the point here is: what features of psychology, that we attribute purely to “Evol-psych”, are in fact “inevitable” features of a generalized intelligence?
I don’t think that’s an question we’re prepared to approach an answer for, not at our current understanding of Cog.Sci.
June 19th, 2007 at 4:24 pm
In the end it may be easier to create an AI “from scratch”. I know that can certainly seem to be the case in engineering something new, but we’re not really starting from scratch in most cases. Even though the drawing board starts blank we all have preconceived notions and expectations as we go about attacking a problem from a fresh or new perspective…from scratch.
This is fundamentally different and more like basic research, right? It seems to me that it is more often easier to emulate or mimic something that exists in nature, to deconstruct and reconstruct it than to invent or discover something fundamentally “from scratch.”
I’m still very much of a mind that we’ll continue to see advances in computing, memory, etc. We’ll continue to see advances in simulations. We’ll continue to see advances in biology, genetics, brain and neuroscience. I’m a bit skeptical of the pace for psychology and some of the social sciences, but they’re beginning to come around to computing and simulation as well.
Essentially what I’m saying is that there are a lot more resources aligned and acting in synergy being aimed toward simulating the human brain and/or human intelligence than there are aimed at creating intelligence “from scratch.”
And a simulation of our intelligence is going to, indirectly at least, carry forward our evolutionary past.
How do you focus or re-focus those energies and synergies toward an architectural change aimed at FAI??? Especially when most wouldn’t even begin to agree that they are working to support simulating or emulating human intelligence.
June 19th, 2007 at 6:51 pm
“What proof do we have that this evolutionarily achieved GI-feature would not be a mandatory feature to exist?”
It doesn’t work like that. If you want to say that every single possible general intelligence has a certain characteristic, the burden of proof is on you. It’s like saying “What proof do we have that God does not exist?”.
“It seems to me that it is more often easier to emulate or mimic something that exists in nature, to deconstruct and reconstruct it than to invent or discover something fundamentally “from scratch.””
It is, but the mimics are usually nonfunctional. It is fairly easy, for instance, to create a fairly accurate model of a bird, but in order to actually fly you have to deconstruct the problem.
June 19th, 2007 at 7:06 pm
Michael,
In originally formulating my comment, I almost used the word “struggle” instead of “story”. Most stories told today are built around some form of conflict, so I thought that would be implicit enough.
The struggle I was referring to is the struggle for common people to level the playing field and be considered equals in power and respect despite the tendency of elites of every stripe to do the opposite.
Yes, yes, yes, I can concede (for the moment) the point that AGI does not necessarily have to be created with evolutionarily-derived human biases to (among other things) “undermine one another”, as you put it. My very point, in which I am agreeing with your assessment, is that the story of creating Friendly AI, the struggle that surrounds that prospect, is indeed, as you say, to insure that what “could” be made Friendly actually is made so. The very drive for egalitarianism and democracy, and the fear of inequality and domination, is exactly what galvanizes the Friendly AI project in the first place.
But none of that happens automatically. We have to work at it. That work is the story and the struggle that we face. Sure, AGI could be Friendly. But that isn’t to say that it will be, or, more importantly, that, if it is, that it will have been achieved automatically, or for free. And that much of it is not new, except in the casting of the roles in the struggle.
June 19th, 2007 at 8:56 pm
Tom wrote:
All human-level general intelligences will have some form of pattern recognition and some form of memory. All usefulAGI’s will have some means of interfacing with the world outside of itself.
I will suggest to you, Tom, that your thought that “it doesn’t work like that” is erroneous. Right now, we have no means of saying for certain what features of cognition are ubiquitous and which are simply the result of evolutionary peculiarities that will never be reproduced in other entities.
I posit that in this case, Tom, I’m playing the role of the agnostic and you’re playing the role of the atheist: the “onus of evidence” is not on me: How do you prove that something isn’t known?
June 20th, 2007 at 8:06 am
The reason I was talking about mimicry was exactly because I’m skeptical that we “know” what intelligence is and how it works.
I guess a good question I can throw out would be, “Does anyone know of any great books on the topic of intelligence?”
On Intelligence by Hawkins is the only one I’ve read in recent years.
June 20th, 2007 at 10:24 am
I just had a stumbling block lifted in my mind as I was reading the Wikipedia discussion page for the entry, “Friendly artificial intelligence”. This is something I’m going to touch on, if very briefly, in my blog (It’s not AI directed, but is AI friendly, of course).
It seems to me that one way to couch the terms for a sufficiently “Friendly” goal is to first define what is or is not “sentient”, and then provide the overarching goal of maximizing the range of non-coercive action-potential for all individuals (including itself.) Leave an exclusionary subgoal of the use of coercion to prevent coercion. Let the AI define what is or is not coercive, but leave its original definition as the use of force/violence or the threat of force/violence, without the consent of the recipient.
“Slightly” O/T, but it applies to the whole power asymmetry thing — because if this sort of goal can be ‘inculcated’ into the process, then disparity is irrelevant.
June 20th, 2007 at 12:47 pm
And… Functionalism in Action: Future Government is up. (If you feel this is spamming, Michael, please delete this comment.)
June 20th, 2007 at 1:08 pm
“Right now, we have no means of saying for certain what features of cognition are ubiquitous and which are simply the result of evolutionary peculiarities that will never be reproduced in other entities.”
We have no means of saying for certain, but the list of characteristics that any mind could have is obviously much longer than the list of characteristics that all minds will have. Thus, even given that a characteristic is on the former list, the probability is still very small that it will be on the latter list.
“How do you prove that something isn’t known?”
That’s not the point. If it’s extremely unlikely, we might as well assume it isn’t true until we learn otherwise.
June 20th, 2007 at 1:37 pm
Tom Wrote:
It seems to me that we’re reaching a point where Occam’s Razor swings both ways. Part of this, I’m sure, is poor wording on my part. The only thing I would suggest is that the approach you’re discussing seems dangerously — to me — susceptible to negative bias. Instead of starting from a presumption of “False”, why not start from a presumption of “Unproven”? Of course, this seems a recurring thread of disagreement between Tom and I (When and how to suspend bias.)
We only have, after all, just the one sampling to go off of; so we have no way to confirm what will or will not be a universal.
By way of example: any AGI that co-exists with human beings will, by definition, have some form of social interaction “system”: in humans this is called “politics”. It is extremely unlikely that AGIs will share that political drive. But if they are designed to interact with us, they will definitely be possessed of some mechanism by which to answer the “political problem”. We have no way of knowing, right now, what problems evolution solved which are inherent to the “intelligence problem”. This isn’t to say that evolution’s answers are universal: rather –
“We have no way, now, to know what ‘problems’ all general intelligences require solutions for in order to operate.”
It seems that every time we learn another step in what problems human intelligence is solving — and moreover, how it solves those problems — we take another step in the creation of “human-level” AGI. This suggests a limitation in the realm of possible minds that we are going to be capable of creating.
It was suggested in this blog’s comments (a thread I cannot remember) by someone more eloquent than myself that the range of possible minds might have physical limitations that force similarities with human psyches for any AGI that might actually be constructed. I think, perhaps, I’m somewhat reinventing the wheel here with regards to that line of thinking.
June 20th, 2007 at 4:00 pm
IanC: Congrats on the *Functionalism in Action* blogsite. Looks pretty good…
June 20th, 2007 at 4:10 pm
“Instead of starting from a presumption of “False”, why not start from a presumption of “Unproven”?”
Because “unproven” applies to absolutely everything! You can never “prove” or “disprove” a real-world statement.
“By way of example: any AGI that co-exists with human beings will, by definition, have some form of social interaction “system”: in humans this is called “politics”.”
Beware of misusing the language. Human politics has many, many more implications that “some form of social interaction”. A full explanation of all the strings attached when we say “politics” would take an encyclopedia.
“This suggests a limitation in the realm of possible minds that we are going to be capable of creating.”
This is true. However, the space of minds humans can create is far smaller than the space of possible minds.
June 20th, 2007 at 5:09 pm
Nato,
Good point! Your comment clarifies what you meant. It can be construed as the thrust for egalitarianism and democracy, but definitely in a different context than the struggle has been throughout history. You probably realize how I’m on hair-trigger with regards to even the slightest scent of anthropomorphism… in the last year or so it seems like you’ve begun to look at FAI as a genuinely new problem rather than a human-esque political problem, which is really encouraging and signifies another person that can actually contribute to the discussion.
Hawkeye, How the Mind Works by Steven Pinker and The Adapted Mind, ed. Leda Cosmides and John Tooby are great books. On Intelligence is practically a joke - he takes what we’ve already known for decades about the visual cortex, pretends like he invented it, and then says that that one pattern can be applied to implementing all of intelligence. It’s futile.
Ian, I did grant this to whoever it was, but these similarities are likely to be extremely broad. Any AI will have data on the behavior of gravity, for instance. Behavioral outputs could still be incredibly varied.
June 20th, 2007 at 6:19 pm
Michael Wrote:
I don’t think I was suggesting otherwise. Just reiterating my usual admonition of not taking this as an excuse to believe that there will be no similarities at all, as that is nothing more than “reverse anthropocentrism”.
June 20th, 2007 at 10:30 pm
Oh — MCP2012: Thanks!
June 21st, 2007 at 2:01 pm
IanC: You’re VERY welcome; keep on keepin’ on!!
To my colleagues here: Forgive my more sparce postings lately. Several reasons: Most of the (other) comments are so well-informed (and thus informative), cogent, and spot-on, and the ensuing discussion(s) so superb, I rarely have much to say that hasn’t already been said as well or better than I could anyway; I’m also currently engaged in building no less than 2—yeah, 2—businesses, which, fortunately, will have a partially overlapping client base. But still, building one is work enough, but doing two (in addition to, at the moment, and for the foreseeable future, having to hold-done a “9-to-5″ wage-slave gig to boot) is really spreading me a bit then; and, in all candor & honesty, my areas/subjects with which I have some proficiency, much less expertise, only extend so far. Many of you are beyond me in some (if not, indeed, many) areas: Certainly this is true of Tom (McCabe), Nato (Welch), IanC, and Brian Wang, to name merely those who come immediately to mind in this regard, and so I oftentimes can only resonate with their (at least usually) superb (or at least cogent) comments. To paraphrase the late great F.A. Hayek: I’m just an unrepentent old transhumanist, Singularitarian, liberal-eudaimonist. And, like the late great F.M. Esfandiary: I have a fond nostalgia for the FUTURE.
Hey, btw, Tom (McCabe): I hear tell you have your own new blog too. Did I miss the address? Have you listed the address here? I’m sure a blog by you, Tom, would be, like Michael’s here (and indeed IanC’s for that matter), a good one in terms of not only raising awareness but also in terms of allowing-for and promoting a robust, vigorous debate/discussion on cutting-edge issues. I wish you the very best in terms of “traffic”…
Ciao for now…
June 21st, 2007 at 2:30 pm
MCP2012: Tom’’s Blog. He still has the wrong url set up for his comments here. (Hint, Tom! ;))
June 21st, 2007 at 2:30 pm
Gyah. Link was bad. Tom’s Blog
June 21st, 2007 at 2:45 pm
THANKS, yet AGAIN, Ian!! You ROCK, kid-o!!
And, yeah, TOM: Fix your comments section (if it still needs fixin’…)
June 21st, 2007 at 5:45 pm
“MCP2012: Tom’’s Blog. He still has the wrong url set up for his comments here. (Hint, Tom! ;))”
The link works for me.
June 22nd, 2007 at 12:52 pm
Today is the first time it’s worked for me; the other URL still does, though.
Weird.
January 14th, 2008 at 4:43 pm
Tramadol….
Cheap tramadol….