More on Kurzweil's Predictions

After I wrote my last blog post reacting to Alex Knapp's critique of Ray Kurzweil's predictive accuracy, Ray Kurzweil wrote his own rebuttal of Alex's argument.

Ray then emailed me, thanking me for my defense of his predictions, but questioning my criticism of his penchant for focusing on precise predictions about future technology. I'm copying my reply to Ray here, as it may be of general interest...


Hi Ray,


I wrote that blog post in a hurry and in hindsight wish I had framed things more carefully there....  But of course, it was just a personal blog post not a journalistic article, and in that context a bit of sloppiness is OK I guess...


Whether YOU should emphasize precise predictions less is a complex question, and I don't have a clear idea about that.  As a maverick myself, I don't like telling others what to do!  You're passionate about predictions and pretty good at making them, so maybe making predictions is what you should do ;-) ....  And you've been wonderfully successful at publicizing the Singularity idea, so obviously there's something major that's right about your approach, in terms of appealing to the mass human psyche.


 I do have a clear feeling that the making of temporally precise predictions should play a smaller role in discussion of the Singularity than it now does.   But this outcome might be better achieved via the emergence of additional, vocal Singularity pundits alongside you, with approaches complementing your prediction-based approach -- rather than via you toning down your emphasis on precise prediction, which after all is what comes naturally to you...


One thing that worries me about your precise predictions is that in some cases they  may serve to slow progress down.  For example, you predict human-level AGI around 2029 -- and to the extent that your views are influential, this may dissuade investors from funding AGI projects now ... because it seems too far away!  Whereas if potential AGI investors more fully embraced the uncertainty in the timeline to human-level AGI, they might be more eager for current investment.


Thinking more about the nature of your predictions ... one thing that these discussions of your predictive accuracy highlights is that the assessment of partial fulfillment of a prediction is extremely qualitative.  For instance, consider a prediction like “The majority of text is created using continuous speech recognition.”   You rate this as partially correct, because of voice recognition on smartphones.  Alex Knapp rates this as "not even close."   But really -- what percentage of text do you think is created using continuous speech recognition, right now?  If we count on a per character basis, I'm sure it's well below 1%.  So on a mathematical basis, it's hard to justify "1%" as a partially correct estimate of ">50%.   Yet in some sense, your prediction *is* qualitatively partially correct.  If the prediction had been "Significant subsets of text production will be conducted using continuous speech recognition", then the prediction would have to be judged valid or almost valid.


One problem with counting partial fulfillment of predictions, and not specifying the criteria for partial fulfillment, is that assessment of predictive accuracy then becomes very theory-dependent.  Your assessment of your accuracy is driven by your theoretical view, and Alex Knapp's is driven by his own theoretical view. 


Another problem with partial fulfillment is that the criteria for it, are usually determined *after the fact*.   To the extent that one is attempting scientific prediction rather than qualitative, evocative prediction, it would be better to rigorously specify the criteria for partial fulfillment, at least to some degree, in advance, along with the predictions.


So all in all, if one allows partial fulfillment, then precise predictions become not much different from highly imprecise, explicitly hand-wavy predictions.   Once one allows partial matching via criteria defined subjectively on the fly, “The majority of text will be created using continuous speech recognition in 2009” becomes not that different from just saying something qualitative like "In the next decade or so, continuous speech recognition will become a lot more prevalent."  So precise predictions with undefined partial matching, are basically just a precise-looking way of making rough qualitative predictions ;)


If one wishes to avoid this problem, my suggestion is to explicitly supply more precise criteria for partial fulfillment along with each prediction.  Of course this shouldn't be done in the body of a book, because it would make the book boring.  But it could be offered in endnotes or online supplementary material.  Obviously this would not eliminate the theory-dependence of partial fulfillment assessment -- but it might diminish it considerably.


For example the prediction “The majority of text is created using continuous speech recognition.” could have been accompanied with information such as "I will consider this prediction strongly partially validated if, for example, more than 25% of the text produced in some population comprising more than 25% of people is produced by continuous speech recognition; or if more than 25% of text in some socially important text production domain is produced by continuous speech recognition."   This would make assessment of the prediction's partial match to current reality a lot easier.


I'm very clear on the value of qualitative predictions like "In the next decade or so, continuous speech recognition will become a lot more prevalent."  I'm much less clear on the value of trying to make predictions more precisely than this.   But maybe most of your readers actually, implicitly interpret your precise predictions as qualitative predictions... in which case the precise/qualitative distinction is largely stylistic rather than substantive


Hmmm...


Interesting stuff to think about ;)
ben

Ray Kurzweil's (Sometimes) Wrong Predictions

Note: there was a followup blog post to this one, presenting some complementary views that I also hold, and linking to some more recent comments by Ray Kurzweil on the  matter.


Forbes blogger Alex Knapp, who often covers advanced technology and futurist topics, recently wrote a post titled Ray Kurzweil's Predictions for 2009 Were Mostly Inaccurate ...

Some of Knapp's posts are annoyingly opinionated and closed-minded, but this one was well-put together, and I made a lengthy comment there, which I repeat here.  You should read his post first to get the context...

And also, once you read his post, you might want to read Ray's rebuttal to Michael Anissimov's earlier critique of his predictions. 

Ray rates himself as 90% right out of 100+ predictions; Michael looks at only a handful of Ray's predictions and finds most of them unfulfilled.

Looking at the "90% right" that Ray claims, it seems to me about half of these are strong wins, and the other half are places where the technologies Ray has forecast DO now exist, but aren't as good or as prevalent as he had envisioned.

On the other hand, Alex Knapp in Forbes took Ray's top 10 predictions rather than the full 100+, and found a lower accuracy for these.

An excerpt from my comment to Alex's post on the Forbes site (with light edits) is:


Alex,
 
...

One thing that should be clarified for the general readership is that the vast majority of those of us in the "Singularitarian" community do not, and never did, buy into all of Ray Kurzweil's temporally-specific predictions.  We love Ray dearly and respect him immensely -- and I think the world owes Ray a great debt for all he's done, not only as an inventor, but to bring the world's attention to the Singularity and related themes.  However, nearly all of us who believe a technological Singularity is a likely event this century, prefer to shy away from the extreme specificity of Ray's predictions.

Predicting a Singularity in 2045 makes headlines, and is evocative.  Predicting exactly which technologies will succeed by 2009 or 2019 makes headlines, and is evocative.  But most Singularitarians understand that predictions with this level of predictions aren't plausible to make.

The main problem with specific technology forecasts, is highlighted by thinking about multiple kinds of predictions one could make in reference to any technology X:

1) How long would it take to develop X if a number of moderately large, well-organized, well-funded teams of really smart people were working on it continuously?

2) How long would it take to develop X if a large, well-funded, bloated, inefficient government or corporate bureaucracy were working on it continuously?

3) How long would it take to develop X if there were almost no $$ put into the development of X, so X had to be developed by ragtag groups of mavericks working largely in their spare time?

4) How long would it take to develop X if a handful of well-run but closed-minded large companies dominated the X industry with moderately-functional tools, making it nearly impossible to get funding for alternate, radical approaches to X with more medium-term potential

When thinking about the future of a technology one loves or wants, it's easy to fall into making predictions based on Case 1.  But in reality what we often have is Case 2 or 3 or 4.

Predicting the future of a technology is not just about what is "on the horizon" in terms of science and technology, but also about how society will "choose" to handle that technology.   That's what's hard to predict.

For example a lot of Ray's failed top predictions had to do with speech technology.  As that is pretty close to my own research area, I can say pretty confidently that we COULD have had great text to speech technology by now.  But instead we've had Case 4 above -- a few large companies have dominated the market with mediocre HMM-based text to speech systems.  These work well enough that it's hard to make something better, using a deeper and more ultimately promising approach, without a couple years effort by a dedicated team of professionals.  But nobody wants to fund that couple years effort commercially, because the competition from HMM based systems seems too steep.  And it's not the kind of work that is effectively done in universities, as it requires a combination of engineering and research.

Medical research, unfortunately, is Case 2.  Pharma firms are commonly bloated and inefficient and shut off to new ideas, partly because of their co-dependent relationship with the FDA.  Radical new approaches to medicine have terrible trouble getting funded lately.  You can't get VC $$ for a new therapeutic approach until you've shown it to work in mouse trials or preferably human trials -- so how do you get the $$ to fund the research leading up to those trials?

Artificial General Intelligence, my  main research area, is of course Case 3.  There's essentially no direct funding for AGI on the planet, so we need to get AGI research done via getting funding for other sorts of  projects and cleverly working AGI into these projects....  A massive efficiency drain!!

If speech-to-text, longevity therapy or AGI had been worked on in the last 10 years with the efficiency that Apple put into building the iPad, or Google put into building its search and ad engines, then we'd be a heck of a lot further advanced on all three.

Ray's predictive methodology tries to incorporate all these social and funding related factors into its extrapolations, but ultimately that's too hard to do, because the time series being extrapolated aren't that long and depend on so many factors.

However, the failure of many of his specific predictions, does not remotely imply he got the big picture wrong.  Lots of things have developed faster than he or anyone thought they would in 2009, just as some developed more slowly.

To my mind, the broad scope of exponential technological acceleration is very clear and obvious, and predicting the specifics is futile and unnecessary -- except, say, for marketing purposes, or for trying to assess the viability of a particular business in a particular area.

The nearness of the Singularity does not depend on whether text-to-speech matures in 2009 or 2019 -- nor on whether AGI or longevity pills emerge in 2020 or 2040.  

To me, as a 45 year old guy, it matters a lot personally whether the Singularity happens in 2025, 2045 or 2095.  But in the grand scope of human history, it may not matter at all....

The overall scope and trend of technology development is harder to capsulize in sound bites and blog posts than specific predictions -- hence we have phenomena like Ray's book with its overly specific predictions, and your acute blog post refuting them. 

Anyway, anyone who is reading this and not familiar with the issues involved, I encourage you to read Ray's book the Singularity is Near -- and also Damien  Broderick's book "The Spike."  

Broderick's book made very similar points around a decade earlier, -- but it didn't get famous.  Why?  Because "Spike" sounds less funky than "Singularity", because the time wasn't quite ripe then, and because Broderick restricted himself to pointing out the very clear general trends rather than trying and failing to make overly precise predictions!

--
Ben Goertzel
http://goertzel.org


P.S. Regarding Ray's prediction that "“The neo-Luddite movement is growing.” -- I think that the influence of the Taliban possibly should push this into the "Prediction Met" or "Partially Met" category.  The prediction was wrong if restricted to the US, but scarily correct globally...

Will Corporations Prevent the Singularity?

It occurred to me yesterday that the world possesses some very powerful intelligent organisms that are directly and clearly opposed to the Singularity -- corporations.

Human beings are confused and confusing creatures -- we don't have very clear goal systems, and are quite willing and able to adapt our top-level goals to the circumstances.  I have little doubt that most humans will go with the flow as Singularity approaches.

But corporations are a different matter.  Corporations are entities/organisms unto themselves these days, with wills and cognitive structures quite distinct from the people that comprise them.   Public corporations have much clearer goal systems than humans: To maximize shareholder value.

And rather clearly, a Singularity is not a good way to maximize shareholder value.  It introduces way too much uncertainty.  Abolishing money and scarcity is not a good route to maximizing shareholder value -- and nor is abolishing shareholders via uploading them into radical transhuman forms!

So one can expect corporations -- as emergent, self-organizing, coherent minds of their own -- to act against the emergence of a true Singularity, and act in favor of some kind of future in which money and shareholding still has meaning.

Sure, corporations may adapt to the changes as Singularity approaches.  But my point is that corporations may be inherently less pliant than individual humans, because their goals are more precisely defined and less nebulous.  The relative inflexibility of large corporations is certainly well known.

Charles Stross, in his wonderful novel Accelerando, presents an alternate view, in which corporations themselves become superintelligent self-modifying systems -- and leave Earth to populate space-based computer systems where they communicate using sophisticated forms of auctioning.   This is not wholly implausible.   Yet my own intuition is that notions of money and economic exchange will become less relevant as intelligence exceeds the human level.  I suspect the importance of money and economic exchange is an artifact of the current domain of relative material scarcity in which we find ourselves, and that once advanced technology (nanotech, femtotech, etc.) radically diminishes material scarcity, the importance of economic thinking will drastically decrease.  So that far from becoming dominant as in Accelerando, corporations will become increasingly irrelevant post-Singularity.  But if they are smart enough to foresee this, they will probably try to prevent it.

Ultimately corporations are composed of people (until AGI advances a lot more at any rate), so maybe this issue will be resolved as Singularity comes nearer, by people choosing to abandon corporations in favor of other structures guided by their ever-changing value systems.   But one can be sure that corporations will fight to stop this from happening.

One might expect large corporations to push hard for some variety of "AI Nanny" type scenario, in which truly radical change would be forestalled and their own existence persisted, as part of the AI Nanny's global bureaucratic infrastructure.  M&A with the AI Nanny may be seen as preferable to the utter uncertainty of Singularity.

The details are hard to foresee, but the interplay between individuals and corporations as Singularity approaches should be fascinating to watch.

Are Prediction and Reward Relevant to Superintelligences?

In response to some conversation on an AGI mailing list today, I started musing about the relationship between prediction, reward and intelligence.

Obviously, in everyday human and animal life, there's a fairly close relationship between prediction, reward and intelligence.  Many intelligent acts boil down to predicting the future; and smarter people tend to be better at prediction.  And much of life is about seeking rewards of one kind or another.  To the extent that intelligence is about choosing actions that are likely to achieve one's goals given one's current context, prediction and reward are extremely useful for intelligence.

But some mathematics-based interpretations of "intelligence" extend the relation between intelligence and prediction/reward far beyond  human and animal life.  This is something that I question.

Solomonoff induction is a mathematical theory of agents that predict the future of a computational system at least as well as any other possible computational agents.  Hutter's "Universal AI" theory is a mathematical theory of agents that achieve (computably predictable) reward at least as well as any other possible computational agents acting in a computable environment.   Shane Legg and Marcus Hutter have defined intelligence in these terms, essentially positing intelligence as generality of predictive power, or degree of approximation to the optimally predictive computational reward-seeking agent AIXI.   I have done some work in this direction as well, modifying Legg and Hutter's definition into something more realistic -- conceiving intelligence as (roughly speaking) the degree to which a system can be modeled as efficiently using its resources to help it achieve computably predictable rewards across some relevant probability distribution of computable environments.  Indeed, way back in 1993 before knowing about Marcus Hutter, I posited something similar to his approach to intelligence as part of my first book The Structure of Intelligence (though with much less mathematical rigor).

I think this general line of thinking about intelligence is useful, to an extent.  But I shrink back a bit from taking it as a general foundational understanding of intelligence.

It is becoming more and more common, in parts of the AGI community, to interpret these mathematical theories as positing that general intelligence, far above the human level, is well characterized in terms of prediction capability and reward maximization.  But this isn't very clear to me (which is the main point of this blog post).  To me this seems rather presumptuous regarding the nature of massively superhuman minds!

It may well be that, once one gets into domains of vastly greater than human intelligence, other concepts besides prediction and reward start to seem more relevant to intelligence, and prediction and reward start to seem less relevant.

Why might this be the case?

Regarding prediction: Consider the possibility that superintelligent minds might perceive time very differently than we do.  If superintelligent minds' experience goes beyond the sense of a linear flow of time, then maybe prediction becomes only semi-relevant to them.  Maybe other concepts we don't now know become more relevant.  So that thinking about superintelligent minds in terms of prediction may be a non-sequitur.

It's similarly quite quite unclear that it makes sense to model superintelligences in terms of reward.  One thinks about the "intelligent" ocean in Lem's Solaris.  Maybe a fixation on maximizing reward is an artifact of early-stage minds living in a primitive condition of material scarcity.

Matt Mahoney made the following relevant comment, regarding an earlier version of this post: "I can think of 3 existing examples of systems that already exceed the human brain in both knowledge and computing power: evolution, humanity, and the internet.  It does not seem to me that any of these can be modeled as reinforcement learners (except maybe evolution), or that their intelligence is related to prediction in any of them."

All these are  speculative thoughts, of course... but please bear in mind that the relation of Solomonoff induction and "Universal AI" to real-world general intelligence of any kind is also rather wildly speculative...  This stuff is beautiful math, but does it really have anything to do with real-world intelligence?  These theories have little to say about human intelligence, and they're not directly useful as foundations for building AGI systems (though, admittedly, a handful of scientists are working on "scaling them down" to make them realistic; so far this only works for very simple toy problems, and it's hard to see how to extend the approach broadly to yield anything near human-level AGI).  And it's not clear they will be applicable to future superintelligent minds either, as these minds may be best conceived using radically different concepts.

So by all means enjoy the nice math, but please take it with the appropriate fuzzy number of grains of salt ;-) ...

It's fun to think about various kinds of highly powerful hypothetical computational systems, and fun to speculate about the nature of incredibly smart superintelligences.  But fortunately it's not necessary to resolve these matters -- or even think about them much -- to design and build human-level AGI systems.

Free Will without Counterfactuals?

This brief post is an afterthought to the just-previous post about the nature of reality.

As a side point in that post, I observed that one can often replace counterfactuals with analogies, thus making things a bit clearer.

It occurred to me this morning as I lay in bed waking up, that one can apply this method to the feeling of free will.

I've previously written about the limitations of the "free will" concept, and made agreeable noises about the alternate concept of "natural autonomy." Here, however, my point is a slightly different (though related) one.

One of the key aspects of the feeling of free will is the notion "In situation S, if I had done X differently, then the consequences would have been different." This is one of the criteria that makes us feel like we've exercised free will in doing X.

Natural autonomy replaces this with, roughly speaking "If someone slightly different than me had done something slightly different than X, in a situation slightly different from X, then the result would likely have been different than when I did X in S." This is no longer a counterfactual, it's a probabilistic statement about actions and consequences drawn from an ensemble of actions and consequences done by various actors.

But perhaps that rephrasing doesn't quite get at the essence. It may be more to the point to say: "In future situations similar to S, if I do something that's not analogous to X, then something not analogous to what happened after S in situation X is likely to happen."

Or in cases of binary choice: "In future situations similar to S, if I do something analogous to Y instead of something analogous to X, then a consequence analogous to CY instead of a consequence analogous to CX is likely to occur."

This is really the crux of the matter, isn't it? Not hypothesizing about alternate pasts, nor choices from an ensemble of similar beings -- but rather, resolutions about what to do in the future.

In this view, an "act of will" is something like "an action in a situation, corresponding to specific predictions about which of one's actions will predictively imply which consequences in analogous future situations."

That's boring-sounding, but avoids confusing talk of possible worlds.

Mathematically, this is equivalent to a formulation in terms of counterfactuals ... but, counterfactuals seem to lead human minds in confusing directions, so using them as sparingly as possible seems like a good idea...

What Are These Things Called "Realities"?

Here follow some philosophical musings, pursued by my rambling mind one evening during the Xmas / New years interval.... I inflicted these ramblings on my kids for a while then finally decided to shut up and write them down....

The basic theme: What is this thing called "reality"? Or if you prefer a broader view: What are these things called realities??

After yakking a while, eventually I'll give a concrete and (I think) somewhat novel definition/characterization of "reality."

Real vs. Apparent

Where did this idea come from -- the "real" world versus the "apparent" world.

Nietzsche was quite insistent regarding this distinction -- in his view, there is only the apparent world, and talk of some other "real world" is a bunch of baloney. He lays this idea out quite clearly in The Twilight of the Idols, one of my favorite books.

There's certainly some truth to Nietzsche's perspective in this regard.

After all, in a sense, the idea of a "real world" is just another idea in the individual and collective mind -- just another notion that some people have made up as a consequence of their attempt to explain their sense perceptions and the patterns they detect therein.

But of course, the story told in the previous sentence is ALSO just another idea, another notion that some people made up … blah blah blah …

One question that emerges at this point is: Why did people bother to make up the idea of the "real world" at all … if there is only the apparent world?

Nietzsche, in The Twilight of the Idols, argues against Kant's philosophical theory of noumena (fundamentally real entities, not directly observable but underlying all the phenomena we observe). Kant viewed noumena as something that observed phenomena (the perceived, apparent world) can approximate, but never quite find or achieve -- a perplexing notion.

But really, to me, the puzzle isn't Kant's view of fundamental reality, it's the everyday commonsense view of a "real world" distinct from the apparent world. Kant dressed up this commonsense view in fancy language and expressed it with logical precision, and there may have been problems with how he did it (in spite of his brilliance) -- but, the real puzzle is the commonsense view underneath.

Mirages

To get to the bottom of the notion of "reality", think about the example of a mirage in the desert.

Consider a person wandering in the desert, hot and thirsty, heading south toward a lake that his GPS tells him is 10 miles ahead. But suppose he then sees a closer lake off to the right. He may then wonder: is that lake a mirage or not?

In a sense, it seems, this means he wonders: is that lake a real or apparent reality?

This concept of "reality" seems useful, not some sort of philosophical or mystical trickery.

The mirage seems real at the moment one sees it. But the problem is, once one walks to the mirage to drink the water in the mirage-lake, one finds one can't actually drink it! If one could feel one's thirst being quenched by drinking the mirage-water, then the mirage-water wouldn't be so bad. Unless of course, the quenching of one's thirst wasn't actually real… etc. etc.

The fundamental problem underlying the mirage is not what it does directly in the moment one sees it -- the fundamental problem is that it leads to prediction errors, which are revealed only in the future. Seeing the mirage leads one to predict one will find water in a certain direction -- but the water isn't there!

So then, in what sense does this make the mirage-lake "only apparent"? If one had not seen the mirage-lake, but had seen only desert in its place, then one would not have made the prediction error.

This leads to a rather mundane, but useful, pragmatic characterization of "reality": Something is real to a certain mind in a certain interval of time, to the extent that perceiving it leads that mind to make correct predictions about the mind's future reality.

Reality is a Property of Systems

Yeah, yeah, I know that characterization of reality is circular: it defines an entity as "real" if perceiving it tends to lead to correct predictions about "real" things.

But I think that circularity is correct and appropriate. It means that "reality" is a property attributable to systems of entities. There could be multiple systems of entities, constituting alternate realities A and B, so we could say

• an entity is real_A if perceiving it tends to lead to correct predictions about real_A things
• an entity is real_B if perceiving it tends to lead to correct predictions about real_B things

I think this is a nicer characterization of reality than Philip K. Dick's wonderful quote, "Reality is whatever doesn't go away when you stop believing in it."

The reason certain things don't go away when you stop believing in them, I suggest, is that the "you" which sometimes stops believing in something, is actually only a tiny aspect of the overall mind-network. Just because the reflective self stops believing in something, doesn't stop the "unconscious" mind from assuming that thing's existence, because it may be bound up in networks of implication and prediction with all sorts of other useful things (including in ways that the reflective self can't understand due to its own bandwidth limitations).

So, the mirage is not part of the same reality-system, the same reality, as the body which is thirsty and needs water. That's the problem with it -- from the body's perspective.

The body's relationship to thirst and its quenching is something that the reflective self associated with that body can't shake off -- because in the end that self is just one part of the overall mind-network associated with that body.

Counterfactuals and Analogies

After one has seen the mirage and wandered toward it through the desert and found nothing -- then one may think to oneself "Damn! If I had just seen the desert in that place, instead of that mirage-lake, I wouldn't have wasted my time and energy wandering through the desert to the mirage-lake."

This is a philosophically interesting thought, because what one is saying is that IF one had perceived something different in the past, THEN one would have made more accurate predictions after that point. One is positing a counterfactual, or put differently, one is imagining an alternate past.

This act of imagination, of envisioning a possible world, is one strategy that allows the mind to construct the idea of an alternate "real" world that is different from the "apparent" world. The key mental act, in this strategy, is the one that says: "I would have predicted better if, 30 minutes ago, I had perceived desert over there instead of (mirage-) lake!"

But in discussing this with my son Zar, who doesn't like counterfactuals, I quickly realized, one can do the same thing without counterfactuals. The envisioning of an alternate reality is unnecessary -- what's important is the resolution that: "I will be better off if, in future cases analogous to the past one where I saw a mirage-lake instead of the desert, I see the analogue of the desert rather than the analogue of the mirage-lake." This formulation in terms of analogues is logically equivalent to the previous formulation in terms of counterfactuals, but is a bit more pragmatic-looking, and avoids the potentially troublesome postulation of alternate possible worlds….

In general, if one desires more accurate prediction within a certain reality-system, one may then seek to avoid future situations similar to past ones in which one's remembered perceptions differ from related ones that would have been judged "real" by that system.

Realities: What and Why

This seems a different way of looking at real vs. apparent reality than the one Kant proposed and Nietzsche rejected. In the perspective, we have

• reality-systems -- i.e. systems of entities whose perception enables relatively accurate prediction of each other
• estimations that, in future situations analogous to one's past experiences, one will do better to take certain measures so as to nudge one's perceptions in the direction of greater harmony with the elements of some particular reality-system

So, the value of distinguishing "real" from "apparent" reality emerges from the value of having a distinguished system of classes of phenomena, that mutually allow relatively accurate prediction of each other. Relative to this system, individual phenomena may be judged more or less real. A mind inclined toward counterfactuals may judge something that was NOT perceived as more "real" than something that was perceived; but this complication may be avoided by worrying about adjusting one's perceptions in future analogues to past situations, rather than about counterfactual past possibilities.

Better Half-Assed than Wrong-Headed!

After I explained all the above ideas to my son Zar, his overall reaction was that it generally made sense but seemed a sort of half-assed theory of reality.

My reaction was: In a sense, yeah, but the only possible whole-assed approaches seem to involve outright assumption of some absolute reality, or else utter nihilism. Being "half assed" lets one avoid these extremes by associating reality with systems rather than individual entities.

An analogue (and more than that) is Imre Lakatos's theory of research programs in science, as I discussed in an earlier essay. Lakatos observed that, since the interpretation of a given scientific fact is always done in the context of some theory, and the interpretation of a scientific theory is always done in the context of some overall research program -- the only things in science one can really compare to each other in a broad sense are research programs themselves. Research programs are large networks of beliefs, not crisp statements of axioms nor lists of experimental results.

Belief systems guide science, they guide the mind, and they underly the only sensible conception of reality I can think of. I wrote about this a fair bit in Chaotic Logic, back in the early 1990s; but back then I didn't see the way reality is grounded in predictions, not nearly as clearly as I do now.

Ingesting is Believing?

In practical terms, the circular characterization of reality I've given above doesn't solve anything -- unless you're willing to assume something as preferentially more real than other things.

In the mirage case, "seeing is believing" is proved false because one gets to the mirage-lake, one can't actually drink any of that mirage-water. One thing this proves is that "ingesting is believing" would be a better maxim than "seeing is believing." Ultimately, as embodied creatures, we can't get much closer to an a priori assumptive reality than the feeling of ingesting something into our bodies (which is part of the reason, obviously, that sexual relations seem so profoundly and intensely real to us).

And in practice, we humans can't help assuming something as preferentially real -- as Phil Dick observes, some things, like the feeling of drinking water, don't go away even if we stop believing in them … which is because the network of beliefs to which they belong is bigger and stronger than the reflective self that owns the feeling of "choice" regarding what to believe or not. (The status of this feeling of choice being another big topic unto itself, which I've discussed before, e.g. in a chapter of the Cosmist Manifesto.).... This is the fundamental "human nature" with which Hume "solved" the problem of induction, way back when....

Now, what happens to these basic assumptions when we, say, upload our mind-patterns into robot bodies ... or replace our body parts incrementally with engineered alternatives ... so that (e.g.) ingesting is no longer believing? What happens is that our fundamental reality-systems will change. (Will a digital software mind feel like "self-reprogramming is believing"??1) Singularity-enabling technologies are going to dramatically change realities as we know them.

And so it goes…

My Goal as an AGI Researcher

In a recent thread on the AGI email list, Matt Mahoney pressed me regarding my high-level goals as an AGI researcher, and a leader of the OpenCog project. This blog post repeats my answer, as I posted it on that email list. This is familiar material to those who have followed my work and thinking, but maybe I've expressed things here slightly differently than in the past....

My goal as an AGI researcher is not precisely and rigorously defined. I'm OK with this. Building AGI is a human pursuit, and human pursuits aren't always precisely and rigorously defined. Nor are scientific pursuits. Often the precise, rigorous definitions come only after a lot of the research is done.

I'm not trying to emulate human beings or human minds in detail. But nor am I trying to make a grab-bag of narrow agents, without the capability to generalize automatically to new problems radically different from the ones for which they were originally designed. I am after a system that -- in the context of the scope of contemporary human activities -- possesses humanlike (or greater) capability to generalize its knowledge from one domain to other qualitatively different domains, and to learn new things in domains different than the ones its programmers had explicitly in mind. I'm OK if this system possesses many capabilities that a human doesn't.

There are probably many ways of achieving software with this kind of general intelligence. The way I think I understand (and am trying to realize with OpenCog), is to roughly emulate the process of human child development -- where I say roughly because I'm fine with the system having some capabilities beyond those of any human. Even if it does have some specialized superhuman capabilities from the start, I think this system will develop the ability to generalize its knowledge to qualitatively different domains in the rough manner and order that a human child does.

What will I do once I have a system that has a humanlike capability of cross-domain generalization (in the scope of contemporary human activities)? Firstly I will study it, and try to create a genuine theory of general intelligence. Second I will apply it to solve various practical problems, from service robotics to research in longevity and brain-computer interfacing etc. etc. There are many, many application areas where the ability to broadly generalize is of great value, alongside specialized intelligent capabilities.

At some point, I think this is very likely to lead to an AGI system with recursive self-improving capability (noting that this capability will be exercised in close coordination with the environment, including humans and the physical world, not in an isolation chamber). Before that point, I hope that we will have developed a science of general intelligence that lets us understand issues of AGI ethics and goal system stability much better than we do now.

Why Time Appears To Move Forwards

On a long drive to my mom's house earlier this weekend, my son Zar and I got into a long conversation about the nature of causality, which got me thinking about the old puzzle of where the feeling of the directionality of time comes from...

Where does the feeling that "time moves forward" come from?

It's interesting to look at this view from two sides -- from the reductionist approach, in terms of the grounding of minds in physical systems; and also the phenomenological approach, in which one takes subjective experience as primary.

Putting together these two perspectives, one arrives at the conclusion that the directionality of time, as perceived by a mind, has to do with: entropy increase in the mind's environment, and entropy decrease in the mind's "theater of decisive consciousness."

A Reductionist View of the Origin of the Directionality of Time

Microphysics, as we currently understand it, doesn't seem to have this. In both classical and quantum physics, there is no special difference between the forward and backward direction in time.

Julian Barbour, in his excellent book The End of Time, argues that the directionality of time is an artifact of psychology -- something added by the experiencing mind.

It's commonly observed that thermodynamics adds an arrow of time to physics. The increase of entropy described by the Second Law of Thermodynamics implies a directionality to time. And the Second Law has an intriguing observer-dependence to it. If one assumes a conservative dynamical system evolving according to classical mechanics, there is no entropy increase -- until one assumes a coarse-graining of the system's state space, in which case the underlying complex dynamics of the system will cause an information loss relative to that coarse-graining. The coarse-graining is a simple sort of "observer-dependence." For a detailed but nontechnical exposition of this view of entropy, see Michel Baranger's essay "Chaos, Complexity and Entropy."

In this view, an argument for the origin of the directionality of time is as follows: The mind divides the world into categories -- i.e. "coarse-graining" the set of possible states of the world -- and then, with respect to these categories, there emerges an information loss corresponding to one temporal direction, but not the other.

A Psychological View of the Origin of the Direction of Time

Next, what can we say about the origin of the directionality of time from the psychological, subjectivist, phenomenological perspective?

Subjectively, it seems that our perception of the directionality of time is largely rooted in our perception of causality. Confronted with a pool of semi-organized sensations, we perceive some as causally related to others, and then assign temporal precedence to the cause rather than the effect.

Now, grounding temporal direction in causation may seem to introduce more confusion than clarification, since there is no consensus understanding of causality. However, there are certain approaches to understanding causality in the philosophical literature, that happen to tie in fairly naturally with the reductionist approach to grounding the directionality of time given above, and bear particular consideration here for that reason. I'm thinking especially of the view of causality as "information transmission across mechanistic hierarchies," summarized nicely by Phyllis Illari in this paper.

If causality is viewed as the transmission of information from cause to effect via channels defined by "mechanistic hierarchies", then we may see the direction of time as having to do somehow with information flow. This is loosely similar to how Baranger sees entropy emerging from the dynamics of complex systems as perceived relative to the coarse-graining of state space. In both cases, we see the flow of time as associated with the dynamics of information. However, to see exactly what's going on here, we need to dig a bit.

(I don't necessarily buy the reduction of causality to information transmission. But I do think this captures an important, interesting, relevant aspect of causality.)

Another point made by Illari in the above-linked article is the relation between causality and production. However, I find it more compelling to link causality and action.

It seems to me that the paradigm case of causality, from a subjective, psychological point of view, is when one of our own actions results in some observable effect. Then we feel, intuitively, that our action caused the effect.

We then interpret other phenomena we observe as analogous to instances of our own enaction. So, when we see an ape push a rock off a cliff, we can imagine ourselves in the position of the ape pushing the rock, so we can feel that the ape caused the rock to fall. And the same thing when it's not an ape but, say, another rock that's rolling into the first rock and knocking it off the cliff.

In this hypothesis, then, the root of temporal directionality is cause, and the root of causation is our interpretation of our own actions -- specifically, the assumption that the relation between an action and its preconditions, is fundamentally conceptually different than the relation between an action and its results.

Another way to say this is: the carrying-out of an action is viewed as a paring-down of possibilities, via a choosing of one action among many possibilities. Thus, carrying-out of an action is viewed as a decrease of entropy.

So, psychologically: The directionality of time ensues from the decrease of entropy perceived as associated with enaction -- via means of analogical reasoning which propagates this perceived entropy decrease to various perceptions that are not direct enact ions, causing them to be labeled as causative.

Putting the (Reductionist and Subjectivist) Pieces Together

On the face of it, we seem to have a paradox here: physically, the directionality of time comes from entropy increase; but psychologically, it comes from entropy decrease.

However, there's not really any paradox at all. This is merely a relative of the observation that living systems habitually decrease their entropy, at the cost of increasing the entropy of their environments.

The directionality of time, from the perspective of a given mind, appears to ensue from a combination of

• entropy decrease in the foreground (the "acutely conscious", explicitly deciding mind -- the "global workspace")
• entropy increase in the background (the environment of the mind)

This is a somewhat short-term perspective: it explains the feeling of temporal directionality over a certain brief interval. But then a mind with an episodic memory and a desire for logical coherence, will naturally piece together the temporal directions of its various memories in a coherent way, forming a linear sequence of time pointing from the past up till the present. And this same mind will then naturally reason about the future by analogy to the past, thus mentally building up a subjective timeline pointing into the future. And so the subjective sense of a linear axis of time emerges -- not because it's the only way to look at the world, but because it naturally emerges from the dynamics of foreground/background information flow, together with the quest for logical coherence.

A New Approach to Computational Language Learning

I've been thinking about a new approach to computational language learning for a while, and finally found time to write it down -- see the 2 page document here.

Pursued on its own, this is a "narrow AI" approach, but it's also designed to be pursued in an AGI context, and integrated into an AGI system like OpenCog.

In very broad terms, these ideas are consistent with the integrative NLP approach I described in this 2008 conference paper. But the application of evolutionary learning is a new idea, which should allow a more learning-oriented integrative approach than the conference paper alluded to.

Refining and implementing these ideas would be a lot of work, probably the equivalent of a PhD thesis for a very good student.

Those with a pure "experiential learning" bent will not like the suggested approach much, because it involves making use of existing linguistic resources alongside experiential knowledge. However, there's no doubt that existing statistical and rule-based computational linguistics have made a lot of progress, in spite of not having achieved human-level linguistic performance. I think the outlined approach would be able to leverage this progress in a way that works for AGI and integrates well with experiential learning.

I also think it would be possible for an AGI system (e.g. OpenCog, or many other approaches) to learn language purely from perceptual experience. However, the possibility of such an approach, doesn't imply its optimality in practice, given the hardware, software and knowledge resources available to us right now.

A Mind-World Correspondence Principle

I had some more ideas, working toward a general theory of general intelligence, which I wrote in a paper posted online at Dynamical Psychology.

(Please note: it's fairly abstract theoretical/mathematical material, so if you're solely interested in current AGI engineering work, don't bother! The hope is that this theory will be able to help guide engineering work once it's further developed, but it's not at that stage yet. So for now my abstract mathematical AGI theory work and practical AGI engineering work are only loosely coupled.)

The crux of the paper is:

MIND-WORLD CORRESPONDENCE PRINCIPLE: For an organism with a reasonably high level of intelligence in a certain world, relative to a certain set of goals, the mind-world path transfer function is a goal-weighted approximate functor

To see what those terms mean and why it might be a useful notion, you'll have to read the paper.

A cruder expression of the same idea, with fewer special defined terms is:

MIND-WORLD CORRESPONDENCE-PRINCIPLE: For a mind to work intelligently toward certain goals in a certain world, there should be a nice mapping from goal-directed sequences of world-states into sequences of mind-states, where “nice” means that a world-state-sequence W composed of two parts W1 and W2, gets mapped into a mind-state-sequence M composed of two corresponding parts M1 and M2.

As noted toward the end of the paper, this principle gives us systematic way to approach questions like: Why do real-world minds seem to be full of hierarchical structures? The answer is probably that the real world is full of goal-relevant hierarchical structures. The Mind-World Correspondence Principle explains exactly why these hierarchical structures in the world have to be reflected by hierarchical structures in the mind of any system that's intelligent in the world.

As an aside, it also occurred to me that these ideas might give us a nice way to formalize the notion of a "good mind upload," in category-theoretic terms.

I.e., if we characterize minds via transition graphs in the way done in the paper, then we can argue that mind X is a valid upload of mind Y if there is a fairly accurate approximate functor from X's transition graph to Y's.

And, if Y is a nondestructive upload (so X still exists after the uploading), it would remain a good upload of X over time if, as X and Y both changed, there was a natural transformation governing the functors between them. Of course, your upload might not WANT to remain aligned with you in this manner, but that's a different issue...

Creating/Discovering New States of Mind

Just some quasi-random musings that went through my head yesterday…

Our society puts a fair bit of energy, these days, into creating new technologies and discovering new scientific facts.

But we don’t put hardly any effort at all into creating/discovering new states of mind.

I think maybe we should – and at the end of this odd, long, rambling blog post I’m going to suggest a specific type of new mind-state that I think is well worth trying to create/discover: one synthesizing spiritual mindfulness and intense scientific creativity.

On Old and New States of Consciousness

First, bear with me while I spend a few paragraphs framing the issue…

When I read Stcherbatsky’s book Buddhist Logic years ago, I was struck by the careful analysis of 128 states of consciousness. Allan Combs’ book The Radiance of Being provides a simpler, smaller conceptual analysis of states of consciousness, with similar foundations. These and other similar endeavors are very worthy – but how can we really know that the scope of all possible varieties of human consciousness-state has been thoroughly explored?

All sorts of amazing new states of consciousness will become possible once the human brain has been enhanced with technology – brain-computer interfacing, genetic engineering, mind uploading, etc. Advanced AGI systems may enjoy states of consciousness far beyond human comprehension. However, it seems quite possible that ordinary human brains may be capable of many states of consciousness not yet explored.

The individual human mind is not all that individual – so the states of mind accessible to an individual may depend to some extent on the culture in which they exist. The catalogue of states of mind available in medieval India when Buddhist logic was invented, may include some states that are extremely hard for modern people to get into, and may omit some states of which modern people are capable.

The Perceived Conflict Between Scientific and Spiritual Mind-States

I’ve often wondered whether there’s some intrinsic conflict between the states of mind labeled “spiritual enlightenment”, and the states of mind consistent with profound scientific discovery.

Great scientific creation often seems to involve a lot of struggle and persistence – along with long stretches of beautiful “flow” experience. Great scientific work seems to involve a lot of very hard thinking and analysis, whereas enlightenment is generally described as involving “stopping all thought.”

Personally, I find it a lot easier to be mindful (in the Zen sense) while walking through the park, washing the dishes, lying in bed, or building a chair -- than while analyzing genomic data, working out the details of a new AI algorithm, writing a novel, or debugging complex software code. Subjectively, this feels to me like it’s because being mindful requires a bit of mental effort at first – to actively pay attention to what my mind and body are doing. Once the effort is done, then mindfulness can flow along effortlessly for a while. But then I may drift away from it, and that little jump of effort is needed to become mindful again. This dynamic of mindfulness drifting and returning, or almost drifting but then not actually drifting after all, seems not to function properly when I’m doing something highly cognitively intensive. When I’m doing the highly intensive thing, I get deeply “into” the process, which puts me in a wonderful flow state for a while – but then when the flow state ends, I’m not necessarily in a quasi-enlightened mindful state. I may be elated, or I may be exhausted, or I may be frustrated. I can then try to be mindful of my elation, exhaustion or frustration – but this is then a moderately substantial effort; and definitely my degree of mindfulness is lower than if I hadn’t bothered to do the cognitively intensive thing.

Now, it might just be that I’m not a particularly enlightened guy. Indeed, I have never claimed to be! I do have my moments of spiritual purity and cosmic blissful wisdom and all that -- but then I also have some pretty boring routine moments, and also moments of being totally un-mindfully overcome with various kinds of positive or negative emotion. However, observing other humans around me, I note that the same dichotomy I feel in my mind occurs in the outside world. I know some enlightened minds, and I know some productive, brilliant artists and scientists – but I don’t know anyone in the intersection. Maybe someone of this nature does exist; but if they do, they’re an awfully rare bird.

You could argue that, since being a spiritual genius is rare and being a scientific genius is rare, it’s not surprising that few people lie in the intersection! But I’m not just talking about genius. I’m talking about passion. Who has true devoted passion for spiritual enlightenment, and also true devoted passion for doing revolutionary science? Most people I know, if they like either, pursue one as a central goal and the other as a sort of sideline.

I don’t particularly want to be this way myself – I’d like to pursue both simultaneously, without feeling any conflict between the two. But in practical life I do feel a conflict, and I tend to choose science and art most of the time. Yes, from the enlightened view, the dichotomy and the conflict are just constructs of my mind. And when I’m in certain states of mind, I feel that way – that dichotomy and all the rest feel bogus and mildly amusing. But when I’m in those states of mind, I’m not doing my best art or science! Similarly, thinking about playing the piano, it clear that my best music has been played in states of heightened emotion – not states of enlightened emptiness.

I think the difficulty of maintaining a mindful mind-state and scientifically intensely creative mind-state, is deeply tied with the conflict between modern scientific culture and some older cultures like those of ancient India or China, that were more spiritually focused. The enlightened master was one of the ideals of India and China; and the great scientist or artist is one of the ideals of the modern world. The differences in ideals reflect more thoroughgoing cultural differences.

You could say that both the great scientist and the enlightened master are exaggerations, and the right thing is to be more balanced – a little bit scientific, a little bit spiritual. Maybe, as someone said to me recently, an enlightened master is like an Arnold Schwarzenegger of the spirit – hyper-developed beyond what is natural or useful (except in contexts like the Mr. Universe contest where being at the extreme is useful in itself!). And maybe great super-scientists are unnecessarily and unhealthily obsessive, and science would progress OK without them, albeit a little more slowly. But something in me rebels against this kind of conclusion. Maybe it’s just that I’m an unbalanced individual – reeling back and forth endlessly between being excessively scientific and excessively spiritual, instead of remaining calmly in the middle where I belong -- but maybe there’s more to it than that.

A New Scientific/Spiritual Mind-State?

What if, instead of being frustrated at the apparent contradiction between the mind-states of spiritual enlightenment /mindfulness and intense scientific creativity, we took it as a multidimensional challenge: to create a new state of mind, synergizing both of these aspects?

The ancient Indians and Chinese didn’t include this sort of mind-state in their catalogue, but they didn’t have science or modern art … they had a very different culture.

Can we discover a new, intrinsically mindful way of doing science and art? Without sacrificing the intensity or the creativity?

What if we pursued the discovery/creation of new states of mind as avidly as we pursue the creation of new machines or chemical compounds? What if there were huge multinational organizations devoted to mind-state discovery, alongside our chemical and pharmaceutical and computer engineering firms?

Zum: A Thought-Experiment

To make the above idea a little more concrete, let’s imagine a specific social structure designed to produce a synergetically scientific-spiritual state of mind. Imagine an agile software development team – a group of software developers working closely together on a project – that was also, simultaneously, a “zendo” or “dojo” or whatever you want to call it … a group of people gathered together in the interest of their own enlightenment. That is, they were simultaneously trying to get stuff done together, and to help each other maintain a state of mindfulness and individual & collective spiritual awareness.

I can’t think of a good name for this kind of combination, so I’m going to call it a “Zum”, because that word currently has no English meaning, and it reminds me of Zen and scrum (the latter a term from agile software development), and I like the letter “Z.”

I have heard of a new type of Vipassana meditation, in which a group of people sit together and while they meditate, verbalize their feelings as they pass through – “cold”, “breathing”, “warm”, “stomach”, etc. One can imagine a Zum engaging in this kind of discussion at appropriate moments, in the midst of technical discussions or collaborative work. Would hearing others describe their state like this interrupt thought in an unacceptable way? Possibly. Or would people learn to flow with it, as I flow with the music I listen to as I work?

What would a Zum be like? Would it help to have a couple enlightened masters hanging around? – maybe sitting there and meditating, or playing ping pong? That would produce a rather different vibe than a usual software development lab!

The key ingredient of the Zum is the attitude and motivation of the individuals involved. They would need to be dedicated both to producing great software together, and to helping each other remain mindful and joyful as much as possible.

One thing that might come out of this is, simply, a kind of balance, where the team does reasonably good work and is also rather happy. This certainly wouldn’t be a disaster. Maybe they’d even be a bit more effective than an average team due to a diminished incidence of personality conflicts and fewer stress-induced errors.

Another possibility is that, if this sort of experiment were tried in a variety of different styles and places, eventually a new state of mind would evolve – one bypassing the dichotomy of spiritual mindfulness versus intensely creative science or art production.

Solo Zum?

But do we really need a Zum? Organizing groups of people in novel configurations involves considerable practical difficulty. Why not become a one-person Zum? Experiment with different ways of practicing intense scientific creation and mindfulness at the same time – maybe you’ll come up with something new. Try to describe your internal methodology so others can follow in your footsteps. This sort of experimentation is every bit as valid and important as scientific experimentation, or personal experimentation with smart drugs. The human brain is far more flexible than we normally realize, it’s hard to say what may be possible even without technological brain modification.

Heh... well I'm really not sure how much any of that means, but it was an amusing train of thought! Now, it's time to pick up my daughter from school, and then get back to work.... I will be trying to be as cosmically aware as possible while my work proceeds ;O ;-) ... and probably not succeeding all that well !! So it goes... bring on the brain chips please...

This blog post was written while repetitively listening to various versions of A Tear for Eddie by Ween. This one is perhaps my favorite, though the studio version is great too.

Musings on future technologies for cognitive enhancement

A former college classmate of my son's, researching a magazine article on cognitive enhancement, just emailed me asking my opinion on future technologies for cognitive enhancement.... Here's the reply I gave him -- not much new information for the well-educated transhumanist reader, but I figured I'd paste it here anyways...



Regarding technologies for cognitive enhancement, present and future..



Firstly, I am not an expert on nootropics, but I can remember seeing various studies indicating potential positive benefits for cognitive aging. The racetams and modafinil come to mind, among many others. Anecdotally I am aware of plenty of folks who say these improve cognitive function, including older folks, but I'm not up on the literature.



I also see a huge future for neural stem cell therapy, and you can find a substantial literature on that online, though I'm not an expert on it. The regulatory issues here become interesting -- I know a number of individuals operating stem cell therapy clinics in Asia and Latin America, that cater substantially to US clients. So far these aren't focusing on neural stem cell therapy but I think that's not far off. The US regulatory environment has become archaic and highly problematic. One can envision a future in which Americans routinely fly to foreign countries for neural stem cell therapy and other medical interventions aimed at maintaining or increasing their intelligence. And the ones who stay home won't be as smart. One hopes that as these technologies mature, the American regulatory infrastructure will eventually mature as well.



I have also heard rumor (from reliable sources) of a device under development by the Chinese government in Beijing, in collaboration with some Western scientists, going by the name of the "head brain instrument" (three Chinese characters). This device uses transcranial magnetic stimulation, and has the dual effects of increasing learning rate, and also increasing susceptibility to suggestion. Interesting. I read an article a few months ago about a different but related device being tested in Australia, using transcranial stimulation to increase creativity. This sort of research seems fascinating and promising. No doubt one could advance even faster and further in this direction using direct brain-computer interfacing, but no one has yet developed an inexpensive and safe method of installing a Neuromancer-style "cranial jack" in the brain, alas. I'm sure the cranial jack is coming, but it's hard to estimate exactly when.



In terms of ongoing and future research, I think that a combination of genomics, experimental evolution and artificial intelligence is fairly shortly going to lead us to a variety of therapies to improve cognitive performance throughout the human lifespan, as well as to extend the healthy human lifespan overall. I'm seeing this now in the work my bioinformatics firm Biomind is doing in collaboration with the biopharma firm Genescient Corp. Genescient has created a set of populations of long-lived fruit flies, which live over 4x as long as control flies, and also display enhanced cognitive capability throughout their lives, including late life. We've gathered gene expression and SNP data from these "superflies" and are using AI technology to analyze the data -- and the results are pretty exciting so far! We've discovered a large number of gene-combinations that are extremely strongly associated with both longevity and neural function, and many of these correspond to likely-looking life-extension and cognitive-enhancement pathways in the human organism. The supplement Stem Cell 100, now on the market, was inspired by this research; but that's just the start ... I think we're going to see a lot of new therapies emerge from this sort of research, including nutraceuticals, pharmaceuticals, gene therapy, and others.



I'm currently in San Francisco, where I just got finished with 4 days of the Artificial General Intelligence 2011 conference, which was held on Google's campus in Mountain View. Now I'm at the larger AAAI (Association for the Advancement of AI) conference in San Francisco. I think that AI research, as it matures, is going to have a huge effect on cognitive enhancement research among many other areas. Right now my own Biomind team and others are using AI to good effect in bioinformatics -- but the AI tools currently at our disposal are fairly narrow and specialized, albeit with the capability to see pattens that are inaccessible to either unassisted humans or traditional statistical algorithms. As AI gradually moves toward human-level artificial general intelligence, we're going to see a revolutionary impact upon all aspects of biomedical science. Already there's far more biomedical data online than any human mind can ingest or comprehend -- an appropriately constructed and instructed AGI system could make radical advances in cognitive enhancement, life extension and other areas of biomedicine, just based on the data already collected ... in addition to designing new experiments of its own.



Down the road a bit, there's the potential for interesting feedback effects to emerge regarding cognitive enhancement, conceivably resulting in rapid exponential growth. The better science and technology we have, the better cognitive enhancers we can create, and the smarter we get. But the smarter we get, the better the science and technology we can develop. Et cetera, and who knows where (or if) the cycle ends! We live in interesting times, and I suspect in the next few decades they will become dramatically *more* interesting....

Unraveling Modha & Singh's Map of the Macaque Monkey Brain

(... plus some semi-related AGI musings at the end!)

On July 27 2010, PNAS published a paper entitled "Network architecture of the long-distance pathways in the macaque brain" by Dharmendra Modha and Raghavendra Singh from IBM, which is briefly described here and available in full here. The highlight of the paper is a connectivity diagram of all the regions of the macaque (monkey) brain, reproduced in low res right here:



See here for a hi-res version.

The diagram portrays "a unique network incorporating 410 anatomical tracing studies of the macaque brain from the Collation of Connectivity data on the Macaque brain (CoCoMac) neuroinformatic database. Our network consists of 383 hierarchically organized regions spanning cortex, thalamus, and basal ganglia; models the presence of 6,602 directed long-distance connections; is three times larger than any previously derived brain network; and contains subnetworks corresponding to classic corticocortical, corticosubcortical, and subcortico-subcortical fiber systems."

However, I found that the diagram can be somewhat confusing to browse, if one wants to look at specific brain regions and what they connect to. So my Novamente LLC co-conspirator Eddie Monroe and I went back to the original data files, given in the online supplementary information for the paper, and used this to make a textual version of the information in the diagram, which you can find here.

Our goal in looking at this wiring diagram is as a guide to understanding the interactions between certain human brain regions we're studying (human and monkey brains being rather similar in many respects). But I think it's worth carefully perusing for anyone who's thinking about neuroscience from any aspect, and for anyone who's thinking about AGI from a brain-simulation perspective.

Semi-Related AGI Musings

Complexity such as that revealed in Modha and Singh's diagrams always comes to my mind when I read about someone's "brain inspired" AGI architecture -- say, Hierarchical Temporal Memory architectures (like Numenta or DeSTIN, etc.) that consist of a hierarchy of layers of nodes, passing information up and down in a manner vaguely reminiscent of visual or auditory cortex. Such architectures may be quite valuable and interesting, but each of them captures a teensy weensy fraction of the architectural and dynamical complexity in the brain. Each of the brain regions in Modha and Singh's diagram is its own separate story, with its own separate and important functions and structures and complex dynamics; and each one interacts with a host of others in specially configured ways, to achieve emergent intelligence. In my view, if one wants to make a brain-like AGI, one's going to need to emulate the sort of complexity that the actual brain has -- not just take some brain components (e.g. neurons) and roughly simulate them and wire the simulations together in some clever way; and not just emulate the architecture and dynamics of one little region of the brain and proclaim it to embody the universal principles of brain function.

And of course this is the reason I'm not pursuing brain-like AGI at the moment. If you pick 100 random links from Modha and Singh's diagram, and then search the neuroscience literature for information about the dynamical and informational interactions ensuing from that link, you'll find that in the majority of cases the extant knowledge is mighty sketchy. This is an indicator of how little we still know about the brain.

But can we still learn something from the brain, toward the goal of making loosely brain-inspired but non-brain-like AGI systems? Absolutely. I'm currently interested in understanding how the brain interfaces perceptual and conceptual knowledge -- but not with a goal of emulating how the brain works in any detailed sense (e.g. my AGI approach involves no formal neurons or other elementary brainlike components, and no modules similar in function to specific brain regions), rather just with a goal of seeing what interesting principles can be abstracted therefrom, that may be helpful in designing the interface between OpenCog and DeSTIN (a hierarchical temporal memory designed by Itamar Arel, that we're intending to use for OpenCog's sensorimotor processing).

And so it goes... ;-)

Why is evaluating partial progress toward human-level AGI so hard?

This post co-authored by Ben Goertzel and Jared Wigmore

Here we sketch a possible explanation for the well-known difficulty of measuring intermediate progress toward human-level AGI is provided, via extending the notion of cognitive synergy to a more refined notion of ”tricky cognitive synergy.”

The Puzzle: Why Is It So Hard to Measure Partial Progress Toward Human-Level AGI?

A recurrent difficulty in the AGI field is the difficulty of creating a good test for intermediate progress toward the goal of human-level AGI.

It’s not entirely straightforward to create tests to measure the final achievement of human-level AGI, but there are some fairly obvious candidates here. There’s the Turing Test (fooling judges into believing you’re human, in a text chat) the video Turing Test, the Robot College Student test (passing university, via being judged exactly the same way a human student would), etc. There’s certainly no agreement on which is the most meaningful such goal to strive for, but there’s broad agreement that a number of goals of this nature basically make sense.

On the other hand, how does one measure whether one is, say, 50 percent of the way to human-level AGI? Or, say, 75 or 25 percent?

It’s possible to pose many ”practical tests” of incremental progress toward human-level AGI, with the property that IF a proto-AGI system passes the test using a certain sort of architecture and/or dynamics, then this implies a certain amount of progress toward human-level AGI based on particular theoretical assumptions about AGI. However, in each case of such a practical test, it seems intuitively likely to a significant percentage of AGI researcher that there is some way to ”game” the test via designing a system specifically oriented toward passing that test, and which doesn’t constitute dramatic progress toward AGI.

Some examples of practical tests of this nature would be

• The Wozniak ”coffee test”: go into an average American house and figure out how to make coffee, including identifying the coffee machine, figuring out what the buttons do, finding the coffee in the cabinet, etc.
• Story understanding – reading a story, or watching it on video, and then answering questions about what happened (including questions at various levels of abstraction)
• Passing the elementary school reading curriculum (which involves reading and answering questions about some picture books as well as purely textual ones)
• Learning to play an arbitrary video game based on experience only, or based on experience plus reading instructions

One interesting point about tests like this is that each of them seems to some AGI researchers to encapsulate the crux of the AGI problem, and be unsolvable by any system not far along the path to human-level AGI – yet seems to other AGI researchers, with different conceptual perspectives, to be something probably game-able by narrow-AI methods. And of course, given the current state of science, there’s no way to tell which of these practical tests really can be solved via a narrow-AI approach, except by having a lot of people try really hard over a long period of time.

A question raised by these observations is whether there is some fundamental reason why it’s hard to make an objective, theory-independent measure of intermediate progress toward advanced AGI. Is it just that we haven’t been smart enough to figure out the right test – or is there some conceptual reason why the very notion of such a test is problematic?

We don’t claim to know for sure – but in this brief note we’ll outline one possible reason why the latter might be the case.

Is General Intelligence Tricky?

The crux of our proposed explanation has to do with the sensitive dependence of the behavior of many complex systems on the particulars of their construction. Often-times, changing a seemingly small aspect of a system’s underlying structures or dynamics can dramatically affect the resulting high-level behaviors. Lacking a recognized technical term to use here, we will refer to any high-level emergent system property whose existence depends sensitively on the particulars of the underlying system as tricky. Formulating the notion of trickiness in a mathematically precise way is a worthwhile pursuit, but this is a qualitative essay so we won’t go that direction here.

Thus, the crux of our explanation of the difficulty of creating good tests for incremental progress toward AGI is the hypothesis that general intelligence, under limited computational resources, is tricky.

Now, there are many reasons that general intelligence might be tricky in the sense we’ve defined here, and we won’t try to cover all of them here. Rather, we’ll focus on one particular phenomenon that we feel contributes a significant degree of trickiness to general intelligence.

Is Cognitive Synergy Tricky?

One of the trickier aspects of general intelligence under limited resources, we suggest, is the phenomenon of cognitive synergy.

The cognitive synergy hypothesis, in its simplest form, states that human-level AGI intrinsically depends on the synergetic interaction of multiple components (for instance, as in the OpenCog design, multiple memory systems each supplied with its own learning process). In this hypothesis, for instance, it might be that there are 10 critical components required for a human-level AGI system. Having all 10 of them in place results in human-level AGI, but having only 8 of them in place results in having a dramatically impaired system – and maybe having only 6 or 7 of them in place results in a system that can hardly do anything at all.

Of course, the reality is almost surely not as strict as the simplified example in the above paragraph suggests. No AGI theorist has really posited a list of 10 crisply-defined subsystems and claimed them necessary and sufficient for AGI. We suspect there are many different routes to AGI, involving integration of different sorts of subsystems. However, if the cognitive synergy hypothesis is correct, then human-level AGI behaves roughly like the simplistic example in the prior paragraph suggests. Perhaps instead of using the 10 components, you could achieve human-level AGI with 7 components, but having only 5 of these 7 would yield drastically impaired functionality – etc. Or the same phenomenon could be articulated in the context of systems without any distinguishable component parts, but only continuously varying underlying quantities. To mathematically formalize the cognitive synergy hypothesis in a general way becomes complex, but here we’re only aiming for a qualitative argument. So for illustrative purposes, we’ll stick with the ”10 components” example, just for communicative simplicity.

Next, let’s suppose that for any given task, there are ways to achieve this task using a system that is much simpler than any subset of size 6 drawn from the set of 10 components needed for human-level AGI, but works much better for the task than this subset of 6 components(assuming the latter are used as a set of only 6 components, without the other 4 components).

Note that this supposition is a good bit stronger than mere cognitive synergy. For lack of a better name, we’ll call it tricky cognitive synergy. The tricky cognitive synergy hypothesis would be true if, for example, the following possibilities were true:

• creating components to serve as parts of a synergetic AGI is harder than creating components intended to serve as parts of simpler AI systems without synergetic dynamics
• components capable of serving as parts of a synergetic AGI are necessarily more complicated than components intended to serve as parts of simpler AGI systems.

These certainly seem reasonable possibilities, since to serve as a component of a synergetic AGI system, a component must have the internal flexibility to usefully handle interactions with a lot of other components as well as to solve the problems that come its way. In terms of our concrete work on the OpenCog integrative proto-AGI system, these possibilities ring true, in the sense that tailoring an AI process for tight integration with other AI processes within OpenCog, tends to require more work than preparing a conceptually similar AI process for use on its own or in a more task-specific narrow AI system.

It seems fairly obvious that, if tricky cognitive synergy really holds up as a property of human-level general intelligence, the difficulty of formulating tests for intermediate progress toward human-level AGI follows as a consequence. Because, according to the tricky cognitive synergy hypothesis, any test is going to be more easily solved by some simpler narrow AI process than by a partially complete human-level AGI system.

Conclusion

We haven’t proved anything here, only made some qualitative arguments. However, these arguments do seem to give a plausible explanation for the empirical observation that positing tests for intermediate progress toward human-level AGI is a very difficult prospect. If the theoretical notions sketched here are correct, then this difficulty is not due to incompetence or lack of imagination on the part of the AGI community, nor due to the primitive state of the AGI field, but is rather intrinsic to the subject matter. And if these notions are correct, then quite likely the future rigorous science of AGI will contain formal theorems echoing and improving the qualitative observations and conjectures we’ve made here.

If the ideas sketched here are true, then the practical consequence for AGI development is, very simply, that one shouldn’t worry all that much about producing compelling intermediary results. Just as 2/3 of a human brain may not be much use, similarly, 2/3 of an AGI system may not be much use. Lack of impressive intermediary results may not imply one is on a wrong development path; and comparison with narrow AI systems on specific tasks may be badly misleading as a gauge of incremental progress toward human-level AGI.

Hopefully it’s clear that the motivation behind the line of thinking presented here is a desire to understand the nature of general intelligence and its pursuit – not a desire to avoid testing our AGI software! Truly, as AGI engineers, we would love to have a sensible rigorous way to test our intermediary progress toward AGI, so as to be able to pose convincing arguments to skeptics, funding sources, potential collaborators and so forth -- as well as just for our own edification. We really, really like producing exciting intermediary results, on projects where that makes sense. Such results, when they come, are extremely informative and inspiring to the researchers as well as the rest of the world! Our motivation here is not a desire to avoid having the intermediate progress of our efforts measured, but rather a desire to explain the frustrating (but by now rather well-established) difficulty of creating such intermediate goals for human-level AGI in a meaningful way.

If we or someone else figures out a compelling way to measure partial progress toward AGI, we will celebrate the occasion. But it seems worth seriously considering the possibility that the difficulty in finding such a measure reflects fundamental properties of the subject matter – such as the trickiness of cognitive synergy and other aspects of general intelligence.

Is Software Improving Exponentially?

In a discussion on the AGI email discussion list recently, some folks were arguing that Moore's Law and associated exponential accelerations may be of limited value in pushing the world toward Singularity, because software is not advancing exponentially.

For instance Matt Mahoney pointed out "the roughly linear rate of progress in data compression as measured over the last 14 years on the Calgary corpus, http://www.mailcom.com/challenge/ "

Ray Kurzweil's qualitative argument in favor of the dramatic acceleration of software progress in recent decades is given in slides 104-111 of his presentation here.

I think software progress is harder to quantify than hardware progress, thus less often pointed to in arguments regarding technology acceleration.

However, qualitatively, there seems little doubt that the software tools available to the programmer have been improving damn dramatically....

Sheesh, compare game programming as I did it on the Atari 400 or Commodore 64 back in the 80s ... versus how it's done now, with so many amazing rendering libraries, 3D modeling engines, etc. etc. With the same amount of effort, today one can make incredibly more complex and advanced games.

Back then we had to code our own algorithms and data structures, now we have libraries like STL, so novice programmers can use advanced structures and algorithms without understanding them.

In general, the capability of programmers without deep technical knowledge or ability to create useful working code has increased *incredibly* in the last couple decades…. Programming used to be only for really hard-core math and science geeks, now it's a practical career possibility for a fairly large percentage of the population.

When I started using Haskell in the mid-90s it was a fun, wonderfully elegant toy language but not practical for real projects. Now its clever handling of concurrency makes it viable for large-scale projects... and I'm hoping in the next couple years it will become possible to use Haskell within OpenCog (Joel Pitt just made the modifications needed to enable OpenCog AI processes to be coded in Python as well as the usual C++).

I could go on a long time with similar examples, but the point should be clear. Software tools have improved dramatically in functionality and usability. The difficulty of quantifying this progress in a clean way doesn't mean it isn't there...

Another relevant point is that, due to the particular nature of software development, software productivity generally decreases for large teams. (This is why I wouldn't want an AGI team with more than, say, 20 people on it. 10-15 may be the optimal size for the core team of an AGI software project, with additional people for things like robotics hardware, simulation world engineering, software testing, etc.) However, the size of projects achievable by small teams has dramatically increased over time, due to the availability of powerful software libraries.

Thus, in the case of software (as in so many other cases), the gradual improvement of technology has led to qualitative increases in what is pragmatically possible (i.e. what is achievable via small teams), not just quantitative betterment of software that previously existed.

It's true that word processors and spreadsheets have not advanced exponentially (at least not with any dramatically interesting exponent), just as forks and chairs and automobiles have not. However, other varieties of software clearly have done so, for instance video gaming and scientific computation.

Regarding the latter two domains, just look at what one can do with Nvidia GPU hardware on a laptop now, compared to what was possible for similar cost just a decade ago! Right now, my colleague Michel Drenthe in Xiamen is doing CUDA-based vision processing on the Nvidia GPU in his laptop, using Itamar Arel's DeSTIN algorithm, with a goal toward providing OpenCog with intelligent visual perception -- this is directly relevant to AGI, and it's leveraging recent hardware advances coupled with recent software advances (CUDA and its nice libraries, which make SIMD parallel scientific computing reasonably tractable, within the grasp of a smart undergrad like Michel doing a 6 month internship). Coupled acceleration in hardware and software for parallel scientific computing is moving along, and this is quite relevant to AGI, whereas the relative stagnation in word processors and forks really doesn't matter.

Let us not forget that the exponential acceleration of various quantitative metrics (like Moore's Law) is not really the key point regarding Singularity, it's just an indicator of the underlying progress that is the key point.... While it's nice that progress in some areas is cleanly quantifiable, that doesn't necessarily mean these are the most important areas....

To really understand progress toward Singularity, one has to look at the specific technologies that most likely need to improve a lot to enable the Singularity. Word processing, not. Text compression, not really. Video games, no. Scientific computing, yes. Efficient, easily usable libraries containing complex algorithms and data structures, yes. Scalable functional programming, maybe. It seems to me that by and large the aspects of software whose accelerating progress would be really, really helpful to achieving AGI, are in fact accelerating dramatically.

In fact, I believe we could have a Singularity with no further hardware improvements, just via software improvements. This might dramatically increase the financial cost of the first AGIs, due to making them necessitate huge server farms ... which would impact the route to and the nature of the Singularity, but not prevent it.