Wealth of Papers from Recent Artificial General Intelligence Conference Held on Google Mountain View Campus

Brian Wang presents a nice overview of the recent 4th Conference on AGI, held in the heart of Silicon Valley. This year's AGI conference seems to represent an important evolution in much of the thinking in the AGI field, with a growing depth and sophistication of approach to the problems involved.

To get a better idea of what I am talking about, here are links to most of the papers presented at the conference.

And here are papers from a special workshop on "Self Programming in AGI Systems"

Videos from 3rd Conference on AGI

Artificial general intelligence of human level or higher, would be a radically disruptive technology to modern societies. Along with breakthroughs in scalable robotics, universal nano-assemblers, and a mastery of biological gene expression, a breakthrough in AGI would quickly overhaul most of the bases of modern economics and most other important foundations of everyday life in high tech societies.

Al Fin cognitive scientists have presented many criticisms to mainstream AI approaches -- particularly to the idea that human intelligence can be represented algorithmically. One of the papers presented at this year's AGI conference elaborates on this idea: "Real World Limits to Algorithmic Intelligence"

The biological basis of mathematical competencies is an interesting look by Aaron Sloman at the development of spatial and mathematical concepts in humans. (via Brian Wang) Sloman touches on the idea of the non-verbal or pre-verbal metaphor, an important key to understanding human learning and thought.

Overall, Al Fin cognitive scientists are pleased at the direction the AGI movement is taking, on the basis of the AGI-4 papers they have read, and on the topics covered generally.

There is no doubt a great deal of hidden treasure in the many papers provided at the conference links above. For those who find this sort of thing interesting, enjoy.

Brain from the Bottom Up: Spontaneous Birth of Synchrony in Small Neuronal Networks

More 13 July 2011: Brian Wang looks at the same research, with an emphasis on the hardware (electronic) aspect. It is fitting to look at both the neurons and the electronics, since the coming cybernetic biosingularity will be dependent upon both.

Human intelligence and consciousness are poorly understood, even by cognitive scientists, neuroscientists, and consciousness specialists. No one understands how to build a human intelligence from scratch, much less how to build a non-human intelligence capable of interacting with humans and the outside world on its own terms. But researchers at Tel Aviv University from the departments of Electrical Engineering and Physics, have taken a fascinating approach to building the basic components of brains: networks of biological neurons. Something wonderful happened when enough cultured neurons linked together in network: They spontaneously "synched up."

Background


Information processing in neuronal networks relies on the network's ability to generate temporal patterns of action potentials. Although the nature of neuronal network activity has been intensively investigated in the past several decades at the individual neuron level, the underlying principles of the collective network activity, such as the synchronization and coordination between neurons, are largely unknown. Here we focus on isolated neuronal clusters in culture and address the following simple, yet fundamental questions: What is the minimal number of cells needed to exhibit collective dynamics? What are the internal temporal characteristics of such dynamics and how do the temporal features of network activity alternate upon crossover from minimal networks to large networks?


Methodology/Principal Findings


We used network engineering techniques to induce self-organization of cultured networks into neuronal clusters of different sizes. We found that small clusters made of as few as 40 cells already exhibit spontaneous collective events characterized by innate synchronous network oscillations in the range of 25 to 100 Hz. The oscillation frequency of each network appeared to be independent of cluster size. The duration and rate of the network events scale with cluster size but converge to that of large uniform networks. Finally, the investigation of two coupled clusters revealed clear activity propagation with master/slave asymmetry.

Conclusions/Significance


The nature of the activity patterns observed in small networks, namely the consistent emergence of similar activity across networks of different size and morphology, suggests that neuronal clusters self-regulate their activity to sustain network bursts with internal oscillatory features. We therefore suggest that clusters of as few as tens of cells can serve as a minimal but sufficient functional network, capable of sustaining oscillatory activity. Interestingly, the frequencies of these oscillations are similar those observed in vivo. _PLoS

More papers by Mark Shein Idelson

Brain synchrony is an important topic of study, linked to consciousness, memory, learning, and normal function of general human brain activity. But synchronous oscillations are also programmed into the neurons themselves, at the smallest level of neuronal organisation. The challenge now, is to build "networks of networks", to discover the communications strategies which interconnected networks will evolve.

Contrast such a biological, bottom up approach with complex machine models of brain function such as the SpiNNaker project out of the University of Manchester, or the Human Brain Project (HBP) led by Henry Markram at Ecole Polytechnique de Lausanne.

Both of the above brain modeling approaches using computers, are based upon bottom-up theories of how brains work. The Lausanne project (HBP) is far more detailed -- going down to the ion channel level of neurons. The Manchester approach is impressive in its parallel computing ambitions, but it begins at the individual "neuronal spiking" level. SpiNNaker is more of a hybrid CompSci:Neurosci approach, than an actual model of the brain like the HBP.

Conventional artificial intelligence approaches do not mimic brain function closely, and are generally more "top-down" approaches, utilising conventional algorithmic concepts of mainstream computer science. Such approaches are doomed to failure before they even begin, as the last 70 years of conventional AI attempts continue to demonstrate.

In reality, brains must be grown. And new types of brains have to be evolved. Not necessarily from biological materials, but up until now the only working brains we know are biological. The first successful autonomous brains are likely to be evolved either from biological materials, or using ingenious abstractions of processes which emerge from biological mechanisms.

Al Fin cognitive scientists suggest that both the Lausanne approach and the Manchester approach are abstracted at the wrong level, to provide rapid paths to evolved intelligences. Creative human beings will have to discover the appropriate balance, but they will certainly be aided by computing systems in doing so. This is not gobbledygook nor is it AI-psychobabble. It is the genuine crux and pivot point of the problem.

What are the implications for the singularity? There will be no "uploading of consciousness" for the foreseeable future. The cyborg biosingularity is still on schedule for the decade between 2020 and 2030, if humans can avoid an extended Obama Dark Ages. The main question is how many of the cyborg components will be biological in origin, and how many will be non-biological (probably utilising nanotechnology).

Evolving Technium Landscapes of Mind

Just because we are conscious does not mean we have the smarts to make consciousness ourselves. Whether (or when) AI is possible will ultimately depend on whether we are smart enough to make something smarter than ourselves. We assume that ants have not achieved this level. We also assume that as smart as chimpanzees are, chimps are not smart enough to make a mind smarter than a chimp, and so have not reached this threshold either. While some people assume humans can create a mind smarter than a human mind, humans may be at a level of intelligence that is below that threshold also. We simply don't know where the threshold of bootstrapping intelligence is, nor where we are on this metric. _KevinKelly

Technium
Kevin Kelly has created a "Taxonomy of Minds" as a way of classifying different types of minds and what they might be able to do.

Precisely how a mind can be superior to our minds is very difficult to imagine. One way that would help us to imagine what greater intelligences would be like is to begin to create a taxonomy of the variety of minds. This matrix of minds would include animal minds, and machine minds, and possible minds, particularly transhuman minds, like the ones that science fiction writers have come up with.

Imagine we land on a alien planet. How would we describe or measure the level of the intelligences we encounter there -- assuming they are greater than ours? What are the thresholds of superior intelligence? What are the categories of intelligence in animals on earth? _Read the rest...TaxonomyofMinds

Technium
The actual development of superior minds is more likely to occur via evolutionary mechanisms, rather than from straightforward design from principle. The adaptive landscape graphic above provides a small portion of an evolutionary adaptive landscape. Creatures that achieve the higher peaks may be capable of achieving greater feats, but also may be more subject to extinction when the environment shifts -- or when the adaptive landscape is enlarged by merging with a previously separate adaptive landscape (building a bridge between islands, tunneling through a mountain chain, digging a canal through an isthmus, or the emergence of an intergalactic wormhole).

Rather than waiting until our minds become capable of creating other minds, it is more likely that humans will create an evolutionary landscape from which a more intelligent mind than human minds might emerge.

Recently, in conversations with George Dyson, I realized there is a fifth type of elementary mind:

5) A mind incapable of designing a greater mind, but capable of creating a platform upon which greater mind emerges.

This type of mind cannot figure out how to birth an intelligence equal to itself, but it does figure out how to set up conditions of evolution so that a new mind emerges from the forces pushing it. _Technium

This is the approach to AI which Al Fin cognitive scientists have been promoting and utilising. It would be fooling one's self to imagine that it will be easy to evolve a smarter mind. But at least it is not impossible, as most conventional approaches to AI are proving themselves to be. (conventional AI researchers are attempting quantitative solutions where qualitative solutions apply)

There is something quite amusing here: The human mind itself can flit among the taxonomy of minds, at any given time. Because of how the human brain evolved, and the paths we have taken in development, each one of us is multitudes. Without a doubt, we all need better training in using our minds.

More: An interesting set of links to sources which expect or assume the imminent creation of a super-human machine intelligence (and a consequent "singularity") and a few sources which are critical of such a "hard take-off" to singularity superintelligence

Al Fin is among the skeptics of the "techno-singularity" concept. Rather, Al Fin expects any near-term singularity to be of the "bio-singularity" variety.

Human Brain Project Moves Toward Human Cortex Model

Spiegel
Henry Markram's Human Brain Project in Lausanne, is competing for funding from the FET Flagship Initiative, to the tune of 1 billion Euros, disbursed over a ten year period. Markram's goals are extremely ambitious, and unprecedented. He aims to model the human cerebral cortex to an exquisite degree of precision. Markram expects that his model of the human brain will be so exact, that he will be able to study inaccessible brain diseases and devise impossible brain cures by using his model. He may be right. But in only ten years?

Scientists are paying particular attention to the cerebral cortex. This layer on the outside of brain, only a few millimeters thick, is the most important condition of it evolution. It is the starting point for efforts to understand what makes us tick -- and for endeavors to find solutions when things go wrong. Our brain builds its version of the universe in the cerebral cortex. The vast majority of what we see doesn't enter the brain through the eye. It is instead is based on the impressions, experiences and decisions in our brain.

Markham already completed important preparatory work for the computer modeling of the brain with his Blue Brain Project, an attempt to understand and model the molecular makeup of the mammalian brain. He modeled a tiny part of a rat brain, a so-called neocortical column, at the cell level. To understand what one of these columns does, it's helpful to imagine the cerebral cortex as a giant piano. There are millions of neocortical columns on the surface, and each of them produces a tone, in a manner of speaking. When they are simulated, the columns produce a symphony together. Understanding the design of these neocortical columns is a holy grail of sorts for neuroscientists.

It is important to understand the rules of communication among the nerve cells. The individual cells do not communicate at random, but instead seek specifically targeted communication partners. The axes of nerve cells intersect at millions of different points, where they can form a synapse. This makes communication between individual neurons possible. In a recent article in the journal Proceedings of the National Academy of Sciences, Markram writes that such connections are also developed entirely without external influence. This could indicate a sort of innate knowledge that all people have in common. Markram refers to it as the "Lego blocks" of the brain, noting that each person assembles his own world on the basis of this innate knowledge. _Spiegel

The object of study for the Human Brain Project may be the most complex dynamic system in the universe. The attempt would be impossible without the most sophisticated computing hardware and software available. And one must have more than a mere fistful of Euros to acquire such advanced goodies.

Modeling all of this in a computer is extremely complex. Markram's current model encompasses tens of thousands of neurons. But this isn't nearly enough to come within striking range of the secret of our brain. To do that, scientists will have to assemble countless other partial models, which are to be combined to create a functioning total simulation by 2023.

The supercomputers at the Jülich Research Center near Cologne are expected to play an important role in this process. The brain simulation will require an enormous volume of data, or what scientist Markram calls a "tsunami of data." One of the challenges for scientists working under Thomas Lippert, head of the Jülich Supercomputing Centre, is to figure out how to make the computer process only a certain part of the data at a given time, but without completely losing sight of the rest. They also have to develop an imaging method, such as large, three-dimensional holograms, to depict the massive amounts of data.

All it takes is a look at the work of Jülich neuroscientist Katrin Amunts to understand the sheer volume of information at hand. The team she heads is compiling a detailed atlas of the human brain. To do so, they cut a brain into 8,000 slices and digitized them with a high-performance scanner. The brain model generated in this way consists of cuboids, each measuring 10 by 10 by 20 micrometers, and the size of the data set is three terabytes. Brain atlases with higher resolutions, says Amunts, would probably consist of more than 700 terabytes _Spiegel

The answer to the question posed above is: No, this goal cannot be met within a time frame of ten years. Because the challenge is not merely quantitative -- a matter of compiling the precise assembly of terabytes to create a brain atlas. The goal is to create a dynamic, interactive model of incredible plasticity -- a model which changes itself moment to moment. The "700 terabyte" requirement mentioned above is just the starting point -- the bare beginning -- in the assembly of such a dynamic and ever-changing model.

But the problem is even harder -- much, much harder. The quantitative complexity -- even in dynamic flow -- is nothing when compared to the qualitative complexity, which is nowhere near to being solved by Markram's team.

The project as described in brief above is an excellent starting point. Much can be learned from such an approach. But starting points do not necessarily point directly toward the end that one seeks. Rather, they point somewhere "out there." It is for the questers to continuously adjust their headings -- and often they are forced to adjust their goals.

Good luck to Henry and his team -- with the funding and with the ongoing project. It is an ambitious goal worthy of any scientist.

Artificial Intelligence: Experts Admit that AI Stinks

Marvin Minsky, Patrick Winston, Noam Chomsky, and other thinkers, researchers, engineers, and scientists, think that artificial intelligence has gone wrong. And the way things have been going, it is not likely that AI researchers will get on "the right track" for some time -- since most of them cannot seem to understand the problem.

Several up and coming AI workers (plus Ray Kurzweil) have predicted that human-level AI will be achieved within the next 10 to 20 years. Where is the progress which they can point to, in order to justify these claims? Within their own imaginations, apparently. According to some of the thinkers and AI pioneers quoted at the link above, the problem of AI is not being approached correctly. One of the most piercing criticisms of the field seems to be that it is focusing on tactics rather than strategy.

...clearly, the AI problem is nowhere near being solved. Why? For the most part, the answer is simple: no one is really trying to solve it. This may come as a surprise to people outside the field. What have all those AI researchers been doing all these years? The reality is that they have largely given up on the grand ambitions of AI and are instead working on increasingly specialized subproblems: not just machine learning or natural-language understanding, say, but issues within those areas, like classifying objects or parsing sentences. _TechnologyReview

This is clearly true. What is tragic is that a large proportion of the current crop of top-level AI researchers do not seem to understand the difference between strategy and tactics, in the context of AI. Strategy in AI calls for much deeper level thinking than tactics, which may in fact be beyond the capacity of most researchers -- and even beyond the range of philosophers such as Dan Dennett, who has relatively low expectations for human-level AI in the foreseeable future.

As brilliant as earlier AI researchers a John McCarthy, Marvin Minsky, Seymour Papert may be (and may have once been) in their hay-day, the extent of the problem of human-level AI was too poorly defined for anyone to grasp the challenge.

Modern researchers who make boastful claims for the likelihood of achieving human-level AI in 10-20 years do not have that excuse. Clearly neither Kurzweil nor the other AI hopefuls truly have a grasp of the problem as a whole. And that failure will be their undoing.

What would it take to succeed at human-level AI? A closely-knit, multidisciplinary team of thinkers willing to try outlandish and counter-intuitive approaches to the problem -- well out of the mainstream. To achieve human-type machine intelligence, high level expertise in one's field is but the barest beginning, hardly a single step in the journey of a thousand miles. One must also have a child-like approach to the world, be both brilliant and incredibly humble in the face of the ineffable, and be able to integrate complex ideas from both within and without his own field of expertise into a new, working holism.

Of course it sounds like psycho-babble, but when approaching something far too complex for words, such failures of communication are inevitable. Al Fin cognitive engineers recommend that the concepts of "embodied cognition" and "preverbal metaphor" be kept foremost in in the minds of any hopeful AI developers.

For everyone else, don't get your hopes up too high. Higher education in the advanced world -- particularly in North America -- is moving into serious difficulty, which means that research facilities and funding are likely to be cut back, perhaps severely. The economies of the advanced nations, and the US in particular, are being badly mismanaged, which means that private sector efforts will also likely be cut back. The problem may require the emergence of an individual with the brilliance of Einstein, the persistence of Edison, and the wide-ranging coherent creativity of a da Vinci.

In other words, people need to become smarter -- at least some people. Then they can learn to evolve very smart machines. And perhaps to interface with networks of these very smart machines. Then, possibly, using this symbiotic intelligence to design even smarter people.

Because let's face it: Humans -- especially PC-cultured humans -- will only take us to the Idiocracy, sooner or later. And the Idiocracy will be not only stupid, but downright brutal.

Double Plus Overhype Ado About Artificial Synapse?

There's a news story replicating on the web right now about a "Functioning Synapse Created Using Carbon Nanotubes," for instance here and here.

....the circuit has not actually been constructed, so the "apparatus" photo there is kind of silly. It just gives the false impression that a synapse model was actually built physically with analog components.

...[ed: all that we have is] an electrical circuit schematic that in turn depends on certain SPICE models of carbon nanotube FETs (which have apparently been available since 2006). So in other words, this circuit is a particular model of a synapse being simulated with a simple circuit. _Science20

Samuel Kenyon points out at the Science20 article linked above, that the "artificial synapse" is only a simulated circuit using the SPICE electronic simulation program. But the overhype is doubly overdone, because even if the researchers had actually built a real, physical circuit that functioned as an "artificial synapse", it would still not put them any closer to actually building an "artificial brain."

PDF Source (via Science20) This overhyped excitement is reminiscent of IBM scientist Dharmendra Modha's claim that he had built a computer that was the equivalent of "a cat's brain." Initially, most science blogs (except Al Fin) accepted Modha's claim at face value. Then when Henry Markram came out publicly to refute the claim, Modha backed off and clarified, and almost everyone agreed in the end, that it was all pretty much ado about nothing.

It is the same thing here, where science and tech blogs initially rush to accept exaggerated claims in press releases. Then, little by little, sceptics step up and insist upon clarifications and qualifications of claims, until the claims are downgraded to the point that eventually no one can remember what the fuss was about.

In the case of "the artificial synapse", it is important to understand why this real world device is not even a synapse, much less a possible ingredient for an artificial brain.

The USC and Stanford researchers have designed a computer model of "an artificial synapse," not an actual artificial synapse. But even if it were a real synapse, would engineers be able to use it to assemble an artificial brain? No. And the reason why one thing does not naturally lead to the next is crucial to an understanding of how real brains work -- real brains being the only working proof of concept of intelligence in the known universe.

Artificial intelligence enthusiasts will rush to say that working brains do not necessarily have to work just like the bio-brains we know now. But then, what is the point of emulating a tiny component of a bio-brain in the first place, if you cannot use it to build a functioning brain, as we understand it? In other words, if your objective is to build a new class of brain, why start with a poor imitation of a low level component of a bio-brain? Why not start with something "better" from the get-go? [By "better", I mean faster, more versatile, etc etc]

Here is the reason: Because artificial intelligence researchers do not have a clue as to how to build an intelligent brain. And so they are practising a subtle form of cargo cult science.

It's okay. We all understand that rents and utilities must be paid, the price of gasoline is high, everything costs money. Academics must publish or perish, and getting research grants to build "artificial synapses" does sound kind of sexy. Anything to keep the lights on, right?

But all the same, it is important to understand that brains are not the plural of "synapse." It is time to stop pretending that one has made progress toward AI, when nothing of the sort has happened.

More: It is important to understand that the bulk of the exaggeration comes from press releases and media coverage. Here is the actual conclusion from the research study referred to:

A carbon nanotube synapse typical of cortical synapses has been designed and simulated using SPICE. While the simulations were successful, the design of a single typical synapse is only a small step along the path to a synthetic cortex. The variations in synapses, including inhibitory synapses, will be the focus of future research. Predicting the interconnection capabilities of nanotube circuits is also important in understanding the future prospects for a synthetic cortex. _PDFeve.usc.eduPDF

Unfortunately, as computer modelers try to more realistically model the events in the brain at cellular and molecular levels, computing power and computing time demands explode out of control very rapidly. More, the researchers above do not seem to understand the key facts of brain function upon which conscious intelligence is balanced: time-dependent high level cross-brain synchronisation (via evolved white matter pathways) of evolved multiple modular (grey matter) brain centers from brain stem to neocortex, dancing alongside sensory input, jostled by memory, under the changing lights of emotion, and swept up in hormonal tides and chaotic flows of molecules...

More complex than one imagines? More complex than one can possibly imagine. The job is simply too hard for intelligent design. Only evolution will do. We need to get better at intelligently designing evolution. ;-)

Playthings of the Gods

Technology Review
Leon Chua -- father of Amy Chua -- conceived the memristor in a paper back in 1971. The memristor is a resistor with a memory of an earlier state. It behaves differently, depending upon its history. Because inter-neuronal synapses typically also behave differently, depending upon their histories, the memristor is often seen as a building-block for creating more brain-like computers. Researchers are already simulating what a memristor-based computing system might look like:

Memristors are resistors that "remember" the state they were in, which changes according to the current passing through them. They are expected to revolutionise the design and capabilities of electronic circuits and may even make possible brain-like architectures in silicon, since neurons behave like memristors.

Today, we see one of the first revolutionary circuits thanks to Yuriy Pershin at the University of South Carolina and Massimiliano Di Ventra at the University of California, San Diego, two pioneers in this field. Their design is a memristor processor that solves mazes and it is remarkably simple.

...Pershin and Di Ventra begin by creating a kind of a universal maze in the form of a grid of memristors, in other words an array in which each node is connected to another by a memristor and a switch. This can be made to represent any regular maze by switching off certain connections within the array.

Solving this maze is then simple. Simply connect a voltage across the start and finish of the maze and wait. "The current flows only along those memristors that connect the entrance and exit points," say Pershin and Di Ventra. This changes the state of those memristors allowing them to be easily identified. The chain of these memristors is then the solution.

That's potentially much quicker than other maze solving strategies which effectively work in series. "The maze is solved in a massively parallel way, since all memristors in the network participate simultaneously in the calculation," they say. _TechnologyReview_via_NextBigFuture

One of the problems with asking a physicist, engineer, or computer scientist to devise a brain-like computer, is that persons trained strictly within these disciplines are not likely to know which elements of brain functioning should be "simplified" or "abstracted", and which elements should be closely copied.

The pursuit of artificial intelligence is rife with failed promises and predictions, over the past 60+ years. If we are not to go at least another 60 years without meaningful success, we will need researchers who are cross-trained in multiple disciplines relating to the problem.

The research described in the Technology Review article above was based upon the simulation of an array of memristors -- not on an actual memristor circuit. But even with real memristors, the circuit is simplistic in the extreme. The idea that one could assemble large numbers of simplified "synapses" into something that might behave like a biological brain -- in any meaningful way -- appears silly to anyone with even a basic understanding of how the brain works. And yet such silliness represents one of many parallel hopes for a so-far failed endeavour: artificial intelligence.

The synapse is not the basic unit of human intelligence or consciousness. The basic unit of human consciousness is something far less substantial and more ephemeral. It exists at multiple logical levels above the synaptic level. It is dependent upon the simultaneous function of trillions of synapses of distinctly multiple types, involving efferent, afferent, and re-entrant activity at multiple logical levels.

What the researchers describe in the Technology Review article is the simulation of a toy. Not the toy itself -- a simulation of the toy. The human brain is not a toy. Unless, of course, you are a god.

Human vs. Computer: How About a Real Challenge?

Update 18Feb2011: AI researcher Ben Goertzel presents some thoughts on Watsons ascendancy at H+Magazine online. H/T Brian Wang

Much has been made about the recently televised Jeopardy victory by IBM Watson, and earlier victories and impressive performances by specialised chess computers such as Blue Gene etc. Even in poker, computers are increasingly seen as threats to human dominance, thanks to clever human programmers. In the game world, Go is most often seen as a game where computers have not come close to experts.

What do these massive, power gobbling, ultra-pampered, spoon-fed, one-trick-pony game-playing super computers tell us about the human vs computer rivalry? Realistically, the rivalry is still human vs. human, with one team of humans utilising ultra-fast electronics devices to store, "analyse", and retrieve massive quantities of data, to gang up on a single human opponent.

Watson is certainly faster to the button than its human opponents, but we already knew that electrons were faster than nerves. Once the "natural language processing" trick was mastered, Watson could more readily lock out his opponents from responding -- even when they both clearly knew the answer.

But Watson could not drive itself home after the game, could not flush away its excretions (heat) by itself, could not feed itself, etc. In the end, Watson is a very expensive gimmick which served as a showcase for various specialised programming problems.

Perhaps if Watson could master all the games mentioned above, at once, and defeat experts in all of the games, it would be impressive as a game-player. But not really. Look at all the money, mass, and energy tied up in the junkpile called Watson. How would IBM make it more capable of playing multiple games? By throwing more mass, money, and energy into the already-huge junkpile. Not very clever, really, compared with the three pound human brain and all the things it can do -- including designing, building, repairing, and programming "smart" computers.

It all points out the fact that the state of artificial intelligence is pretty pathetic, all in all. Despite over 60 years of promises to create human-level intelligence "within 10 years", AI still stinks badly, and promises more of the same into the forseeable future.

The Jeopardy challenge -- like all similar challenges -- was a huge and expensive publicity hullabaloo. It is quite likely to damage the Jeopardy brand in the long run. It certainly puts forth an entirely false idea about the modern capability of computers, vis-a-vis humans, to reason and make decisions.

What would be a real challenge for Watson? How about a spontaneous, unplanned race over an extensive, lengthy, novel, 3-D obstacle course with ladders, walls, tunnels, slides, sand, and foot-deep water traps -- against a 5 year old human child?

Let's face it: Modern life requires humans to overtly or covertly (via proxies) partner with computers to achieve optimum performance in large areas of our lives. But what will it take to get computers to the point where they are consciously setting the agenda for humans, rather than the other way around?

It will take an entirely new "substrate of thought" than the high speed digital architectures currently used to such great -- if ultra-specialised -- effect. Worse, modern AI researchers for the most part have no idea what form such a new substrate would take. Certainly they do not understand the substrate for the only proof of concept of conscious intelligence which currently exists -- the human brain.

Too much like robots themselves, too many AI researchers unwittingly plod along artificial pathways leading to nowhere but diminutive local optima. Watson is only one illustration of the kludgy phenomenon.

What will it take, and how long will it take to discover it? There are limits to pure reason and speculation. Experimentation is necessary. Hands must be dirtied and hypotheses must be generated and tested. For the luggiest of lugheads out there, we need much better challenges than chess, Jeopardy -- or even Go -- to spur the effort required.

Memristor Brains? No, But Likely a Step in the Right Direction

IEEE
Brian Wang presents a fascinating glimpse at the next stage of attempted machine intelligence -- driven by DARPA grants. The approach will likely involve the use of the Chua memristor -- or similar nano-scaled electronic devices. DARPA has specified its requirements for its new family of scalable and adaptive electronic thinking systems, and it appears that the memristor family of devices may be the best approach for government contractors wishing to collect their fees.

Researchers have suspected for decades that real artificial intelligence can't be done on traditional hardware, with its rigid adherence to Boolean logic and vast separation between memory and processing. But that knowledge was of little use until about two years ago, when HP built a new class of electronic device called a memristor. Before the memristor, it would have been impossible to create something with the form factor of a brain, the low power requirements, and the instantaneous internal communications. Turns out that those three things are key to making anything that resembles the brain and thus can be trained and coaxed to behave like a brain. In this case, form is function, or more accurately, function is hopeless without form.

Basically, memristors are small enough, cheap enough, and efficient enough to fill the bill. Perhaps most important, they have key characteristics that resemble those of synapses. That's why they will be a crucial enabler of an artificial intelligence worthy of the term.

The entity bankrolling the research that will yield this new artificial intelligence is the U.S. Defense Advanced Research Projects Agency (DARPA). When work on the brain-inspired microprocessor is complete, MoNETA's first starring role will likely be in the U.S. military, standing in for irreplaceable humans in scout vehicles searching for roadside bombs or navigating hostile terrain. But we don't expect it to spend much time confined to a niche. Within five years, powerful, brainlike systems will run on cheap and widely available hardware. _IEEE

A step in the right direction? Yes. The memristor family of devices will allow for a nanoscale fabrication of devices which function very much like a inter-neuronal synapse. Creating massively parallel circuits with such devices will allow designers to produce some fascinating -- and possibly quite functional -- computing devices.

But will these devices work anything like the human (or animal) brain? Not anytime soon. Because the designers seem focused on one small, rudimentary aspect of the human brain -- the neuronal synapse -- it is unlikely that they will achieve the "bigger picture" view of how human brains actually work for a long, long time, and after many failures.

But the development of electronic devices which imitate the synapse more accurately will place the pursuit of the machine brain on an entirely different level, above and away from the diminutive local optima which previous AI researchers have been struggling to achieve.

What will it take for memristor family devices to approach human brain level of function? First, it will require the knowledge that the brain has many distinct types of neurons, which form many distinct types of synapses. Next, it will require the awareness that synapses are just the meager beginning of the spark of intelligence. It is actually a vast ensemble of synaptic actions occurring in precise ways at precise times, and affecting precise modular systems of processors, which makes animal-style consciousness and intelligence possible.

Then, it will require the insight that intelligence is "embodied," to start the research down a long, difficult, but final road toward the creation of a rudimentary working machine intelligence.

If you are thinking that there are other approaches to intelligence than the animal or human approach, Al Fin cognitive scientists respond, "of course." But where are these alternative approaches? Where are their proofs of concept, their working prototypes? No closer today, than in the late 1940s and 1950s when absolutely brilliant computer scientists first believed they were within easy reach.

Human level machine intelligence would create a radical revolution of human existence at many levels, in many ways. But such a development does not appear to be very close. Certainly, humans are not ready for it. But a lot of things happen which humans are not prepared to experience. Better start getting ready now.

More: Brain Inspired Computing by Versace (via Brian Wang)

Moneta Neuromorphics Laboratory (via Brian Wang)

The Limits of Intelligence; The Farce of Artificial Intelligence

Sato, et al

The only working model of human-level intelligence, as far as we know, is the human brain. We have no evidence of any higher form of intelligence anywhere in the universe. Yet scientists from widely varied areas of cognitive studies continue to make unlikely claims that they will achieve reverse-engineering of the human brain within 10 or 20 years. The problem with humans attempting to use machines to emulate intelligence, is that humans do not understand intelligence very well at all.

Recent progress in "memristor synapses" has given reverse-engineers of the brain hope, that they may finally be developing a hardware substrate that is better capable of emulating brain function. But even if that is true, how close do these developments place us to the goal of reverse-engineering a functioning human brain? Bluntly put, not close at all.

Scientists are slowly gaining an appreciation for how human memories are encoded -- within and by the hippocampus. For example, new memory formation requires the hippocampus to be able to produce new nerve cells of various types from stem cells. Some neuroscientists apparently feel that this understanding will help them to discover new "drug targets" for treating memory dysfunction, such as dementia. We should hope so, because dementia and brain atrophy of one form or another waits for virtually all of us -- if we live long enough.

But successful treatment of dementia does not help us to understand how our intelligence works -- except insofar as it provides tools for further research into the intricate mechanisms of human learning, memory, and creative imagination.

The encoding and decoding of human memories (more) has virtually nothing in common with what is generally thought of as "computation." Consequently the substrate of ordinary computation -- such as digital computers -- should not be seen as likely substrates for reverse engineering a human brain.

Human intelligence evolved over millions of years by natural selection, in the course of solving a variety of problems of survival. Human brains are not well evolved to solve the most pressing problems currently facing human societies. The average IQ for human populations is just below 90 points, and on a downward, dysgenic trajectory. Most humans are simply not intelligent enough to solve complex problems -- except those for which the human brain is evolved to solve. Most of the "big" problems of today do not fall within that category.

Even most humans with IQs in the 130 to 180 ++ range are generally not well suited to understand the basis for their own intelligence on any logical level -- much less most or all of them. If the potential to understand our own intelligence rests within the developing embryo and infant child, its critical window of development inevitably passes without the proper training. And so it goes, almost certainly, for a significant number of potential human abilities -- lost out of ignorance. But I digress.

Artificial intelligence research suffers from the lack of individuals with a special combination of trained aptitudes. Brilliant researchers abound in the disparate disciplines of computer science, neuroscience, cognitive psychology, linguistics, anthropology, philosophy, electrical engineering, and a wide array of creative, inventive, and speculative arts and sciences. But workers with the right combinations of skills and attitudes are extremely rare. The potential accomplishments of the uni-disciplinary approach to higher education evaporate very quickly when it comes to solving the extremely hard problems with which we are faced.

Solving the problem will require a different way of thinking about the problem. But that is a virtual impossibility for most people -- no matter how "intelligent."

Contemplate what may be involved in the efficient teaching and learning of "lateral thinking." The most rewarding known examples of lateral thinking occurred by accident. But de Bono claims to be able to teach the skill. It is virtually certain that such teaching is more effective if initiated during childhood -- and more effective in some children than in others.

Modern human knowledge is "full of holes", like a Sierpinski gasket. No matter how conscientiously we set about to fill in the holes, we only create more holes. Humans need to learn to relish this creation of holes, because the more holes we create, the more we have filled in. But the development of such a relishing of the fractal world of knowledge must likely begin in childhood.

Which brings us back to the creation and upbringing of children, their training and the societal milieu in which they are to be raised. We are botching the job rather badly at this time.

More on these topics later.

Beyond Kurzweil and Myers: A Useful Brain Emulation Viewpoint

George Dvorsky provides a measured and reasonable approach to the question of machines emulating the human brain in this well written article on "making brains". While quite short and lightly documented, Dvorsky's piece provides a useful outline of the problem, and a fairly sound description of a good approach for attacking the problem.

While I believe that reverse engineering the human brain is the right approach, I admit that it's not going to be easy. Nor is it going to be quick. This will be a multi-disciplinary endeavor that will require decades of data collection and the use of technologies that don't exist yet. And importantly, success won't come about all at once. This will be an incremental process in which individual developments will provide the foundation for overcoming the next conceptual hurdle.

But we have to start somewhere, and we have to start with a plan...The idea of reverse engineering the human brain makes sense to me. Unlike the rules-based approach, WBE works off a tried-and-true working model; we're not having to re-invent the wheel. Natural selection, through excruciatingly tedious trial-and-error, was able to create the human brain—and all without a preconceived design. There's no reason to believe that we can't figure out how this was done; if the brain could come about through autonomous processes, then it can most certainly come about through the diligent work of intelligent researchers.

...A number of critics point out that we'll never emulate a human brain on account of the chaos and complexity inherent in such a system. On this point I'll disagree. As Bostrom and Sandberg have pointed out, we will not need to understand the whole system in order to emulate it. What's required is a functional understanding of all necessary low-level information about the brain and knowledge of the local update rules that change brain states from moment to moment. What is meant by low-level at this point is an open question, but it likely won't involve a molecule-by-molecule understanding of cognition. _SentientDevelopments

Dvorsky goes on to describe the type of multi-disciplinary approach he has in mind, and bravely makes a prediction as to how long the effort will likely take: 50 to 75 years. This is a much longer timespan than Kurzweil and most AI researchers are giving, but I suspect it is closer to a realistic mark.

There are a couple of small criticisms I have to make. Dvorsky expects a workable brain emulation to be built within a "digital substrate":

.... if you believe that there's something inherently physical about intelligence that can't be translated into the digital realm, you've got your work cut out for you to explain what that is exactly—keeping in mind that any informational process is computational, including those brought about by chemical reactions. Moreover, intelligence, which is what we're after here, is something that's intrinsically non-physical to begin with.

Here, it seems that Dvorsky has it backwards. It is the persons who believe that intelligence can be made to work in a different physical substrate than the brain who bear the burden of proof to show that intelligence can be "transferred" to the "digital realm." We only have one proof of concept of intelligence up until now, which is a bloody ball of fat resting on a stalk rising between the shoulders of homo sapiens.

In another place Dvorsky asserts:

... the brain contains masterful arrays of redundancy; it's not as complicated as we currently think.

In truth, the brain is far more complicated than we can currently imagine. The question should be: Is the relevant functionality within the brain/mind which generates consciousness and intelligence, perhaps "not as complicated as we currently think?" Al Fin cognitive theorists believe that such a thing is possible, as long as we take care not to stumble amongst the numerous overlapping logical levels which present themselves whenever attempting to deal with this problem.

Dvorsky is quite right that the brain emulation problem is going to require extensive multi-disciplinary effort. We will need multi-disciplinary teams, as well as team members who themselves have multi-disciplinary training.

The great online debate between Ray Kurzweil and PZ Myers continues unabated, but it has very little to do with the eventual creation of a machine intelligence modeled after the brain.

If I had to choose one or the other to lead an effort to create an artificial brain, I would choose Kurzweil, hands down. Myers is an academic on the "intellectual" side -- an intellectual being someone who is rarely challenged by reality when he makes a mistake. Kurzweil's inventions and products have to work. That puts Kurzweil firmly in the reality-based camp, regardless of how many in the media and academia call him a kook.

Neither Ray Kurzweil nor PZ Myers Understand the Brain

Irrepressible bio-blogger PZ Myers has attacked futurist inventor and author Ray Kurzweil on his blog, accusing Mr. Kurzweil of failing to understand the human brain. But it seems that Mr. Myers was unwittingly attacking second-hand accounts of a talk given by Kurzweil, rather than responding to Mr. Kurzweil's actual claims. Kurzweil takes Myers to the woodshed for that mistake.

Other prominent tech- and mind-bloggers such as Brian Wang and George Dvorsky have reacted to this tiff, appropriately pointing readers to Mr. Kurzweil's actual words on the topic.

Lost in all the ballyhoo is the obvious fact that in reality, neither Kurzweil nor Myers understand very much about the brain. But is that clear fact of mutual brain ignorance relevant to the underlying issue -- Kurzweil's claim that science will be able to "reverse-engineer" the human brain within 20 years? In other words, Ray Kurzweil expects humans to build a brain-functional machine in the next 2 decades based largely upon concepts learned from studying how brains/minds think.

Clearly Kurzweil is not claiming that he will be able to understand human brains down to the most intricate detail, nor is he claiming that his new machine brain will emulate the brain down to its cell signaling proteins, receptors, gene expression, and organelles. Myers seems to become a bit bogged down in the details of his own objections to his misconceptions of what Kurzweil is claiming, and loses the thread of his argument -- which can be summed up by Myers' claim that Kurzweil is a "kook."

But Kurzweil's amazing body of thought and invention testifies to the fact that Kurzweil is probably no more a kook than any other genius inventor/visionary. Calling someone a "kook" is apparently considered clever in the intellectual circles which Mr. Myers' and the commenters on his blog travel, but in the thinking world such accusations provide too little information to be of much use.

Clearly if Mr. Kurzweil understood the brain, he could simply sit down and design an artificial brain based upon the principles which he already understands. The fact that Kurzweil places the development of such a human-level thinking machine 2 decades in the future, suggests that Kurzweil himself is not attempting to disguise his lack of comprehensive understanding of the brain.

I should point out that it is neuroscientist Henry Markram who is attempting to reverse-engineer a human brain to incredibly exquisite levels of biological detail -- in an attempt to study the function and potential pathologies of the brain. Kurzweil is not taking that path of reverse-engineering, but is rather attempting to extract principles of higher level mental functioning from the study of the brain. These "higher level mental functions" may appear to be quite low-level to a lay-person, but to a neurobiologist they will seem quite high-level indeed.

Life scientists do not understand life, really. Take this interesting Spiegel Online interview with Craig Venter on the genome. You would think that if anyone would understand the genome, it would be Craig Venter. But no, he admits that he does not -- not nearly to the extent that he intends to, at least. That is why he goes to work every day, because he understands just enough to want to understand more.

So while PZ Myers apparently fell off the wavelength upon which Ray Kurzweil was transmitting, that is no reason why the rest of us cannot follow Kurzweil's progress in his quest -- as food for thought.

Full Disclosure: Al Fin has in the past criticised Ray Kurzweil's approach to artificial intelligence as being insufficiently nuanced -- based upon Kurzweil's own writings. But a man with Mr. Kurzweil's track record of accomplishments is not one who should be written off. Such a person has been wrong innumerable times in his past, and has come back to correct his mistakes and move far beyond them. Every person of high achievement must go through such a process of being wrong and learning from it. It is one of the disgraces of modern education, culture, and child-raising that "being wrong" or "failing" at something, is considered to be an object of shame or disgrace. Far from being disgraceful, it is a necessary part of living and learning.

It is the dogmatist who is unwilling or unable to learn from his mistakes who should be avoided. The person who is unwilling to put in the necessary hard work to correct his own faulty assumptions and innate prejudices, is the person who will achieve little in the end. Except, perhaps, for calling everyone who disagrees with him a kook.