Showing posts with label artificial intelligence. Show all posts
Showing posts with label artificial intelligence. Show all posts

Tuesday, February 07, 2012

Laggard Science of Artificial Intelligence Belatedly Discovers Embodiment

Researchers at the Artificial Intelligence Lab of the University of Zurich are challenging the basic underpinnings of artificial cognition research. The implications of their research -- if they are correct -- may shake the AI establishment to its roots.
...the notion of intelligence makes no sense outside of the environment in which it operates.

For them, the notion of embodiment must, of course, capture how the brain is embedded in a body but also how this body is embedded in the broader environment.

Today, Pfeifer and Matej Hoffmann, also at the University of Zurich, set out this thinking in a kind of manifesto for a new approach to AI. And their conclusion has far reaching consequences. They say it's not just artificial intelligence that we need to redefine, but the nature of computing itself. _MIT Technology Review
The researchers of Zurich have stumbled upon the concept of intelligence as having evolved within a complex and dynamic environment. While their ideas are likely to be dismissed by devotees of "algorithmic cognition" and "designed intelligence," these concepts actually appear rather obvious and elementary to anyone with a biological background, in addition to computational and cognitive training. One might even ask: "What took you so long?"
...Pfeifer has been shouting from the roof tops for several years, with some success, about the role that shape and form play in biological computation.

But today he and Hoffman go even further. They say that various low level cognitive functions such as locomotion are clearly simple forms of computation involving the brain-body-environment triumvirate.

That's why our definition of computation needs to be extended to include the influence of environment, they say.

For many simple actions, such as walking, these computations proceed more or less independently. These are 'natural' actions in the sense that they exploit the natural dynamics of the system.

But they also say it provides a platform on which more complex cognitive tasks can take place relatively easily. They think that systems emerge in the brain that can predict the outcome of these natural computations. That's obviously useful for forward planning.

Pfeifer and Hoffmann's idea is that more complex cognitive abilities emerge when these forward-planning mechanisms become decoupled from the system they are predicting. _MIT TR
Implications of Embodiment for Behaviour and Cognition

Artificial intelligence has been trying for the "easy score" ever since its founding, in the aftermath of World War II, and at the beginnings of the long-simmering cold war. But devoting one's resources to the easy score means that the deep underlying concepts are often neglected. And so it was and has been for over half a century. Advanced computing has been incredibly helpful over that time frame, demonstrating continuous progress, but there has never been a clear path to AI from anything coming out of computer science.

It is not clear that the Zurich researchers have gone beyond the insights of robotics researchers such as Rodney Brooks. But the fact that at least some AI workers are willing to think outside the box, is a hopeful sign.

Friday, October 21, 2011

Is the Singularity Near? Kurzweil vs. Allen

When will humanity reach Singularity, that now-famous point in time when artificial intelligence becomes greater than human intelligence? It is aptly called the Singularity proponents like Ray Kurzweil: like the singularity at the center of a black hole, we have no idea what happens once we reach it. However, the debate today is not what happens after the Singularity, but when will it happen. _BigThink
In the video below, Kurzweil discusses some of his ideas about the coming singularity, including timelines and cautionary notes.

Microsoft co-founder and billionaire Paul Allen recently expressed skepticism about Kurzweil's timeline for the singularity, in a Technology Review article.
Futurists like Vernor Vinge and Ray Kurzweil have argued that the world is rapidly approaching a tipping point, where the accelerating pace of smarter and smarter machines will soon outrun all human capabilities. They call this tipping point the singularity, because they believe it is impossible to predict how the human future might unfold after this point. Once these machines exist, Kurzweil and Vinge claim, they'll possess a superhuman intelligence that is so incomprehensible to us that we cannot even rationally guess how our life experiences would be altered. Vinge asks us to ponder the role of humans in a world where machines are as much smarter than us as we are smarter than our pet dogs and cats. Kurzweil, who is a bit more optimistic, envisions a future in which developments in medical nanotechnology will allow us to download a copy of our individual brains into these superhuman machines, leave our bodies behind, and, in a sense, live forever. It's heady stuff.

While we suppose this kind of singularity might one day occur, we don't think it is near.

...Kurzweil's reasoning rests on the Law of Accelerating Returns and its siblings, but these are not physical laws. They are assertions about how past rates of scientific and technical progress can predict the future rate. Therefore, like other attempts to forecast the future from the past, these "laws" will work until they don't. More problematically for the singularity, these kinds of extrapolations derive much of their overall exponential shape from supposing that there will be a constant supply of increasingly more powerful computing capabilities. For the Law to apply and the singularity to occur circa 2045, the advances in capability have to occur not only in a computer's hardware technologies (memory, processing power, bus speed, etc.) but also in the software we create to run on these more capable computers. To achieve the singularity, it isn't enough to just run today's software faster. We would also need to build smarter and more capable software programs. Creating this kind of advanced software requires a prior scientific understanding of the foundations of human cognition, and we are just scraping the surface of this. _Technology Review_Paul Allen
Allen goes on to discuss the "complexity brake," which the limitations of the human brain (and the limitations of the human understanding of the human brain) will apply to any endeavour that begins to accelerate in complexity too quickly.

Allen's argument is remarkably similar to arguments previously put forward by Al Fin neurscientists and cognitivists. The actual way that the human brain works, is something that is very poorly understood -- even by the best neuroscientists and cognitivists. If that is true, the understanding of the brain by artificial intelligence researchers tends to be orders of magnitude poorer. If these are the people who are supposed to come up with super-human intelligence and the "uploading of human brains" technology that posthuman wannabes are counting on, good luck!

But now, Ray Kurzweil has chosen the same forum to respond to Paul Allen's objections:
Allen writes that "the Law of Accelerating Returns (LOAR). . . is not a physical law." I would point out that most scientific laws are not physical laws, but result from the emergent properties of a large number of events at a finer level. A classical example is the laws of thermodynamics (LOT). If you look at the mathematics underlying the LOT, they model each particle as following a random walk. So by definition, we cannot predict where any particular particle will be at any future time. Yet the overall properties of the gas are highly predictable to a high degree of precision according to the laws of thermodynamics. So it is with the law of accelerating returns. Each technology project and contributor is unpredictable, yet the overall trajectory as quantified by basic measures of price-performance and capacity nonetheless follow remarkably predictable paths.

...Allen writes that "these 'laws' work until they don't." Here, Allen is confusing paradigms with the ongoing trajectory of a basic area of information technology. If we were examining the trend of creating ever-smaller vacuum tubes, the paradigm for improving computation in the 1950s, it's true that this specific trend continued until it didn't. But as the end of this particular paradigm became clear, research pressure grew for the next paradigm.

...Allen's statement that every structure and neural circuit is unique is simply impossible. That would mean that the design of the brain would require hundreds of trillions of bytes of information. Yet the design of the brain (like the rest of the body) is contained in the genome. And while the translation of the genome into a brain is not straightforward, the brain cannot have more design information than the genome. Note that epigenetic information (such as the peptides controlling gene expression) do not appreciably add to the amount of information in the genome. Experience and learning do add significantly to the amount of information, but the same can be said of AI systems.

...How do we get on the order of 100 trillion connections in the brain from only tens of millions of bytes of design information? Obviously, the answer is through redundancy. There are on the order of a billion pattern-recognition mechanisms in the cortex. They are interconnected in intricate ways, but even in the connections there is massive redundancy. The cerebellum also has billions of repeated patterns of neurons.

...Allen mischaracterizes my proposal to learn about the brain from scanning the brain to understand its fine structure. It is not my proposal to simulate an entire brain "bottom up" without understanding the information processing functions. We do need to understand in detail how individual types of neurons work, and then gather information about how functional modules are connected. The functional methods that are derived from this type of analysis can then guide the development of intelligent systems. Basically, we are looking for biologically inspired methods that can accelerate work in AI, much of which has progressed without significant insight as to how the brain performs similar functions. _TechnologyReview_Ray Kurzweil
Kurzweil's attitude seems to be: "Because difficult problems have arisen and been solved in the past, we can expect that all difficult problems that arise in the future will also be solved." Perhaps I am being unfair to Kurzweil here, but his reasoning appears to be fallacious in a rather facile way.

Al Fin neuroscientists and cognitivists warn Kurzweil and other singularity enthusiasts not to confuse the cerebellum with the cerebrum, in terms of complexity. They further warn Kurzweil not to assume that a machine intelligence researcher can simply program a machine to emulate neurons and neuronal networks to a certain level of fidelity, and then vastly expand that model to the point that it achieves human-level intelligence. That is a dead end trap, which will end up wasting many billions of dollars of research funds in North America, Europe, and elsewhere.

This debate has barely entered its opening phase. Paul Allen is ahead in terms of a realistic appraisal of the difficulties ahead. Ray Kurzweil scores points based upon his endless optimism and his proven record of skillful reductionistic analyses and solutions of previous problems.

Simply put, the singularity is not nearly as near as Mr. Kurzweil predicts. But the problem should not be considered impossible. Clearly, we will need a much smarter breed of human before we can see our way clear to the singularity. As smart as Mr. Kurzweil is, and as rich as Mr. Allen is, we are going to need something more from the humans who eventually birth the singularity.

Friday, October 07, 2011

Evolving 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.

Taken from an earlier Al Fin posting