IFTF's Future Now

David Brooks, cyborg

David Brooks is now augmenting. The piece is over the top, but is one of those "if he's doing it, it's either really big or is really over" data-points.

I have melded my mind with the heavens, communed with the universal consciousness, and experienced the inner calm that externalization brings, and it all started because I bought a car with a G.P.S....

I had thought that the magic of the information age was that it allowed us to know more, but then I realized the magic of the information age is that it allows us to know less. It provides us with external cognitive servants — silicon memory systems, collaborative online filters, consumer preference algorithms and networked knowledge. We can burden these servants and liberate ourselves....

Memory? I’ve externalized it.... [I]f I need to know some fact about the world, I tap a few keys and reap the blessings of the external mind.

Personal information? I’ve externalized it. I’m no longer clear on where I end and my BlackBerry begins....

Now, you may wonder if in the process of outsourcing my thinking I am losing my individuality. Not so. My preferences are more narrow and individualistic than ever. It’s merely my autonomy that I’m losing.

Technorati Tags: collective intelligence, end of cyberspace, geoweb, memory, mobility

What is Web 3.0?

Some friends of mine sell T-shirts that read, "Everytime you say 'Web 3.0,' a startup dies." In a more serious vein, Wade Roush writes in Technology Review about two strains of research that are reaching for Web 3.0 status, but are still "closer to 'Web 2.1.'"

The first category of projects is related to the Semantic Web, a vision for a smarter Web laid out in the late 1990s by World Wide Web creator Tim Berners-Lee. The vision calls for enriching every piece of data on the Web with metadata conveying its meaning. In theory, this added context would help Web-based software applications use the data more appropriately....

A second category of post-Web 2.0 projects focuses not on helping machines understand the meaning and the uses of existing Web content, but on recruiting real people to add their intelligence to information before it's used. The best known example is Amazon Mechanical Turk, a kind of high-tech temp agency introduced by the online retailer in 2005. The service allows people with tasks and questions that computers can't handle--for example, spotting inappropriate images in a collection of photos--to hire other Web users to help.

Wade is always worth reading. Alas, he seems to have led his very promising Continuous Computing blog go fallow.

Technorati Tags: AI, Web 2.0

When is AI no longer AI? When it works

Adobe Advanced Technology Lab researcher Bill McDaniel asks the provocative question, "why is it that when we build AI systems, the ones that work are no longer considered AI?"

This has plagued the AI arena for years. It seems that as soon as we develop a new technique... it is immediately classed as useful algorithms, semantic technology, vision processing, anything but AI....

I suspect some of that is because we do NOT solve these problems with emergent AI, intelligence that both arises out of some "ghostly signature" in our knowledge bases and tta seeks to extend itself. It is, I contend, the very absence of these traits that makes us so reluctant to embrace wht we HAVE accomplished as Intelligence.

In some region of ourselves, we know that what these techiques represent, as amazingly successful as htey are, is Artificial Smarts, not Artificial Intelligence. We sense that Intelligence will be recognizable and will be emergent, not algorithmic in nature. And we won't be happy with AI until we get that.

Is he right? I don't know enough about the state of AI to answer the question, but it's an interesting one.

Technorati Tags: AI

Rudy Rucker @ IFTF

Rudy Rucker is visiting the Institute today, talking about his new book, The Lifebox, the Seashell, and the Soul-- one of the most intriguing book titles of the year. Should be interesting.

Technorati Tags: computer, IFTF, science

New AI Brainwave

Having done corporate work in AI, I have a strong interest in practical artificial intelligence. This Infoweek article is a good overview of the direction of new AI work, by IBM, Intel, Msoft and PARC. Note the mention of Bayesian Techniques as being key.

... Now a new generation of researchers hopes to rekindle interest in AI. Faster and cheaper computer processing power, memory, and storage, and the rise of statistical techniques for analyzing speech, handwriting, and the structure of written texts, are helping spur new developments, as is the willingness of today's practitioners to trade perfection for practical solutions to everyday problems. Researchers are building AI-inspired user interfaces, systems that can perform calculations or suggest passages of text in anticipation of what users will need, and software that tries to mirror people's memories to help them find information amid digital clutter. Much of the research employs Bayesian statistics, a branch of mathematics that tries to factor in common beliefs and discount surprising results in the face of contrary historical knowledge. Some of the new AI research also falls into an emerging niche of computer science: the intersection of artificial intelligence and human-computer interaction ...

Brain Models for Intelligence

I reviewed Jeff Hawkin's book: On Intelligence here last year. The founder of Palm Computing suggested a model using the architecture of the brain to build artificially intelligent systems. This is a novel approach, since most AI systems do not attempt to use the brain as a model. His book was very good, but I suggested that he had considerable work to do to implement his ideas. Its another commendable attempt at the holy grail of machine intelligence.

Now an article in today's WSJ: Next Case for Palm Pilot Creators: The Brain and another in BusinessWeek: Jeff Hawkins' Bold Brainstorm, suggest that some progress has been made. He linked with Dileep George, who has started to implement some mathematics based on Hawkin's ideas, and formed a company called Numenta to work on practical applications. Both articles quote Intel scientist Gary Bradksi who says "Even if he's wrong, his theory is better than nothing. And it's 'attackable' -- and that's a good thing." Likely early applications are in machine vision and drug discovery. Well worth tracking. See also article in the NYTimes.

Pamela McCorduck on AI, again

I read Pamela McCorduck's book: Machines Who Think: A Personal Inquiry into the History and Prospects of Artificial Intelligence some 25 years ago, when the uptick of commercial Artificial Intelligence hype began. In part, her work and that of Edward Feigenbaum, of Stanford attempted to translate the technology for management.

Even then McCorduck's book was seen as being a bit too full of breathless praise of artificial intelligence and its future. At the time industrial practitioners like us were too willing to agree and gave out copies of the book freely. But practitioners soon found that it would be harder than that. And while McCorduck and others included discussions of how to deal with the broad replacement of human intelligence by machine, the practitioners started to elicit knowledge, one simple rule at a time.

This is a new version of her book, with a long afterword, which reviews AI progress since her 1979 book. I read the afterword of the new book first, recalling my impression of her original volume. I appreciated the fact that early on she admitted she was responsible for some of the over-hyping of AI in the 80s. She filled me in on what happened to Japan's Strategic Computing Initiative and what happened to the once voluminous DARPA AI support. Still, what I have now is a book that is unchanged in its first part, with all its hype intact ... yet down-toned, corrected and updated in its afterword. OK for historical purposes, but confusing, even misleading for the beginner. Its a good journalistic tale at times, but I would have preferred more technical detail about changes the 25-year passing of time. So a good read, with caution. Below, Scientific American's View:

The Value of Genetic Design

In the February MIT Technology Review: Unnatural Selection, which discusses the increasing value of design by genetic algorithms. They quote David Goldberg, director of the Illinois Genetic Algorithms Laboratory at the University of Illinois at Urbana-Champaign. He wrote an excellent book on the subject: Genetic Algorithms in Search, Optimization, and Machine Learning , which introduced me to the topic, and inspired me to write my own genetic codes. Great, gentle introduction. AAAI also has a useful link collection.

We are increasingly seeing problems which combine design and quantitative efficiency, and these are often combinatorially difficult problems that can lead to genetic methods. These are useful because they can provide models for problems whose mathematical statement are too difficult to specify or solve. I am in the process of exploring evolutionary methods, like genetic algorithms, so if there is anyone out there with substantive experience with vendors, I would appreciate comments.

...“Just as the steam engine created mechanical leverage to do larger tasks, genetic algorithms are starting to give individuals a kind of intellectual leverage that will reshape work,” Goldberg says. “By automating some of the heavy lifting of thought, we free ourselves to operate at a higher, more creative level.” Such freedom comes at a price, of course. It requires that engineers recognize the impossibility of peering into each and every “dark corner” and put their trust in yet another layer of mechanical assistance. But more and more of them are taking that leap ...

Biological Design: Ants Solving Problems

A colleague pointed out the below piece from NPR to me, the ant metaphor used for solving cellphone routing problems ... It relates to the previous post, to what degree is biological metaphor useful for designing an algorithm, a system?

(Search for audio link) ... As ants attack a breadcrumb, they appear to run into, over and around each other in complete chaos. But an article in the journal Nature posits that ants actually follow simple geometric rules as they forage your food. ... based on observation of foraging (food gathering) behavior of ants. As the ants begin their trail, they secrete a substance known as pheromone. Each ant is guided on a particular path based on its concentration of this pheromone. As the number of ants following a particular path increase, the concentration of pheromone on the path increases as well, further making the path more attractive for more ants to follow. This way the ants can effectively move around obstacles in its path and quickly adapt their movements to reach the food source. In this way, each agent working towards its individual goal actually drives the organization closer towards its larger goals...

Ant-based problem solving techniques are a form of agent-based problem solving methods called swarm-intelligence, developed in part at the Santa Fe Institute. Although a fascinating example of biologically inspired problem solution methods, their actual application is not common. One of the best known researchers in this area is Eric Bonabeau of Icosystem, who is working with DARPA and Pharmaceuticals to solve problems. See his book: Swarm Intelligence: From Natural to Artificial Systems, for a gentle intro.

This is an example of biologically-inspired solution methods. Neural Nets and Genetic Algorithm methods are other examples. I have some experience in the case of neural nets as a pattern recognition method. Nets are certainly useful for nonlinear applications, but their biological metaphor linkage is very weak. Use of biology as an inspiration for human design is also a related idea. Its reviewed in Steve Vogel's book Cats Paws and Catapults, which suggests that we have looked for biological inspiration after rather than before the fact. The broad meta-idea deserves more study.

Neurocomputing Data Mining Environment

NeuCom - A Neurocomputing Environment for Data Mining, Knowledge Discovery and Intelligent Decision support systems. Based on evolving connectionist systems (ECOS). Here is a PDF overview of ECOS. Brought to my attention by Steve Cayzer of HP Labs UK . Several demo downloads available at the site, have just started to take a look at them. An interesting comparision would be with SPSS/Clementine, another mining environment that includes stat and neural methods.

Berners-Lee Takes Chair at Southampton

Richard James of HP points us to the breaking news that Tim Berners-Lee, the inventor of the Web, is moving to the University of Southampton as a chair of their Computer Science Department. He will also retain his MIT appointments. Both of the links in the quote below have useful information about AI/Semantic Web work in the UK. Here is their press release. Richard writes:

....You may remember the name Professor Nigel Shadbolt as this was one of the contacts that was highlighted to us by David Dupplaw. Since HP has strong Semantic links with Southampton University it will be interesting to see what impact (if any) that Tim's appointment will have on that work...

Eats, Shoots and Leaves

A bit off technology topics, just completed ... though I guess it is ultimately about language and issues of precision. It shows how difficult AI analysis of language can be.

Eats, Shoots and Leaves, The Zero Tolerance Approach to Punctuation by Lynne Truss.

An interesting, often quite funny book about ...you guessed it ... punctuation. This was a best-seller in the UK. Its about precision in writing and the use of punctuation to deliver that precision. I doubt, though, I will be much better at punctuation after reading it. So does punctuation matter? ... consider the two following sentences:

“A woman, without her, man is nothing."
"A woman without her man, is nothing."

Bayesian Nets

A very good tutorial on Bayesian Nets, with lots of supporting information. Via the package Netica from Norsys. This methodology is becoming more common for delivering expertise. Because its statistically based it can model aspects of uncertainty in a system. The site has downloadable software for testing. This can be seen as a replacement for the 'rule bases' that we used for delivering expertise back in the 90s in expert systems. Here is another useful online tutorial.

Facial Avatars

The BBC reports on work at the University of Surrey on Avatars that can respond to visual signals. This reminds me once again of some of the work by Byron Reeves of Media-X at Stanford and his Media Equation book. And Rosalind Picard's Affective computing work at MIT. Such work could ultimately lead to more emotive, sympathetic avatars for machine based interaction with people. Since such systems also react to visual cues, such as facial expressions, they could also record reactions to product or product context. Since the avatar also reacts, it can lead to a stream of interaction between people and machine. Brought to my attention by Richard James of HP.

Jeff Hawkins: On Intelligence

Jeff Hawkins, founder of Palm Computing, just published a book: On Intelligence. Subtitled: How a New Understanding of the Brain will Lead to the Creation of Truly Intelligent Machines. Just completed this in my recent travels. This book really struck home to me because it paralleled my own exploration of the possibilities of Artificial Intelligence. From being inspired by the famous Crick article in Scientific American in 1979, where Crick basically admitted we knew very little about the brain, to building rule-based expert systems, to using neural nets to plumb hard pattern recognition problems. This always seemed like great progress to me ... from logic to networks that were at least inspired by brain structures. But all of this work, some very valuable for solving problems, never really got much closer to the goal of producing real 'intelligence'. I think he is right in thinking there is much to learn from the form and function of the brain.

Hawkins has the resources, and he has set up the Redwood Neuroscience Institute. Which is worth exploring, just scanning its publications gives me an impression of the difficulty of the problem, linking low level functionality to high level process. Its like being given the periodic table of elements and being asked to prepare the perfect omelette from sample chemicals. Basic chemistry is still missing and even the cookbook is missing. I look forward to following RNI's progress.

I particularly like the analogy to forecasting ... Hawkin's suggestion that the brain is a forecaster .. continually predicting future events ... aided by a huge amount of memory. Its not always right, but it learns and adapts based on the mistakes it makes and resulting feedback from its evironment.

Hawkins makes the case that with study of physiology and functionality of the neurocortex we are starting to make inroads into defining and ultimately leveraging machine intelligence. I think he is right but I am not as optimistic about how long it will take. I suggest we are mutliple decades rather than just a decade away from real progress. I am still battle-scarred with predictions in the 90s about how artificial neural networks would provide the ability to learn our way to intelligence. We learned a lot, but there was much hype as well. I hope I am wrong, then we can start working on the ethics of AI.

This book is very readable, largely non-technical though the latter parts of it contain some detail about the functioning of the neurocortex and may take some close thought. Also contains fascinating historical context. I strongly recommend this book if you have interest in this area.

Further, an insightful review from Corante.

... With resume bullets like inventor of the PalmPilot and CTO of PalmOne, a popular-science book about the future of computing certainly seems like an obvious choice for Hawkins. But as soon as you open up the sharp, electric-blue dust-cover, you’ll realize On Intelligence was probably the last thing you’d expect from a Silicon-Valley techie. Missing are detailed technology roadmaps and ethereal speculations about fantastical improbable futures. Instead, On Intelligence adeptly intertwines lay-English summaries of decades of research from neurophysiology, computer science, cognitive psychology, and even includes some well-placed philosophical sidebars that mesh into an approachable and well-written narrative addressing the plausible future of computing ...

John Holland and the Future of AI

John Holland is one of the best known investigators of adaptive systems, specifically the biologically-inspired method known as genetic algorithms used to solve very tough problems. He is also a fellow at the Sante Fe Instititute, where we have enjoyed examining his methods, and in some cases reapplying them. It has been claimed that adaptive approaches like Hollands will be the ultimate path towards machine intelligence. If you are interested in more details, see the Echo Simulation system, inspired by Holland and available from SanteFe. In the article below, a good overview of his work, Holland is pessimistic about the development of competent artificial intelligence anytime soon.

Falling Prey To Machines?
ANN ARBOR, Mich -- It's coming, but when? From Garry Kasparov to Michael Crichton, both fact and fiction are converging on a showdown between man and machine. But what does a leading artificial intelligence expert--the world's first computer science PhD--think about the future of machine intelligence? Will computers ever gain consciousness and take over the world?

"Computer sentience is possible," said John Holland, professor of electrical engineering and computer science and professor of psychology at the University of Michigan. "But for a number of reasons, I don't believe that we are anywhere near that stage right now." ...

Alice the Chatbot Wins the Loebner Again

The BBC provides an overview of the recent Loebner chatbot competition, which ALICE won for the third time. The ALICE chatbot was developed by Richard Wallace. A few years ago ALICE was internally proposed for an interactive way to gather consumer comments. A technical tidbit: This form of AI uses Zipf's law as part of the basis for its interaction. Wallace and his team have developed something called the Artificial Intelligence Markup Language (AIML) which is worth a look.

We experimented with web based chatbots using the Extempo system. Neither of our efforts went very far, but the results are worth examining. For purposes of scale, having a system that interacts like a person, using a knowledge base of information, can be very useful. The knowledge simply needs to be uploaded rather than learned. It turns out that people interact with AI systems in remarkably trusting ways, I refer again to Stanford's Byron Reeves work on media interaction, where we were inspired by his seminal book with Clifford Nass: The Media Equation.

Tricky issues of privacy also emerge, should we demand to know if we are talking to an AI versus a human? What is the implication of giving data to an AI that represents itself as a human? I consider the Loebner Prize more of a slight-of-hand trick that takes advantage of our trusting nature, rather than a creating true AI, but it is a step that suggests how we may eventually solve the problem.

Semantic Technologies at IBM

A colleague pointed me to a recent posting at IBM where Juhnyoung Lee of their Hawthorne Labs does a nice job of introducing semantic technologies, and why they have declared them to be one of their Emerging Technologies. They also include a number of download-able toolkits for experimentation, another in particular covers agents, which I posted on here earlier. Since suggested by Berners-Lee, the Semantic Web has produced excitement in academia, but relatively little actual application in industry. Can anyone point us to practical applications we can use as examples?

Autonomous Agents

A good short article in ComputerWorld on Autonomous Agents. Some of my contacts at NASA Ames continue to relate exciting results of their work with autonomous systems. Agents have been a topic here before, but only in a sense of building models that use software agents which represent real world entities. Autonomous agents in this other sense are software systems, often linked with hardware, that make decisions for themselves.

In fact we are well acquainted with agents in the form of self-regulating systems. The thermostat, anti-skid brakes and many other examples are well known. They have knowledge of their environment, link to some intelligence, and make decisions without being checked by people. Their autonomy, though is usually limited to simple (but increasingly complex) tasks.

Robotics also has to address the issue of autonomy. Most robot examples today have minimal autonomy. For example the Roomba sweeper performs a simple task, with simple inputs. Yet the performance of the task is still complex. Many robots that do complex things today are either strongly repetitive in their tasks, or are controlled remotely by people.

NASA is in the business of launching things to far away places, some so far away that controlling them remotely is, because of communications time, not possible. So they have built some of the most complex autonomous systems in existance. These systems must adapt to their environment and make complex choices without human interaction. They may even have complex sets of goals, in the form of agendas, to direct their decisions. The Earth Observer Satellite 1, shown in the picture, has autonomous agendas directing it.

How do we grant different levels of autonomy? How are levels of trust linked to specific decisions? How do we ultimately deal with the responsibility that comes with mistakes? All being addressed at NASA and elsewhere. This will, I believe be an increasingly important element of information technology in the future.

Roger Schank and Cognitive Arts in Retail

I met Roger Schank back in the early 1990s, and were reminded of him by an article in The Edge. At the time he was head of the Yale AI project, and at a talk at an AAAI conference around that time he talked about the interaction of Cognitive Science and Artificial Intelligence.

Our conversations with him at time were combative. We were relatively young corporate proponents of using AI to store and leverage knowledge. He was an academic who believed that much work needed to yet be done to use the tools of the time, expert systems. At least based on subsequent corporate investment, he was right.

We followed his work since then, on and off, over the next dozen years, and just read about some of the work he is doing as CEO of Cognitive Arts.

Most interesting is his ongoing work with Walgreens for training new managers. Started in 2002, this approach is now rolled out to 3500 stores. CA has also started a broader Retail Practice Group.

Its all about Simulation Based Learning, which CA calls the Experience Learnings Solution. This is based generally on the AI concept of Case-Based Reasoning, or CBR.

Simulation, of course, is an excellent means of understanding a complex system. Applications from MS Flight Simulator to SimCity show how people can engage with very complex systems through action and feedback. The idea of CBR links past cases, or experience, with user input to find best matches and alter them to the problem at hand.

Its also a bit ironic that we are now also in the process of working on some simulation-based engagement systems that could likely benefit from some of CA's work ... will write more on that later.

Its worth looking at CA's work with Walgreen's and also another case-study on their site with GE Capital.

Schank appears to have made some useful progress with artificial intelligence over the past years.