Accelerated Learning with AI systems?

Malcolm Gladwell‘s Outliers and Geoff Colvin‘s Talent Is Overrated are two recent, popular books that discuss expertise and “what it takes” to become a high performer in particular fields. Both authors touch on is the necessity of at least 10,000 hours of practice. Colvin qualifies it as very particular, deliberate practice. Others agree.

In Accelerated Learning (?) (I have not been able to find a freely available full text version of the article online yet), Robert R. Hoffmann, Paul J. Feltovich, Stephen M. Fiore, Gary Klein and David Ziebell pose a challenge for intelligent systems: Accelerate learning to reduce the time for people to acquire the necessary knowledge and experience to be high performers in their domain.

This strikes me as a fascinating, important application of AI technology. One reason this is important, as indicated by the authors:

Many organizations such as the US Department of Defense, NASA, and the electric utilities are at risk because of the imminent retirement of domain practitioners who handle the most difficult and mission critical challenges.
To accelerate proficiency, we must facilitate the acquisition of extensive, highly organized knowledge. We must also accelerate the acquisition of expert-level reasoning skills and strategies.

and

A classic estimate states that the development of of very high-level skills in any complex domain takes at least 10 years. But extraordinary experts who conduct mission-critical activities in industry settings have proven their value and earned extraordinary respect through the course of 25 to 35 years of experience.

If it takes fairly long to acquire necessary skills and if the qualitative differences between good and great performers is as significant as the research seems to suggest, then the advantages of time reduction become obvious.

I am guessing, approaches would have to support a high degree of personalization. The technology would have to do a very good job at discovering and assessing the user’s skill level as well as determining the best exercises and methods to stretch the user’s mind in just the right way. I am looking forward to learning about interesting projects that address this challenge.

It is intriguing to think about the impact that drastically accelerated acquisition of expert-level skills might have on us as individuals, society, and progress (scientific or otherwise) in general. Certainly, it may affect our ideas of what constitutes complexity in knowledge domains.

Measuring Machine Consciousness

In Can Machines Be Conscious?, Christof Koch and Giulio Tononi provide supporting evidence that consciousness does not require “many of the things we associate most deeply with being human: emotions, memory, self-reflection, language, sensing the world, and acting in it.”

The authors also propose a new type of Turing Test to measure machine intelligence:

So this is how we can test for machine consciousness: show it a picture and ask it for a concise description [see photos, “A Better Turing Test”]. The machine should be able to extract the gist of the image (it’s a liquor store) and what’s happening (it’s a robbery). The machine should also be able to describe which objects are in the picture and which are not (where’s the getaway car?), as well as the spatial relationships among the objects (the robber is holding a gun) and the causal relationships (the other man is holding up his hands because the bad guy is pointing a gun at him).

The machine would have to do as well as any of us to be considered as conscious as we humans are—so that a human judge could not tell the difference—and not only for the robbery scene but for any and all other scenes presented to it.

(Can Machines Be Conscious?)

This is very interesting. Of course, the machine would have to be able to support language understanding and generation to communicate its interpretation of the image.

The test also seems to require solving of the computer vision problem. I am not sure that that is necessary though. I am thinking this test could be performed entirely text based:

The user would describe a scene to the machine. The machine’s task would again be to provide a meaningful interpretation of the scene, given the description. A vague description should result in useful follow-up questions by the machine. Personalization and/or preference technology would need to be used to provide results most appropriate for the user’s background knowledge and expectations.

Is there a compelling reason that necessitates solving of the vision problem here?

Short notes on IEEE Spectrum’s Singularity Report

IEEE Spectrum published The Singularity – A Special Report back in June 2008. The report, both available online and in print features a number of articles and opinions looking at different aspects of this broad subject area. Here is a very brief survey of some of those articles.

The Consciousness Conundrum by John Horgan discusses brain complexities, neurobiological challenges and examines current work in the field.

Economics Of The Singularity by Robin Hanson defines a singularity as sudden, rapid acceleration of economic output, such as was observable with the agricultural revolution and the industrial revolution. Historical data suggests that we may be nearing another such change. The author discusses potential contributing causes as well as implications for human society.

Reverse Engineering the Brain by Sally Adee takes a look at David Adler’s work (and associated challenges) on building complete images of fruit fly brains as a first step towards imaging the human brain.

Can Machines Be Conscious? by Christof Koch and Giulio Tononi reasons that “consciousness does not seem to require [...] emotions, memory, self-reflection, language, sensing the world and acting in it.” A test to gauge the degree of consciousness is proposed: The ability to correctly interpret images.

Singular Simplicity by Alfred Nordmann rejects the notion of singularity and reasoning on which singularitarians base some of their claims.

In Rupturing The Nanotech Rapture, Richard A.L. Jones points out problems with mechanical approaches to nanotechnology.

I, Rodney Brooks, Am a Robot by Rodney Brooks describes challenges to be solved on the way to Artificial General Intelligence. Brooks describes the singularity as a period, not like an event and provides a speculative view of that future.

Book: Introducing Artificial Intelligence

The book Introducing Artificial Intelligence by Henry Brighton and Howard Selina is a very accessible introduction to the history and some key concepts of Artificial Intelligence.

Here are some of the topics covered:

• Strong AI vs Weak AI
• Psychology, Philosophy and Cognition
• Computation and Turing Machines
• Intelligence, Thought and the Chinese Room argument
• Ontology and the Cyc project
• Neural Networks, Machine Learning, Robotics

This is a small book and clearly the emphasis is introductory breadth and not depth. Like other books in the series, this one his very short and features illustrations on every page, accompanied by very short passages on text. The conversational style makes it well suitable for interested laypeople.