Saturday, September 17, 2005

Prospects for AI @ AC2005

The first panel discussion here at Accelerating Change 2005 was on the Prospects of AI. The panel includes an impressive line-up of people:
Neil Jacobstein, Chair, Innovative Applications of AI 2005; CEO, Teknowledge
Patrick Lincoln, Director, Computer Science Lab, SRI International
Peter Norvig, Director of Search Quality, Google; Author, Artificial Intelligence: A Modern Approach (the world's leading texbook in AI)
Bruno Olshausen, Director, Redwood Ctr for Theoretical Neuroscience
The introduction by Neil gave an overview of the many Task Areas being explored in the development of Artificial Intelligence. The key aspects of development are in Knowledge Engineering, Systems Engineering, and Business & Cultural. In his bullets about Ontologies and the Semantic Web, he referenced examples of early work - Cyc (OpenCyc), SUMO, and OWL.

The second speaker, Patrick, talked to the value of AI - Intelligence Amplification - and why this is necessary. The increasing gap between the complexities of technology, and human capabilities is causing more and more failures. AI can augment our ability to design complex systems, debug complex systems, and even operate complex systems. He talked about AI providing powerful abstracations - at the right levels - for both designers and operators. His examples included the progress and predictions in the uses of UAVs.

Third was Peter, from Google, who started with a slide titled AI in the Middle. His comments were about AI existing between authors and readers. His first point was about Machine Learning ... and joked about the fact that we don't know how to do it. His comments on AI in the Middle included how authors can write trillions of words, systems can detect certain patterns, and intelligent readers can then actually sort through this and find information. He went on to give examples of where apparent intelligence can emerge from larger amounts of data . .. giving examples of the accuracy of Arabic translation based on larger and larger data sets of example translation.

Bruno was the final panel speaker, and his area of research - Theoretical Neuroscience - is looking to the brain to gain insights into AI. The view of his team is to understand intelligence by understanding the brain. Not only the human brain ... but also other animal brains. One example is Jumping Spiders. He reviewed the knowledge that they have gained, and some interesting points that they are exploring. One area they have learned about involves vision, and where for each neural connection of retinal data (vision) coming in from the outside world, there are 10 times as many feedback connections coming from the cortex of the brain. So there is more information coming from the model in our own brain of what we are seeing, then the actual information being sensed! The model that we have in our mind contributes more feedback that what we are actually seeing! He explained that this is only one rich feedback loop that they are working to better understand.

It seems that all of the speakers look at advanced AI arising out of the shear number of patterns and complexities of their foundation work. I have to agree with them ... what we perceive as AI just might end up being an emergent property of the systems that we are creating ... not the explicit result of the planning and construction of the system.


Post a Comment

<< Home