Sunday, January 14, 2007

Artifical Intelligence and Society

Artificial Intelligence and Society
Symposium on AI in Industry (SAII) http://sigai.cdacmumbai.in/, held on January 10, 2007, collocated with IJCAI 2007 http://www.ijcai-07.org/

I contributed to a panel on Artificial Intelligence and Society at the above-mentioned symposium held on January 10, 07 in Hyderabad. Our task had been cut out for us as Prof Raj Reddy had given a keynote at IJCAI 07, focusing on what AI and Robotics can do for society. He had also mentioned an Indian initiative to set up a center to carry out research in the field.

I hope that report on Prof Reddy’s keynote and on the discussions in the Panel would be reported on the net soon, perhaps on the websites mentioned above, or even on this blog.

The large audience we had for the panel very pleasantly surprised me. It was heart-warming to see the enthusiasm displayed in comments from the floor. It is clear that a lot of people feel that is it is worth using challenges drawn form the point of view of benefits to society as drivers to develop AI.

Let me add my own comments here. I believe that AI has a tremendous potential to contribute to education. The world has a few billion people who have not had the opportunity to get a good education. See http://www.ei-india.com/full-report.pdf

Nearly a billion people were illiterate according a UNESCO report of 2003.
http://portal.unesco.org/en/ev.php-URL_ID=10513&URL_DO=DO_TOPIC&URL_SECTION=201.html
Their fate would no longer matter after fifty years, when they would mostly be dead. But, as humanity, is this what we want to see – a few billion people live through their lives in the 21st century with a bad education? Unfortunately, mortality among illiterates has been a major contributor to the so-called “increase in literacy” in many parts of the world. I don’t want to call it increase in literacy.

I believe that the best of available technology, and all technology we can develop, would be needed to solve this problem within the available window of a couple of decades. Of course, mere technology will not do. We will need science, particularly cognitive science and educational psychology. Even more, we will need the political will and the resources that would be required to mount this “grand challenge”.

AI’s potential in assisted living and in helping the sensory handicapped have been pointed out by many people. I will not dwell on them.

Let me give a couple of practical examples to justify my faith in the potential of AI. Dr William McBride, an obstetrician, had published a letter in The Lancet, in December 1961, pointing out that the drug Thalidomide given to pregnant women could cause serious birth defects in children. I have heard an Australian scintist say in a talk that McBride had used a very simple form of cluster recognition in his detective work to track down the cause of birth defects in children he had encountered in unexpected numbers in Sydney. In a great example of the power of epidemiology, he had used a map of Sydney and pinned a small paper flag at every address at which he knew of a child being born with birth defects. Unfortunately, soon enough, he had noticed two big clusters forming on the map. He is said to have tracked these down to two doctors who had prescribed the then relatively new drug Thalidomide to treat nausea in pregnancy. McBride won a Nobel Prize for his discovery.

(I cannot find supporting evidence for the cluster reognition part of this story, and would be grateful to readers for commenting on the accuracy or otherwise of this part).

The other example deals with a case in which simple techniques of analysis were helpless.

I quote Chris Darnell, who is now involed in very large scale financial management:

“I had been discouraged with the failure of traditional statistical methods to analyze financial market data characterized by a low signal-to-noise ratio. After reading some of the early papers by Professor Kanal in the new field of statistical pattern recognition, I called Laveen--out of the blue … He gave me some great suggestions for using new analysis methods to analyze financial data”.

Among other things, Darnell and Kanal explored the use of the open source statistical pattern recognition software ISPAHAN. This was all circa 1981, when people thought that pattern recognition was pretty much AI! Darnell has now acknowledged Prof Kanal’s guidance with a handsome endowment. The University of Maryland announced recently the Darnell/Kanal Professorship in computer science. See

http://www.cs.umd.edu/newsletters/InsideCS_2006f.pdf

I hope that these two examples provide some support to the faith that AI can do some good to the world. “Failure of traditional statistical methods to analyze financial market data characterized by a low signal-to-noise ratio” was a big problem in financial management. Can you imagine a similar problem in epidemiology? Do not many AI researchers have tools to help in this? This may not be a new idea to those working on machine learning. See

http://dimacs.rutgers.edu/Workshops/DataMiningTutorial/dmtutorialreport.pdf

However, that does not detract from the argument in favor of AI research being aimed at social good.

P. S. You might wish to read

http://newstudentresearch.blogspot.com/2006/08/exploiting-search-speech-recognition.html

for some suggestions on creating socially relevant systems using speech interfaces.


Srinivasan Ramani