on Aug 9th, 2007Open World Computing
There was an interesting concept put forward to me by my professor – Open World Computing. A software that is not bound by any restrictions or constraints. A software that learns and adapts to its environment. Think of a person who is taken from a metro and put in a village. Does the person fail and give up like a computer program ? No. the adaptability of living beings is something so hard to understand that it can take probably another 1000 years to just simulate a living being, let alone learn its qualities.
Think of the same thing in software. A software that is programmed based on generic constraints and the software dynamically learns from its environment through input devices like sensors and then adapts to the changed environment. The classical shortest path problem can be taken as an example. If a crawler running through the shortest path is lifted from its path and put somewhere else, will it be ale to comprehend the change and then quickly adapt or will it be lost. What if the graph changes and produces a lot of cycles or what if the problem statement changes during the course of the program.
I know that a very few people are relating to what they are reading, but the belief that a machine can learn and adapt is what scientists are trying to prove everyday. Machine learning is taboo after movies like iRobot and Terminator, but trust me, we are far far away from something like that. If we can solve a subset of problems of adaptability through an expert system, that will be an achievement in itself.
But how do you go about designing software with such requirements. I would go one step ahead and call it no requirements or changing requirements with no defined thresholds. No current methodologies like OO or aspects can cater to such a requirement. Probably a new scheme of designing learning software has to be developed. Taking tips from AI and the Turing thesis, a perfect turing machine is what is required. Its a hard call but we will get there one day.
