Norvig on Adaptive Software

Agent Technology, Decision Theory, Reinforcement Learning, Probabilistic Networks? I say these will overcomplicate your software and likely have scaling problems. Avoid over-engineering. It is not necessary for a product to be "adaptive". I wouldn't want my refrigerator to adapt into a washing machine.

It seems like there are a lot of software engineers that try to achieve the absolute perfect design, just like there are many who don't try at all. As with anything in life, extremes are rarely optimal.

It is advantageous for your source code to be succinct and rewritable, but not for it to adapt its own resource usage patterns. The adaptations could cause cascading performance problems. Just do the simplest design that gets the job done, profile, then use the feedback to optimize your algorithms. Such is software.

Awesome post. I've been thinking a lot about these ideas and its nice to see it presented so clearly and concisely.

Interesting. I wish there was a date on it though.