I have fond recollections of the early years of my "discovering" the field of AI. Coming across it in a graduate course in Michigan back in 1985, I was fascinated and inspired by a field that had the bold vision of nothing short of modeling human intelligence. The field in its early days consisted of a collection of works that spanned everything from CS Theory to Biology, to Psychology, to Neural Sciences, to Machine Vision, to more classical computer science tricks and techniques. The excitement was very high and the expectations even higher. As I decided to start working in the subarea of Machine Learning, I started to realize how difficult the problems are and how far we truly are from realizing the ultimate dream of a thinking machine. What I also realized at the time was that specialization with some deep technical approaches and mathematical rigor were a necessity to make progress. These techniques came at a high price: the romantic ideals of building a machine that could think like humans were shattered in favor of understanding how the simplest problems of induction and data analysis can be addressed. We also realized that solving the technical problems required utilizing many classical techniques developed in other fields: mathematics, statistics, operations, research, and so on. The result was inevitable: AI fragmented into smaller camps, with each camp focused on solving a problem so narrow that all relevance to the original global dream of a thinking entity was erased. While we built systems that could solve narrow problems better than any human could ever hope to solve and at scales that literally boggle the human mind, we were making no progress on understanding how our own brains managed to go from low-level signals to high level thought so elegantly and quickly.
In reflecting back on those days of romantic excitement, I am very pleased at what they drove in terms of engineering achievements and new fields of study. In my own area of Machine Learning, while the vision of pursuing general algorithms that "learn from experience" morphed itself into highly specialized algorithms that solve complex problems at a large scale, the result was the birth of several new subfields of specialization. Combining learning algorithms with database techniques and algorithms from computational statistics resulted in data mining algorithms that work on very large scales. The resulting field of Data Mining is now a vibrant field with many commercial applications and significant economic value. This journey has also taken me personally from the world of basic scientific research, to the business side of realizing economic value from applying these algorithms to commercial problems, to finally working at the "strategy" level on the senior executive team of the largest Internet company in the world: Yahoo!, where data drives many products and strategies. I have been fortunate enough to witness this AI-inspired research solve problems that scientists could not solve in astronomy, planetary geology, remote-sensing, and other medical applications. I also witnessed the power of these algorithms in the business world: from forecasting automobile sales, to churn modeling in telecommunications and banking, to marketing in financial services. AI technologies are a routine operation in our search engines and our indexing of the web, that almost every consumer uses, relies on machine learning techniques to adapt itself to user needs and determine relevance of documents to queries.
In looking back at it, I can only say in wonder: what a ride! In looking at the future, we have much to do and I hope we make some serious progress on that original romantic dream of building machines that truly exhibit learning and thought in general. The dream is still worth pursuing, especially after all we learned over the past decades. The ride will be a lot more exciting for the new researchers entering the AI field.