I went to a great lecture last night. I'm an active member of the British Computer Society (BCS) and every year a distinguished computer scientist is invited to give the Lovelace Lecture. This year, Professor Chris Bishop inspired the audience with a talk on "Machines that Learn: Adaptive Computing in the 21st Century".
Machine learning is used to solve lots of problems: identifying cancerous cells, filtering email spam, face detection, speech recognition etc. The cornerstone of machine learning is modeling uncertainty with Bayesian statistical methods. This method enables an algorithm to take new evidence and 'home in', statistically speaking, on a solution. The algorithm gets better and better at making predictions in the light of new evidence - it learns.
Our goat trail algorithm will use machine learning techniques to 'home in' on the best trail for a given search. Machine learning can also help identify the best search engine on which to blaze a new trail. This in turn will help solve the holy grail of metasearch.