TTOY’s technologies are based on decades of work in modelling and learning using AI, OO,logic programming, data mining, and Unix scripting.
With these tools, we strive to simplify human-in-the-loop theory formation and revision. The goal is to augment what humans do best with what computers do best. Humans, we believe, are very good at proposing new theories while computers are very good at reviewing and changing those theories.
(Note that by “theory formation” includes both knowledge engineering and software engineering tasks. KE uses explicit declarative theories while the thoeries in SE are often tacit and may only be explicated when, say, a test engineer asks the question “what is it, exactly, that this system is trying to do?”)
In the (almost) two decades of building tools for this domain, we have tried various approaches including logic programming, object-oriented programming, rule-based programming, visual programming, qualitative reasonng, design or analysis or inference patterns; and classical abduction.
Right now, our preferred method is a combination of stochastic abduction (which can generate consistent possibilities very fast) and treatment learning (a special kind of data miner invented by TTOY). The hypothesis here is such a combination is simple to implement, fast to run, can scale to very large theories, and which can significantly simplify the modeling process.