An Environ- Ment/classiication Scheme for Evaluation of Domain- Independent Function--nding Programs. in Proceed- Ings of the Ijcai Workshop on Knowledge Discov- Discovering Knowledge from Observational Data. In

indications of the limitations of domain-independent function nding. In concluding, let me anticipate a possible objection, namely that the Bacon algorithm is a weak straw man against which to compare a new function-nding approach. In fact, though Bacon is the oldest and simplest of AI function nders, my experience suggests that it is quite diicult to outperform. I expect that more recent systems|including the new IDS Nord-hausen, 1989]|would do worse than Bacon in a similar test. Certainly, I am prepared to make the trial, if the author of any such system is willing to provide code and suggest appropriate parameter settings.