I think you may find more activity in #data-science. Depending on your definition of AI there's plenty of prior art before http://scicloj.ml, Neanderthal et al give you a great ecosystem of libraries for working with matrices, on the GPU or elsewhere. There are wrappers for Weka, which offers a fairly broad set of algorithms. Way, way back Incanter was a nice environment for stats but it's very hard for large Clojure frameworks to reach critical mass.