Just as electronics are intrinsically faster than ion-pumping nerves, an artificial approach to learning has emerged which fundamentally improves on what is possible in completely natural systems. It is ``statistical inference'', based on the use of statistics and probability theory to formulate and solve a great variety of problems in pattern recognition and learning. Already statistical inference methods are used in a wide variety of space applications including astronomical image understanding. Biosynthetic engineering will make further large improvements in the practice of statistical inference by introducing evolutionary and developmental processes that result in high-quality algorithms. These processes will include (a) evolution and adaptation on graphs that represent knowledge, (b) the operation of powerful statistical models called ``stochastic parameterized graph grammars'', (c) the stepwise creation and refinement of inference algorithms for these models, and (d) the automatic generation of computer implementations for these algorithms (e.g. codes or circuits) by a future compiler incorporating principles from development. The impacts on intelligent systems in space will be enabling, as outlined above. The same graph grammar statistical inference techniques will be amenable to creating intelligent systems for space biomedical research and biotechnology: for example, the core networks of biomolecular interactions within living cells can be understood and then manipulated.
These directions were developed in an exciting workshop at JPL in October 2003 attended by Amy Braverman, Kenneth Hurst, Michael Turmon, Jeffrey Jewell, Andrea Donnellan, Virendra Sarohia (JPL), Bernd Fischer (Ames), David van Dyk, Padhraic Smyth, Max Welling (UCI) and Eric Mjolsness (UCI/JPL).