Rodrick Wallace of the New York Psychiatric Institute has released a draft version of a new paper (PDF format) in which he speculates on the mathematical foundations of machine intelligence and the difficulty of controlling the development of a complex, conscious machine. Even if such a machine could be created in an initially stable state, Wallace believes it likely that the machine would become psychotic. "The most likely use of the first generations of conscious machines will be to model the various forms of psychopathology, since we have little or no understanding of how consciousness is stabilized in humans or other animals." Last year we mentioned another paper by Wallace in which he predicted that biomimetic machines were likely to "fail insidiously, irregularly, and progressively" for similar mathematical reasons.