Ongoing successful operation of AI systems, build credibility and shifts more and more responsibility towards the well functioning automation. Over time human alertness dims, and the machine keeps its good work. Such very successful machine may harbor a logical trap which states that for a given combination of input, do something harmful, even catastrophic. Such disruptive dormant logic is common in regular complex programmed systems. Its origin may be a bug, a mistake, or long range dormant malware. Such logic though, in normal programs can be systematically spotted and neutralized. But AI runs on rules and action imperatives which are deduced from its inference engine. The human input is much earlier, much more subtle, and very difficult to spot and excise.

​

A self driving machine, for example, may harbor a logic element that prescribes a catastrophic move upon a rare combination of input. This rare combination is dormant for thousands of hours of operation before it pops up, and according to Murphy's law, it will pop up in the worst possible moment.

​

A solution approach to this dormancy threat is offered by BiPSA, where every inferential construct is tested against its contrasting construct, comparing likelihoods. Ideally critical decisions should be taken upon agreement between two or more independent inferential pathways.

​

A conservative approach to positive probabilities is also deployed. We build effective approximation routine by using Numerization - an iterative counting of natural number. Numerization allows one to read natural numbers as approximation of a more specified counting.

​

We seek conversation with other outfits which wrestle with AI dormancy threat.

​

​