In March of this year, DARPA announced plans to undergo research into the idea of a “smart camera”, where upon endowing a system with such a smart camera, the system would have a greater understanding of its surrounding world, thereby making the system all the more intelligent.
This project is being called Mind’s Eye, and its purview has been stated as the pursuit of ‘visual intelligence’. Various visual systems thus far designed have proven to be fairly good at object recognition but have yet to demonstrate recognition of action, nor have any systems been so designed as of yet. This move intuitively makes sense. After all, the world in which we live is not merely an aggregate of objects but is actually the interplay between those objects, which ultimately gives rise to context; something that computer scientists and philosophers alike consider epistemologically invaluable when it comes to understanding as humans regard it.
Action recognition is arguably the key for unlocking a whole new wave of intelligence development in humanly designed machines. If a system can be given the capacity for recognizing and tagging certain actions that fall within the view of its visual lens, then this seems tantamount to its being able to attribute reasons for a given action’s occurrence; and reason, of course, has been an enterprise that so far only humans (maybe animals to a lesser degree?) have been fortunate enough to enjoy. With the ability to reason about perceived motions, a system would be capable of constructing narratives on what is being perceived rather than simply treating it as a collection of static objects, and with the ability to infer based on context, the number of possible applications of such a system are as indefinite as they are troubling.
Nevertheless, the military’s purported first use of the Mind’s Eye project, if it succeeds, is to beef up aerial surveillance drones. If the product turns out to be reliable, it is likely to change the video surveillance paradigm in general, making that camera at the nearest 7-11 gas station much more intelligent. The implications of such a development would of course reach much further than mere security, and if machines prove capable of the slightest degree of contextual reasoning, the result would call for a philosophical restructuring of what we consider intelligence to be, and what the human place in the world is.