<A|Q> = <A|I> <I|Q>

The title is a bit of a joke (don't ask) and even vaguely relevant to the subject of this post, which is the utility of artificial intelligence in thinking about hard stuff like IQ, human intelligence and even consciousness. This thought is not remotely original: Daniel Dennett was writing decades ago that AI and cognitive science would revolutionise traditional Philosophy of Mind topics such as free will, the existence of other people ('other minds'), and the nature of intentionality and agency.

It's a stretch to get from neuronal sludge to introspection. Into that gap flow ideas like 'the soul', philosophical dualism and 'the ghost in the machine'. Once you have a passing acquaintance with the architecture of theorem provers, AI planners and robotic control systems, a lot of things which previously seemed to defy a scientific, materialist explanation suddenly seem .. kinda obvious.

AI systems typically use deduction and/or abduction. Deduction lets you plan your way, despite obstacles, from a starting situation to a goal. The paradigmatic example must be the satnav, although we used to quote chess playing programs. The less familiar abduction looks for a conceptual framework which would economically account for the data under consideration, and which would allow a novel problem to be solved (typically by then applying deduction). We call these learning systems and it's the new hot topic in AI. A good example of such a system might be IBM's Watson.

A while ago I wrote about the Gossip Cat concept: an AI 'toy' which could be a companion to the young or old. It seems that Watson is going to add some value here: the dinosaur toy in the video below is powered by Watson in 'the Cloud'.

(The dialogue model has got to stop aping Wikipedia though. The point is to chat!)

Yes, the dialogue is processed by an AI system - IBM's Watson - in the Cloud. This means that everything spoken in the room is sent off to IBM's servers (somewhere in the States, I imagine) to be interpreted. Has anyone thought at all about the privacy implications of this? Apart from the police and intelligence agencies, obviously.

So then I had a further thought. The toy has been placed in the suspect's household ...  'Watson in the Cloud' is surreptitiously taken offline and a spook-interrogator takes over:

"Tell me Luke, what time is Daddy coming home tonight ... ?"

However ... there's lots of our most precious stuff already in the Cloud: our bank accounts, our savings, ... DropBox. I don't think these privacy issues are the kiss of death per se; what we need are sufficient trust relationships for this novel form of Cloud-based surveillance - it's the unpredictability and 'always-on' character of real-time conversation which presents the challenge.