Curiouser and curiouser...

A few days ago I sent this tweet:

“I know that you feel differently about, say, different haircuts. But could you describe exactly how these feelings differ? No. This is why explaining Affective Context is hard.”

One of the problems I have experienced in explaining Affective Context is that people immediately think of global emotional states – happy, angry etc. – rather than the subtle differences in affective reaction to say – different haircuts. So I have been thinking of a different way to explain it – a way that goes something like this:

“Perhaps a better way to think about affective context is from a ‘bottom up’ perspective:

In 1949 Herb famously discovered that ‘neutrons that fire together, wire together’ - when neurons in proximity fire simultaneously the connections between them are formed or strengthened. So as you move through the world - from train station to cafe, from cafe to cinema, your neurons are firing and wiring.

But this begs the question: what, exactly are they recording? It seems reasonable to assume that the firing and wiring and resulting patterns correspond in some way to experiences. But in what regard? Are they recording the way things look? The way they sound? The meaning of things?

No. They are recording how you feel. They are recording your affective responses. It turns out you feel differently about the number ‘7’ than you do about the number ‘18’; you feel differently about different haircuts, about your house versus your Neighbour’s house. Your feelings literally connect you to the world.

Why would neurons encode experience in this way? Two reasons. Firstly, it’s a very efficient system - by encoding how you feel, you encode just enough information to reconstruct an experience in it’s critical characteristics (this is why memory is so unreliable). Secondly (and related to this) you only encode important stuff: since our affective responses are finely tuned to the impact on us, anything that doesn’t elicit an affective reaction is just not stored. It’s ‘boring’.”

But the problem that had been nagging me was why we have so few words to describe emotional states – if I am right why aren’t there words to describe the myriad of emotional states that reflect the subtlety of our emotional experience of the world?

And then it hit me. There are. They are all the words we use. That is what words do: they describe how we feel about things. Chair describes one set of feelings, stool another. Words are the subtle delineations that we draw between emotional states – chair describes a class of feelings (things that make us feel ‘chair’) stool, another.

Don’t believe me? This has long been a problem for AI – how exactly do we define what is a chair and is not a chair? Guess what? We never figured that one out – we just told the machines to copy what humans do. There is no definite category for 'chair'.

How about ethics? What, exactly, are the rules? Could it be that some things just feel right and some wrong?

Imagine trying to explain that to a machine: human words, human concepts – they just describe how we feel about things!