The Theory of Everyone

For around a decade now I have been banging on about the Affective Context Model and how we learn – and finding that no-one has the foggiest idea what I am talking about.

As a result, most of my posts are not about the underlying theory, but about its practical implications:  user-centered learning design (5Di) and the design of resources and experiences.

But I continue to believe that the underlying theory is important, not least because it explains much more than learning & memory: areas such as creativity & storytelling, everyday conversation – even AI (I think it is true to say that one could not begin to design a human-like AI without understanding the Affective Context Model - though you can design inhuman AI, and might make some progress through trial & error).

It occurred to me that there might be a story I could tell that would explain it better – so here goes:

Imagine you had a friend at University who studied computer science - Jo. At the time you lost touch, Jo had just set up a small company to do ‘sentiment analysis’. This was many years ago, and back then you had no idea what sentiment analysis was but as Jo explained ‘It’s a computer program - it looks at reviews that people have written – products and stuff - and figures out if they are positive or negative.’

Many years later you bump into Jo unexpectedly whilst waiting for a flight to Munich and the conversation goes like this:

You: so how’s the business going, are you still doing the – what do you call it – ‘sentiment analysis’ thing?

Jo: It’s going great. Yes – I mean - we’ve come so far… I’m not even sure you’d call it sentiment analysis anymore.

You: how do you mean?

Jo: so – back when we started out the computer would take a couple of sentences – like a product review or something - and figure out if they were positive or negative, right?

You: right.

Jo: So now we use a neural net and instead of ‘positive’ and ‘negative’ sentiments it recognizes around a thousand distinct sentiment markers.

You: uh-huh. And what can you do with that?

Jo: Well - it’s just amazing – for example we can feed it a short story that someone has written – or just a short story that they have read – and it does the sentiment analysis and then, sometime later, it can completely reconstruct the story based only on the sentiment analysis.

You: wow. That does sound cool. How accurate is it?

Jo: that’s the interesting bit. It depends how you measure it: if we show it to the people who wrote or read the original stories it’s around 95 percent accurate – meaning they mostly can’t distinguish the reconstructed stories from the story as they remember it... but – if we compare it to the actual text, it’s only about thirty percent accurate, word for word.

You: what does that mean?

Jo: Well – we don’t know for sure – but our working hypothesis is that our sentiment analysis algorithm is doing exactly what peoples’ minds are doing: in other words it stores lots of sentiment data when processing the story, then rebuilds the story based on that. So people can’t tell the difference between their reconstruction and our reconstruction – even though both are, strictly speaking, inaccurate. Basically: people don’t remember stuff – they just remember how it made them feel.

You: so why only ninety-five percent accurate?

Jo: well - that’s why I’m flying to Munich. It turns out the variation seems to be down to individual differences – when you read a story, different people react a bit differently to the same things. We call it ‘affective idiosyncrasy’. So we’ve started ‘mapping’ people – we get them to tell stories – about their lives that sort of thing – and them the algorithm can pretty much tell you what they will remember – for example as they wander around a city – versus someone else.

You: starting to sound a bit sci-fi!

Jo: No kidding! We found that after a few days of sentiment mapping even close relatives couldn’t tell the difference between stories generated by our algorithm and the real person!

You: I imagine that’s good for advertising too – I mean you can pretty much predict what will stick and what won’t right – on a person-by-person basis?

Jo: Exactly. Kind of ironic though – that it took marketing to figure out what makes people tick! Anyway – I’m being rude – what is it you’re up to these days?

You: Me? Learning and development.

Affective Context is important because it explains a whole bunch of things that are currently inexplicable: from the scientific (such as Elizabeth Loftus and the curious unreliability of eye-witness testimony) to the personal (such as why you remember an embarrassing comment, or school-dinners, rather than algebra), to the blindingly overlooked (such as what ‘aversive’ means in the context of operant or classical conditioning).

It tells us what people will notice on the train-ride into work, and the things they will tell their colleagues when they arrive at the office. Why people gossip and computers don’t. It makes sense of the ‘cognitive biases’ we have recently discovered – which turn out not to be biases at all.

In short, it’s a theory of everyone.