Blog

Following the release of new software for behavioural modelling, well meaning friends have commented ‘so its basically machine learning’ right?

I felt like a tarty reply “we do proper behavioural modelling and use it to develop road networks, cities and healthcare treatments, not to generate clickbait and abstract art.

At a deep mathematical level through, they are the same. Both AI and behavioural modelling produce equations that apply parameters to inputs. They both produce an output that predicts an outcome. Whether it’s filtering spam or predicting uptake of a new softdrink.

The difference is that in a neural network, these parameters cannot be interpreted, they (just) work. In Behavioural Modelling they have an explicit meaning. And why is this? Because while the researcher started with a hypothesis a direct inquiry, the AI started with training data and ‘made-up’ the parameters to get the output to work.

An analogy

Cut to a fancy Berlin restaurant, Bourcharts perhaps, in the evening which in Berlin could be anywhere from 6 to 3pm the next day. The diner scans the menu knowing he can reliably order an amazing meal. Of course he has no real understanding of how it was prepared or why the flavour, ingredients and texture of his creme brulee all work so well together. Its all about mystery, delight and convenience.

In the kitchen, the chef knows well the intricacies of the melting points of milk solids, caramelisation, balance of sweetness and texture to produce a culinery wonder. But not only can he only knock out another one, but can produce something novel for next month’s menu, because he knows after years of experimentation why it all works. The diner is limited to pointing at a menu, in eternity. But without the Chef the whole enterprise falls over.

---

I fear what is driving the interest in AI is the promise of delight with very little input. A couple of keystrokes and voila, out pops something interesting.

And this is perhaps my “gripe with the hype” of AI. Yes its amazing and yes its convenient and for a while it can dodge the classic kid’s question ‘but WHY Dad?’. But this cannot continue forever. “Why” is the ultimate question that set humans on their path.

For researchers who operate in a scientific domain, its critical to know why something is happening, and thats where real development begins. That said I’m sure some clever clogs will work out how to join them and bring meaning to AI.

And that is is why I’m still passionate about the Scientific Method and why we keep pushing something that is hard work to execute – because it answers Why.