Whether it is in media planning or writing for the web, it is clear that software is going to take on lots of the work. There’s a ton of complexity to manage, a mass of stuff to be generated, and lots of it is solvable by relatively simple algorithms.
All of which means we will soon have people in the bars of Soho and Shoreditch talking about "local maxima" – by optimising any kind of system, taking individual elements and tweaking them towards improvement, will get you success. For instance,
A/B testing of elements with lots of users will quickly show you how to make things easier for people. You will climb a hill of betterness.
But eventually – inevitably – diminishing returns will set in and you will get to the top of that hill. You won’t be able to tweak things better. You will have reached the "local maximum". However, it is possible you won’t have reached the "global maximum" – the best possible solution to your problem. You’re stuck on a small hill, gazing at a massive mountain of betterness, and you’ll have to go down and start somewhere else to get there (apologies to any mathematicians out there for the severe mangling).
'Whether it’s in media planning or writing for the web, it’s clear software is going to take on lots of the work'
This is a magnification of a problem that you get with old-fashioned ad testing. But, as more of this stuff gets sped up, we will need to keep an eye on it. That is where people come in. We can collaborate with the machine, reaching down like gods to pluck the software from the small hills of local maxima, and plonking it down somewhere more promising before it starts iterating and tweaking again.
This sounds silly, I know, but as we work with increasingly opaque systems – big data, optimisation algorithms, all that stuff – we need new metaphors and language to understand these things.
Russell Davies is a creative director at Government Digital Service