I have to say something. The siren song of Artificial Intelligence has got so strong that it needs answering. In the last week I have attended 2 events explicitly built around Robots and Artificial Intelligence – so I want to talk about them. But mostly I want to talk about the overclaims of AI. And how we can calibrate the potential of smart software to something that will actually be useful.
When I was a junior planner a very long time ago, the Japanese fifth generation project was in full flow. The Japanese wanted to steal a technical lead. Only it didn’t work. It wasn’t because the computers weren’t fast enough. But the problems that could be solved were limited, had to be contained. From that we got rule-based expert systems. But we stopped calling them intelligent. I used to type AI programme onto my BBC B computer – the code took up a few K – (you might want to google what a K is to remind yourself that you can pack a lot of code in a few thousand bytes). Some of it was very clever. I had a program that you trained to predict the weather based on temperature and barometric pressure and whether it had rained. It worked pretty well But it wasn’t remotely intelligent. But that didn’t stop the AI evangelists at the time bragging that a thermostat was intelligent because it ‘knew’ when to switch on. Only thermostats intelligent at all. Just because they were linked to a clock didn’t mean they ‘knew’ the time either. AI was muddled with consciousness. And as I started to work in direct marketing I taught myself advanced Excel skills. I was the planner on Microsoft after al and worked on the launches of Office and Windows NT. So I read the manual on the train home every night. And started to design spreadsheets that told you what media ads to book based on historic response rates. And the predictable uplift in response. And the reduction in cost per lead. The year was 1994. And Excel wasn’t intelligent either. It was the Solver function which you can still find in Excel if you dig deep into the data menu. Then the neural network sales guys turned up. One of them was actually selling them out of the box at the front of the event for £1000. You could have more variables than data points and there was no technical support because it was so easy to use. This personal neural network could solve any marketing problem you could throw at it. So I learned to be exceedingly sceptical about the claims of artificial intelligence. A generation later with the advent of self driving cars back come the experts again.
What passes for Artificial Intelligence is one of a handful of types of software activity. Pattern recognition, rule based choices, optimisation or automation. All of these within limited pre-defined parameters. but you wouldn’t think that to hear the AI evangelists bang on about how many jobs are going to go in the next 10 years. And how many of us are going to have to retrain to teach machines to do our old jobs. None of these involve intelligence. It isn’t needed. But a lot of our work isn’t particularly intelligent. And its automation is going to make differences. Not always for the better.
So the Account Planning Group in London this week offered us 4 speakers talking about the impact of artificial intelligence on account planning. Amelia Torode and Paul Feldwick based on his IPA paper gave a utopian take on how various automation and optimisation tools might be deployed. I have misgivings on whether agencies would adopt even the simplest of tools. Particularly if they have to pay for them. I had a junior planner design his own timesheet system in Access to save everyone time. He was the only one in the agency using it. Between the two of us, we build a database which would capture all campaign results so the agency developed a memory. Only the suits wouldn’t fill it in or ask the client for the results. Then next wheeze was to talk to the accounts department to build a system for account handler to predict the hours and internal costs for campaigns and to complete what actually happened. Finance didn’t see the point since it was they not the suits who determined productivity. And these were internal projects which would have cost the agencies nothing to implement. What I am saying is that culturally the agencies I have worked for resisted automation.
Ronnie Crosbie gave a more sceptical response ingeniously because AI’s weren’t human enough and couldn’t cope with the contradictions and compromises inherence in human activity. My favourite was the program getting flummoxed when creative was represented to the client 3 times on the basis that each campaign was different and had moved on when really it was the same. We’ve all done that!
Russell Davies concluded the evening by describing some of the simplest things which AI simply couldn’t do (like fold jumpers). And that it was in very narrowly defined fields where it would and could make an impact.
For me the big learning from the evening was the fear that AI would take the most advanced and cleverest bits of the job away when the evidence was that this was the last thing it would do. Because AI isn’t clever enough. And that unless the application was a mass market application no agency would pay for it. So that disposes of the free or low-cost option. So mediocre mass market problems which could be drawn into the agency – like an app for sorting out people’s calendars – that sort of thing. That’s not trivial. There used to be secretaries to do that kind of work. but not any more. Now we have to do that even if there is no budget to train us in using the software. But hardly the high-end clever stuff that puts Elon Musk on Mars.
That leads me to the second event of the week The We are Robots music exhibition at the end of last week just off Brick Lane. There was a panel including Martyn Ware who was a founder member of Human League and Heaven 17. He was sounding off at one of the other panellists whose company had an AI application. She was explaining how there was nothing to be worried about since the AI didn’t write music. It just automated mixing and mastering. That was exactly the issue replied Martyn. When you go on a Spotify playlist the songs you are listening to aren’t there because of multiple plays but because companies paid to list them at the top. And the library songs and sound alike artists were being mastered by those AI programs to make it more likely that you would discover and listen to new music. So AIs were being used to reduce differentials and make every new hit song sound like previous hit songs. This to me is the biggest threat of AI. Which is not to make something the best. But better by mimicking something else. Making everything else the same. the program doesn’t know what Good is – it just knows how to optimise a mix to sound like something else.
So the idea of an ad agency pitching with its own AI would be a nightmare for clients. Since each AI would effortlessly try to produce a strategy it thinks the client would buy. Result all of the presentations from the competing agencies would look the same. It would be more interesting to have transparency so each AI could see what the others were doing and had to try to beat the competitor strategies. But somehow I doubt that transparency will happen!
Let me finish with another presentation – it has been quite a week. This time from Tony Brignull -any shortlist of the best living copywriters would include him. He was talking to a group of students at the SCA advertising school in Brixton. He asked them What is the opposite of creativity? It’s not destruction. It’s copying. Endless mimicking is cheap and improves our lives no end. But it reduces differentials as it does so. Because that’s what copying does – it’s convergent thinking. Creativity is a way to jump out of the loop to somewhere new. And that the even go playing, chess playing, traffic managing machines. Can’t do. And won’t be doing for a long time yet.
Designed by Matthew Pattman