APG Noisy Thinking | Which bits of your job are going to be taken over by robots, and what are you g
You can’t cast your eye over an on-line industry rag, or new thinking from economists and pundits, without hearing sweeping statements about the implications of AI for our industry and the future of the world.
There’s rather less detail on what the march of the robots - if there is such a thing - means for our actual jobs; the things we do day-to-day.
So we asked some strategists what they thought and what, if anything, they plan to do. The evening was fascinating for the variety of their vision and prescription; for the practical tips and tools they suggested, and for the vast array of nuggets on offer that help us think our own way through this highly over-imagined future, by offering some smart guidance and new, positive thinking. We invited Amelia Torode; she of the newly launched fast, flat, flexible Fawnbrake Collective to speak first. In a classic piece of intellectual generosity, she demurred and suggested Oli Feldwick should speak, as she had judged his dissertation for the IPA Excellence Programme and thought he had an exceptional point of view. Introducing Oli, she noted that a good way to think about how to manage the introduction of various versions of AI is to isolate the ‘pain points’ and work backwards from them. If we were starting afresh, how would we re-structure and what would we do as a result? Agencies are very bad embedding new tech and we tend to sprinkle it on the outside rather than making it fundamental to what we do.
What we should be doing is using tech to outsource the bits of our job that feel automated anyway - the tasks that take up too much time and deliver little of real value. And we need to work out how we are going to work with computers, to make us better at what we do. Amelia was first of several to reference the Alpha Go challenge, noting that the person playing the computer may have been been beaten by it, but got better and better at the game and more creative in his thinking, and jumped up the world rankings as a result. Oliver Feldwick, Head of Digital Strategy at CHI&Partners usefully outlined his central thesis (you can read the whole article here) and pulled out some implications for the jobbing planner.
First he reminded us of the central narrative that robots are stealing our jobs because they are faster and more efficient. But his thesis is that we need collaboration between humans and machines, and understanding how that collaboration can make us better planners. The context is an exponential demand for creativity; fewer people producing more stuff in less time, for less money, and where the demand is to work harder (‘or shitter’). He coined the phrase ‘The age of ‘cyborg creativity’’ suggests 4 ways that we can respond:
Automating the simple tasks so that we can focus on the cool stuff
Assisted craft at scale: The fundamental tool for a modern agency is to do it bigger and better
Provide greater depth of insight. The magic of machines is to generate lots of insIghts that humans can’t
Higher order cyborg creativity – making the machine a strategic partner that pushes us to think better.
Oli’s view is that ideally we should be doing all 4 in tandem. So what will it mean practically? Basically it’s all about liberating time as a resource and spending our time on bigger, better problems. And he came up with some useful examples: Assisted craft at scale is manifest in AI assisted journalism. When done well and covering real-time sporting events and elections for example, it’s hard to choose between human and robot copy. An example of a tool of mass creation is 7 million computer generated pack designs for Nutella.
David Bowie used Verbasizer to help him write his lyrics by spewing out loads and loads of words.
But how do we gain a new depth of AI assisted insight? Well we could tag all the elements of our creative development processes on different briefs to help automate the pitch process. Or we could plug all the consumer data in a category into chatbot that’s learned about consumers and has interacted with them, thereby raising the extraordinary spectre of moderated AI focus groups…
And higher order AI thinking can work too. Apparently humans prefer AI assisted art works because they are fresher and more different. I’m not convinced Leonardo da Vinci enthusiasts would entirely agree with this but it’s evidence of AI inspired creativity’s ability to manipulate human responses effectively. So what might this mean for agencies’ competitive creative credentials? We could develop an AI assisted house style that competed against other agencies’ in pitches: Hire our AI – it’s doing really effective work.
As Oli noted the concept and function of a creative team has barely changed in 40 years, so perhaps it’s time to evolve. And we would welcome the chance as planners to use these tools to free up our time and do what we should be doing: getting out there and talking to and observing real people. So why hasn’t it happened already? In short, it’s a bit to early; the biggest disruption is happening behind the scenes; frankly we don’t really get it yet, and ….it might all be bullshit….. So where does that leave us? What should we be doing to work with those robots to our advantage?
Data is the oil for everything we do. We’re wasting it; we should be using it better
We should be hacking our own planner bots to write better briefs
Network agencies should be using AI – it’s a new killer resource
And let’s create our own blueprint for the agency of the future
So a broadly positive, constructive prescription from Oli. And we look forward to him making it happen in his own backyard…. Ronnie Crosbie who is Planning Partner at Saatchi & Saatchi had a highly imaginative vision for the way AI could work in tandem with a planner. Funny, soul searching, practical and genuinely experimental, this is his vision of working with a female robot called …..Alan. This is the text of Ronnie’s speech
For the purposes of answering the question: ‘Precisely which bits of your job could be done by a robot and what can you do about it?' I’ve reimagined a past project, as if an AI Robots existed. My robot is called Alan.
And this is our story. I got selected for a special planning project on P&G because I knew nothing. Magnus our MD, now Global President of Saatchi, explained to me that my lack of knowledge on the brand, the product, the process and the protocol were all seen as the key assets for a successful outcome. Of course I’d never been prouder, never had my lack of knowledge counted for so much. But the point is they want a fresh point of view. If Alan my AI robot were around at in this time, I would have whispered, “Alan given you know everything in the universe, I best handle this one.” QUESTION ONE: IF ALAN KNEW THE ‘OPTIMAL’ THINKING FOR THIS CLIENT BASED ON SEQUENTIAL LEARNING, WOULD I HAVE TO ERASE HER, IN EFFECT KILL HER AND START AGAIN AT THIS POINT AND WOULD I HAVE THE MORAL RIGHT TO DO SO? Next day our elite squad of advertising experts who knew nothing, and presumably all their robots, that knew everything, flew out to Geneva to receive the brief. Of course, I could have sent Alan alone, like a physical extension of Video Conferencing, but it was a first meeting and people like to look people in the eye. QUESTION TWO: PEOPLE BUY FROM ROBOTS ALL THE TIME, THEY HAVE RELATIONSHIPS WITH ROBOTS. I’VE NEVER MET ANYONE FROM AMAZON. WILL PEOPLE BUY SOMETHING THEY CAN’T SEE AND IS COMPLETELY NEW FROM ROBOTS? ‘LAZER EYE SURGERY WALK IN AND JUST LIE DOWN’…. On the flight there, Alan would have told me that Pampers Diapers are an $11.5 billion brand that lives in the afterglow of new-born babies. That their markets around the world are 100% focused on this product, spending a huge portion of their resources maintaining market share in this highly profitable sector. This would have made me like Alan, despite knowing what happened to the guy in Ex Machina. He’d then tell me that, having checked the attendees and their profiles on LinkedIn, he’d concluded we would be getting briefed by the Pampers Wipes team, the unsupported product that no market has ever put any money behind; that was silently at risk of being delisted. This would have made me irrationally angry at Alan. He’d pick this up via my body movements, so I wouldn’t have to tell him. QUESTION THREE: EMPATHY. WOULD ALAN KNOW I WAS ANGRY OR THAT HAD WIND OR WAS CONFUSED? LIP READING EVEN BY AI’S IS ONLY 50% RIGHT AS SOME WORDS ARE SO SIMILAR: SAM, PAN, RAM, DRAM. AND IF HE DID, WOULD HE GIVE A SHIT? Alan would record the whole meeting and what everyone said. I’d ask Alan to delete what the junior client said, then cross reference everything that the main client said that was repeated by everyone else in the room. Alan would then state the brief: Drive awareness of Pampers Wipes by creating communications the markets will run, despite never having run any work they’d been given before. Alan would point out that it is illogical to invest money on a tiny brand that offers very little profit, when you have a really big one that does. QUESTION FOUR: HOW WOULD AN AI DEAL WITH A LOGICAL BUT ILLOGICAL BRIEF? DO WE ASK IT TO IGNORE THIS? THEN COULD WE ASK IT TO IGNORE ANYTHING ELSE WE DON’T LIKE? On the flight back, Alan would have tried to cheer me up by pointing out that babies are highly emotive topic for humans. I would then help Alan’s learning engine by saying “not when they are covered in shit.” Alan would then review the most effective ads against mums over time and conclude they have a baby, in an idyllic setting and they are free/happy/learning/eating well due to the client diaper, shampoo, toy, food, which makes mum happy. He’d then offer to brief the creatives for me, as we have a tight production schedule. I’d ask Alan to prioritise idea over cost and timings. QUESTION FIVE: HOW DOES AN AI PRIORITISE THE VALUE OF AN IRRATIONAL EMOTIONAL IDEA OVER COSTS AND TIMINGS. CAN IT BELIVE IN THE POWER OF AN IDEA? I’d ask Alan to do a global analysis of mums with kids under 3 years of age. Alan would then look simultaneously at millions of mums all around the world on their web and security cameras via a start-up company that one of last year’s grads began as a side project last year. Alan would recognize that their erratic behaviour would not match that of MUMS living idyllic lives, mums are mostly angry, tiered and confused about why they spend their lives cleaning up everyone else’s shit. Facial recognition software would surmise that the key emotions associated with wiping a baby’s bum: apprehension, denial, disgust, concern, followed by a short-lived sense of satisfaction. QUESTION SIX: ALAN COULD COLLECT MASSIVE AMOUNTS OF DATA, BUT WHEN WOULD HE HAVE ENOUGH AND HOW WOULD HE RECOGNISE A UNQIUE INSIGHT? YOUNG MEN LIKE FOOTBALL. CHILDREN LIKE FOOD THAT’S SWEET. MUMS GO ON THE INTERNET at different times in different countries because their THEIR KIDS GO TO BED at different times. Next we’d think about the product in this context. Alan 'summaries' the product's key attributes from the 3-hour science induction session. Alan would recall instantly. 1. Wipes have unique micro pores that pick up poo better than wipes without them 2. they are gentler than water 3. conveniently packaged. Summarise, Alan. ‘They clean poo away’. Access 5 WHY’S Protocol Alan. Why: Because poo isn’t nice for babies. Why: because poo is messy and disgusting. Why: because it smells and will get everywhere if you don’t. Why: because it doesn’t just disappear. Brilliant Alan. Pampers Wipes make the bad stuff disappear. QUESTION SEVEN: COULD AI THINK AT SUCH A STUPIDLY SIMPLE BENIGN LEVEL? WE ARE HALF NURTURE AND HALF NATURE. THIS LEVEL OF THINKING IS A NATURAL TALENT, AI’s WILL ONLY EVER BE NURTURE. Alan, isolate occasions when people who use wipes, don’t have them. List key emotions. PANIC, FEAR, followed by an over whelming desire to shout at the next person who comes into the room. Alan writes the brief: Insight: Mums fear mess. SMP: Pampers Wipes make the bad stuff disappear. Tone: Real world. You brief the creatives, I’ll be working, ahem, from home for this afternoon. QUESTION EIGHT: WOULD ALAN GRASS ME UP FOR SKIVING? CAN YOU EVER TRUST A ROBOT? Four weeks later we’re on a plane again back to Geneva. The ECD has her amazingly expensive next generation AI. I’ve left Alan in the office. This is the third time we’ve presented the same creative and Alan keeps pointing this out, whilst the rest of us keep pretending it’s changed. Plus, Alan is also still having difficulty understanding what a creative idea is. Eat Fresh. That’s not fresh. Taste the Rainbow. It’s light. It gives you wings. No. Maybe She’s Born with It, Maybe It’s Maybelline. It’s Genetics. QUESTION NINE: A LOT OF HUMOUR (JOHN CLEESE) IS BASED ON TAKING A SITUATION AND CHANGING THE CONTEXT. IT MAKES SENSE BUT DOESN’T MAKE SENSE AND THAT’S WHAT MAKES IT FUNNY AND MEMORABLE. RAINBOW COLOURS AS LIGHT OR FLAVOUR. HOW CAN AN AI MAKE SENSE OF SOMETHING THAT DELIBERATELY MAKES NO SENSE AND HAS TO BE UNIQUE EVERY TIME? The meeting goes well. They have the product demonstration they want and are convinced the idea we keep presenting will die in research. Alan arranged focus groups by selecting a company based on their 5-star rating on Trust Pilot. I cancel this and arrange them through a company that is a ‘friend of the agency’, but I tell Alan it is because they are cheaper. The focus groups go well. Our idea is approved. It involves filming babies faces in super slow-mo HD whilst they Poo to the sound track of Space Odyssey. Alan tells me this no job for any sentient species and that I need to evolve. I say I don’t. Alan asks if I can spell Odyssey. I say I can’t, but I don’t need to. You can spell, I’m an ideas man.
And finally. The inimitable Russell Davies. He had 4 thoughts on the subject. Pity the poor APG hack who attempts to get down his thinking (or a version of it) sans music, hilarious video and Russell’s own speech. Luckily it’ll be available on video, on the APG site, quite soon.
In the meantime, here goes.
First off he believes that the impact of AI will not be a big, dramatic thing for us, and in any case it’s not about us and we don’t have the data. People who are genuinely experts in AI will not go anywhere near our industry because they can earn huge piles of cash elsewhere.
So if we do want to play with AI then we should get on with it using open source tools, creative tools and consumer tools. Russell demonstrated the kind of thing we could get involved with by showing a video of a bloke in a self drive car that he had created using downloadable AI from the internet. Not that hard apparently and obviously for us it’s not cars, but that level of adaptable creativity.
An equivalent example is author Robin Sloan and his use of recurrent neural networks – creating novels using an AI which works on the character level and auto completes your sentence. A bit like having an eccentric parrot on your shoulder. And of course the really interesting question here is whether you can – do this with and for planners? There are 40,000 words of APG papers now available to members on our site. And Russell applied the approach to our database of cases by asking an AI to autocomplete sentences like What is a Brand….the results are hilarious and you’ll have to wait for the video to see them.
But more seriously, to train a network takes hours, and actually planning as a discipline, and the words we use, are not that distinctive. So sadly (Or thankfully?) the results you get are a lot less distinctive than pulp science fiction. All of which begs the question whether this exercise is either worthwhile or interesting? (Maybe Oli could have a go?)
So what we need is some useful models, something helpful to describe AI; and it’s not a robotic generalist. Workable AI is very specific. Alpha Go is great but it can’t do anything except Alpha Go and it can only play the game. It can’t go and fetch it for you. And some things like folding towels are incredibly hard for robots.
And washing machines are really good at washing, but you have to take the washing to them So let’s try and think about some really specific problems that we want to solve. And note that it won’t get really interesting until it gets really familiar and boring. Like the first novel written on a word processor by Len Deighton (although in fact he wrote it long hand and his female PA did all the typing). Another unsung hero of progress. And as Russell pointed out, since AI is based on the examination of past data, most AI is inherently sexist and racist. So Russell’s conclusion is that AI can do what we are already doing but in a different, much more specific way. We can use AI to do grunt work and use our judgement to make use of it. So when you’re generating concepts and words for a workshop you get the machine to do it, and you decide what to so with the output. Or you can get a machine to help generate loads of ad taglines and use Facebook to work out which ones are effective. It’s not hard and it’s pretty quick. There is some bad news, however and it’s hard work. We will really have to understand properly what it is is that we do and the tenets of our work that the machine needs to follow. How does advertising work? What is a successful brand? And on that note, cue some really hard thinking by our community. After all, that’s our job, isn’t it?
Sarah Newman 7.11.17
Nice date. Palindromic.
Further Reading: it seems that the Chief Tech and innovation office of Accenture agrees with our speakers (go here)