The Future of Where Ideas Come From

Advertising creativity meets artificial intelligence

Things ain’t what they used to be. We used to write ads with pens and sketchbooks.

I had a dream last week where I proposed three headlines to an artificial intelligence (AI), which promptly:

  1. Displayed them within a carousel of ad templates, indicating which — because of sentence length, use of verb, tense and “Prospect Relate-Ability™” were most likely to engage our target audience
  2. The display also formatted the headlines for all manner of social and gave a score showing which would perform best on Facebook versus Twitter, etc. also versus programmatic banners, depending on time of day and audience segmentation
  3. Flagged which of my ideas were similar to headlines already out in the world
  4. Ranked the headlines for degree of positive search-engine appeal
  5. Ranked the headlines for “Award Show Worthiness™” based on a 10-year analysis of winning headlines from Cannes, the One Show, the Andys and D&AD
  6. Offered 1,600 new headline options to consider that the AI thought might be “better”

Then I woke up.

Yes, “Ads is (still) Art.” Mark Fenske had that right (additional word mine); even as our robot overlords sneak up on all the white collar jobs and intellectual leisure time pursuits. WIRED’s recent guest editing of Digg offers a slew of useful AI links.

But the business of being an ideas person is shifting dramatically, especially in advertising and marketing, whether you’re a writer, art director, designer or strategist. The way things used to be pitted the idea person alone against the world, or maybe just two of you — feet up on the table. The act of idea conception was the idea person’s solitary pursuit. The headlines didn’t write themselves. Now they might.

Consider today’s HTML5 programmatic banner ads. A human designer and copywriter assemble an initial template and list of headlines for a 728×90 animated banner ad. They may even supply a list of phrases and words. It’s up to the code to assemble those ingredients to find the best-performing combination. While not “intelligent,” you’ve got code doing a lot of iteration and decision making that was once the provence of Creative Directors and Clients.

Or consider the path almost any designer takes today when beginning a project. Pinterest, Google Image search, CompFight et al enable visual thinkers to quickly gather then curate potential visual direction. Again, it’s not (yet) “intelligent,” but there’s certainly artificial (i.e. code) help.

We live in a realm of enabled idea development.

(Need a primer? Two long-reads warrant your time: Paul Ford’s treatise “What is Code” in Bloomberg Businessweek is a great place to start. His treatise will help you grasp computational thinking — the bedrock of our age. Second, read The Great AI Awakening in The New York Times. You’ll get the basics and context for what it means for an intelligence to be created artificially, and how the leap towards automation of cognitive processing — e.g. thinking — can impact professional services jobs like those in marketing and advertising.)

To understand this shift I’m talking about, let’s understand (or maybe it’s better to say “clarify”) the creative process. What happens when an ideas person comes up with an idea? And how might that human process be augmented or replaced by technology? When I’ve taught the process of developing advertising ideas, I explain it this way:

We begin with constraints — in the form of an assignment (or creative brief). So it’s not any idea imaginable we’re after, but an idea specifically purpose-oriented. We need to understand who we’re writing headlines for, in order to influence their behavior. What topics or conditions or emotions might influence a change? Under what circumstances (media location, timing and frequency) would we best influence them? Second, how is what we’re selling — product, service, etc. — relevant, distinct, or necessary enough to evoke a behavior change in that specific audience?

Well, we’ve already got AIs helping us in this first phase of idea development. Consider Lucy, the Watson-fueled AI for market research and customer segmentation. Lucy can tell you, from across your entire addressable audience, who best to focus on. And “she” doesn’t need healthcare or take vacation days. Sorry, Planners. Or, as Godin put it recently, “The current era of on-demand, widespread looking things up offers a whole new level of insight for those that care enough to take advantage of it.” An AI such as Lucy helps you take advantage, and clarify the kind of idea you need, and where and when you need it. Granted, the impetus is still human. Someone has to start the process rolling. But it’s not hard to imagine thresholds and timers for optimal marketing circumstances, which automatically trigger the need for an advertising idea complete with bespoke creative brief.

So why not have the AI create the ideas, too? This process evolves, in my experience, across three stages: Themes, Hypothesis, Optimization. How would an AI proceed?

Let’s use the structure of headlines as our example in this scenario, since headlines aren’t highly subjective imagery and we’ve already seen how computational thinking impacts paid search writing. It’s easiest to spell out this theory if we keep it constrained to text only.

And let’s assume Lucy has informed us our best vehicle in which to influence is some kind of static advertisement (e.g. out-of-home billboard or online banner). For the sake of this discussion, let’s assume we’re advertising an update of an existing automobile, a new Volkswagen Jetta. It’s got a more powerful engine, stronger standard brakes and brighter exterior lighting, again, standard.

So, Themes.

I usually start with general topics under which an idea could sprout or inspire more general topics. In other words, let’s not try to solve the headline all at once. So, in this case, our themes might be: German engineering (it’s a VW); Power (engine, brakes); Speed; Easier to See (brighter); Greater Value (since the improvements are now standard features), Urban Life; Four (seating); and Nimble (VW reputation). That’s where I start for now. But I’d prefer to have 12–15 themes. So imagine an AI taking all it can know about the VW Jetta via every car magazine issue ever and all it can know about advertising cars via every automotive advertising solution that’s ever won an award. We get #ThemesForDays. An ideas person can already do this, of course, with assistance via search. But I don’t think it’s a stretch to imagine an AI such as Lucy offering “potent advertising themes” based on readily accessible data. We’re talking about keywords, after all.

This is the earliest stage of idea-making, the rough sorting of generalities. We’re not worried about idea quality yet. We’ll get there. At this point, we’re only worried about potential. The point in the Theme stage is quantity, and computational assistance is fantastic for volume parsing.

Hypothesis is the quintessential stage of idea-making.

It is the writing down, the stating of, a raw concept. It’s what they dramatize in Mad Men. Inspiration strikes! But ideas do not come from thin air. To continue that metaphor, the “air” from which ideas emerge is richly woven with data. So let’s take our Themes, and take them apart. For example, underneath the subject of Easier to See, we might start to gather and assemble words and phrases (data) such as: Blind, Aware, Glasses, Superpowers, Turn on the lights, Lights Camera Action, Illumination, Flashlight, Searchlight, Lifting the veil, Fog clearing away, See further, etc. We’re moving from abstract to concrete. It’s a process of reference and pattern-matching to loose assembly, vetting and editing. All in fractions of a second. So, maybe Easier to See leads to Superpowers leads to flying leads to that scene in The Incredibles where Edna Mode tells Mr. Incredible “No capes!” leads to Cyclops in the X-Men leads to that lyric “The future’s so bright I gotta wear shades.” Do we have a coherent headline yet? No, but that’s not the point. The point is hypothesizing enough — churning through enough shards of data to get to coherent-enough assembly — and get to what look like fragments of ideas. And “hypothesizing enough” is potentially what an AI could be doing for you. Call it ‘cognitive flexibility.’ This ability is at the core of how idea people work, which explains why some at Facebook are apparently “not” <cough> working on a General AI about, “taking ideas learned in one scenario and applying them in another.” In other words, Facebook says it is not outlining the architecture for code that can hypothesize enough, which in a general sense, humans would recognize as “thinking.” But why not?

Imagine an AI rattling of a thousand, “Hey, what about ________?” loose assemblies of phrases that might lead to a great headline — predicated on our inputs above. It’s my dream from earlier, minus the part where I wrote the first three headlines. So the AI suggests headline hypothesis culled from Themes supplied by market, product and audience data points. Then you, the writer, vet the tonnage and edit your favorites into actual headlines. Better yet, you instruct the AI to vet the tonnage for you, based on additional inputs. “Hey Alexa, tell Copywriter to write these in the style of Mark Fenske.” (And let’s just note it would take a programmer a lot of blood, sweat and tears to define the protocols for headline writing in the style of Mark Fenske.)

So AIs could be helping us with the initial, broad due diligence of idea creation by generating and assessing themes. In fact, they already kind of are: witness Meta, “a company that created an artificially intelligent system that reads and analyzes scientific literature, then connects insights across millions of papers,” that was acquired by the Chan Zuckerberg Initiative. The next step would be for an AI to assist in the loose knitting together, the hypothesizing, the tendering of rough phrases and sentences. (Assuming we’re still on our headline writing assignment.) Or let me put it to you this way… One of my favorite assignments to give when teaching advertising copywriting is called “Write 100 Headlines, Due Tomorrow.” And let’s assume it’s for the same VW Jetta task we’ve been talking about. In this assignment, we put the pressure on and see the idea-generating brain in action. Maybe you just start writing complete headlines. If you do, what process do you imagine your brain going through to assemble those initial headlines? I bet you’re following a Theme-to-Hypothesis operation, even if it feels split-second intuitive. Point is, to write 100 headlines by tomorrow you’re going to need to go wide first, generate quantity then hone quality. You will humor all corners of possibility, writing down any word, fragment or phrase that might lead to gold. I argue that all that gathering and hypothesizing could be done by a well-trained AI. Because what matters isn’t the tonnage. Walk into any ad agency during crunch time and you’ll see that tonnage on the floor. What matters isn’t the brute force processing. What matters are great ideas — which you don’t have quite yet. Because “great” ideas are the result of optimization.

Optimization is Creative Direction.

Optimization is the removal of the extraneous, the prioritization of the most compelling. It’s killing your darlings. Some people call this Taste. And right now it appears to be highly subjective, a skill even. How does an artificial intelligence replicate all that? How could an AI tell the difference between what could be a D&AD Gold-winning headline and any other headline? Let’s be clear, I’m talking about undermining my own job here, so please understand I theorize with the greatest respect for all the creative directors I’ve ever worked with. The business of taste is not simple or obvious or oftentimes clear. It’s about feeling as much as rational analysis. Can an AI be taught to “feel” the subjective differences between Headline A and Headline B and surmise — again, based on inputs we humans teach it — which headline is “better?”

I’m of two minds on the answer. First, we work in an era of almost unlimited distribution for ideas, abetted by vast and swift processing. Why not put as many headlines out into the world that you or your AI thinks are great, and test them against each other? In other words, why gate-keep anything; why not test it all? Why not let your audience’s reactions serve as the Creative Director? It’s a math equation, isn’t it? Time + Money = # of Ideas We’ll Run. So run ’em all. Let the robots sort out which is “best.” Why waste the human hours on optimizing?

On the other hand, as the aforementioned Fenske mentioned, “ads is art.” Today I suspect we’d likely feel amused and cheated if celebrated artwork in a museum, or a film in the theater, or a book, or maybe even the ad on that billboard turned out not to be created by an artistic (human) visionary. Fair enough, but is it okay to use an AI to write your newsletter? We’ve been taught the provence of Art is populated only by humans. We enjoy a Pixar film for its humanity, despite all the artificial intelligence that went into making it. How do we react when an AI enters this realm, and subjects us to its taste — even (and especially) if the end result is more effective than one derived by humans?

If you’re on the side of the human idea people, cheer up. Godin, in his article linked earlier, also notes, “It still takes talent and time to find the right thing in the right place at the right time.” That talent is you, an ideas person, generating themes, hypothesizing coherent concepts, and optimizing what your taste tells you is the best solution. And yet, we are moving ever faster towards defining and replicating the synaptic events that “create” taste, and giving those skills and subjectivity to artificial intelligence.

Time to wake up.

Originally published in the Useful Lunacy blog.