At the meeting point of marketing consulting, data technology and creativity, Artefact is a data innovation agency using data and artificial intelligence to revolutionize customer experience.
A few years ago, we often said that Big Data was like sex for teenagers: everyone talks about it, everyone says he does it, but no one really knows what it is.
Today, Artificial Intelligence is the new Big Data.
Let’s pursue with this metaphor: some teenagers actually know what they are talking about, because they’ve had the opportunity to practise. They are “the cool kids”: the prom queen, the school quarterback… But they are all hiding details that could hurt their popularity.
Who are those cool kids for AI? Google, Amazon, Facebook, IBM, Microsoft… and a bunch of startups born from this sector’s specialized branch. Early 21st century tech giants are working really hard to finally “get lucky”; they talk a lot, especially about the bright side of things – even if sometimes some pitfalls are revealed.
But for those who don’t spend billions on R&D, how to understand AI?
This is a complex question and for good reason: there is no clear definition of “Artificial Intelligence” since there is no definition of “intelligence” at all –“What is intelligence?” is typically a philosophical test subject of the baccalaureate. That’s why there are as many answers as nonsense in the debates to know if this algorithm or business case is AI, or machine learning, or data mining, or natural language processing with a bit of internet-of-things but careful it’s not AI…
So let’s take a very pragmatic approach by leaning on Marvin Minsky, one of the field founders: artificial intelligence is a computer program that engages in tasks that are, for the moment, performed more satisfactorily by human beings.
AI is not a particular technology or a precise algorithm. It’s the name of every program at the service of problems we want to solve using machines instead of humans. This pragmatic vision has the advantage of centralizing the debate on the end goal, what we seek to accomplish rather than the mean used.
Marketing has used, uses and will keep using artificial intelligence
A few years ago, to build marketing campaigns, we would form a focus group of people who would think hard, assisted by a couple of TNS survey, marketing study and such, to define the 3-4 segments to target.
Nowadays, programs analyse big amounts of quantitative data, and discover automatically this segmentation – sometimes even revealing new relevant segments.
We had a complex issue requiring attention and human intelligence to be well solved. Today an algorithm can take care of it for us – and sometimes handles it even better than we would have. That’s Artificial Intelligence.
Up until last year, we defined manually the different steps in campaigns, what channels to use and what messages to address, to whom and when. Now, a software makes those choices automatically and finds the best “customer journey” for each type of persona.
What algorithm was used has no importance. It can be random drills. It can be based on a noise canceling auto encoder (it’s a kind of “deep learning”). It can use an inference rules generator. Or else it can also be about dynamic programmation, a 70 year old technique, well before Artificial Intelligence existed – which doesn’t prevent it from being more brilliant than some other more recent techniques calling themselves “AI”. It can even simply have stored the right answers in memory and return them. As long as it fixes the problem we didn’t know how to solve before without using the human brain, it’s AI.
The above examples are two real and current applications of AI to marketing – targeting, and campaign diffusion – that are already bearing fruit. There were others examples and there will be more. Which ones? When and how to start for a CMO?
Here were are talking about weak AI, that is to say focused on a specific task. Strong AI, able to react in every situation as well as a human would, is another matter.
Moore’s law and GPUs
Let’s start with the question of when: AI, it’s now, and even a little before. AI has been evolving since the end of the 50’s and has had ups and downs – as a matter of fact we are talking about “AI’s winters” for the 70-80’s and 90’s-2000’s decades. That being said, since the beginning of 2000s, it is experiencing a continuous boom and ever more applications. The tone has been set in 1997 with the victory of Deep Blue against Gary Kasparov at chess. In 2003, Amazon releases its intelligent product recommendation engine (responsible for 35% of their turn over). In 2006, Google Translate starts to be used industrially for automated translation. In 2011, IBM’s Watson wins Jeopardy, the famous american game, against human champions. In 2016, Google’s AlphaGo defeats Lee Seedol, Go global champion–one of the rare games still resisting computer domination.
What are the reasons of this ascension? 3 major factors:
- Parallelisations and data mining
New statistic learning techniques imagined and developed in the 80’s have been able to spread and evolve at the end of the 90’s, thanks to the ever increasing volume of data available to train them, and to an increasing simplicity of parallel processing, software (hadoop, then spark for example) as well as material (infrastructure-as-a-service, Amazon’s cloud then Google, Microsoft, etc.).
- Increasing calculating power.
Some algorithms imagined from the 60s to the 90s were not very usable on cases of complex uses because they were simply too slow. The famous Moore’s law – the number of transistors in an integrated circuit doubles approximately every 2 years – released them from the speed constraint. Note that this law has reached its limits in recent years and is no longer relevant, as also announced by Moore himself.
- A customized “hardware”
Graphical Processing Units (GPUs), processors originally designed for games and video, turned out to be extremely performant to run some machine learning algorithms–especially the ones designated as “deep learning”–up to a 1000 times faster than classic processors thanks to their specialised architecture.
If we still needed convincing of this new trend, the following two data points should finish to convince you. First, in 2016 the financial investment is nearly 10 times more important than in 2012. Second, nVidia, a NASDAQ listed company that produces graphics cards equipped with GPUs, was previously mainly known by gamers. Its market capitalization has been multiplied by 5 in 18 months.
Marketing’s “What’s next?” – Towards artificial creativity
They say automatisation and learning technologies have a lot to offer to marketing (and to every sectors)–but how?
We hear people talking about augmented reality (A.R.). Today with a smartphone’s camera and tomorrow with specialised glasses – after Google Glass’s failed tentative, Snapchat is trying out the experience again. The concept popularised by Pokémon Go could be democratised. Visual recognition algorithms will allow apps to enrich the visual field; for example by filming a dress with an e-commerce app, one could find its reference; looking in one direction and asking where the nearest ATM is, it will start to flash or an arrow will indicate it (there will be more ATMs that said); each app will have its use for A.R. And for the advertisement, after Adwords there was the display, then the native and the video, then the push notification in-app – the next move would be the push A.R.?
We also hear talking about dynamic emotional marketing: measuring tools are proliferating to know where and when you cast attention. Soon we’ll be able to follow your sight while you watch a video on mobile. Adding this to other signals, it will be soon possible to learn what you like and don’t like during a tv ad depending on your state of mind at the moment, and adapt in real time depending on the individuals: it would be DCO’s next step (Dynamic Creative Optimization).
That being said, those perspectives remain fictional for the time being–few realisations are concrete to date. The main marketing domain where AI already started to prove itself is creativity.
A machine paints the next Rembrandt. Coca-Cola creates his ads with its algorithms. McCann Japan hires an artificial Creation Director. A chatbot predicts future artistic concepts. Of course, there is a lot of ad and buzz effect in all this, but the trend is marked.
What are the realistic projects? Which projects are really solid? What in this proliferation of ideas and innovation will fail or succeed? There is only one true way of knowing it. Take the risk and innovate – possibly by surrounding yourself with experts. By knowing how to leverage Google, Amazon, Facebook, IBM, Microsoft technologies and the APIs that they make available (IBM Watson, Google Cloud Machine Learning, Amazon ML, etc.), or by working with one of the many startups that embark on this type of subject.
On the other hand, developing your own AI technology internally is probably not the right way. Unless the technology you want to work on is completely fundamental to your business (for example if you’re a visual recognition startup), there’s no point in trying to do better than the tech giants who invest heavily in AI.
Above all, let’s be visionaries and dreamers in keynotes – and pragmatic and realistic when launching projects. Poorly controlled ambition becomes the enemy of success; the imagination about AI’s potential is bigger than the Nile during the deluge.
In the 1960s, researchers in Artificial Intelligence predicted the advent of strong AI, which is better than humans without assistance, for the 90s. Epic fail. But in 2000 computers could still read written checks, detect complex cases of bank fraud and beat humans in chess.
In the years 2010, some announce again the end of the work and the complete replacement of the man by programs in 30 years. The story allows to doubt this aggressive timeline. But AI will probably be able to drive your car, do your shopping, find the jobs behind the failures and successes of your digital campaigns, and create the slogan for your next product – which will probably be imprinted IA. In 2040, the machines will not think by themselves but they will market themselves.
This article was first published on the french media Viuz.
Get in touch with our expert team to know more about Artefact’s marketing and artificial intelligence approach.