by James Greaney
Artificial Intelligence, or AI as it’s known on the streets, is all the rage at the moment. The excitement has been fuelled by some mega acquisitions of companies such as MetaMind by Salesforce recently and of course the $600m acquisition of DeepMind by Google in early 2014. The latter shot to fame in March this year for developing AlphaGo; the algorithm that beat world Go champion Lee Sedol four times in a five game series.
The advertising industry in particular is very good at getting excited about new tech and trends, but then not very good at taking advantage of them. Big data was the latest fad, and while medicine, mining and manufacturing applied it quickly, marketing only talked about it. So, how do we prevent that happening here?
Firstly, it’s worth getting past some of the frenzied reports of what AI is going to do. As Gartner’s Hype Cycle neatly outlines, the path to the ‘plateau of productivity’ is preceded by the ‘peak of inflated expectations’, the ‘trough of disillusionment’ and ‘slope of enlightenment’. Let’s help advertising move AI towards that goal of productivity quickly.
To start, let’s acknowledge that Artificial Intelligence isn’t exactly a new thing.
Computer gaming leveraged the concept since its inception in 1951. In 1998 we saw the launch of Half-Life (remember that?) in which you are accompanied during part of the game by a security guard that helps you out. Sim City (released in 1989) boasted the capability and was pushed further in 2005 by F.E.A.R. (First Encounter Assault Recon – a first-person shooter where the player controls supernatural phenomenon and armies of cloned soldiers). In this game the AI used a planner to generate context-sensitive behaviours – the first time in a mainstream game. The enemies were capable of using the environment very cleverly, finding cover behind tables, tipping bookshelves, opening doors and crashing through windows.
So, is the AI we know today shaped by the seismic technology shifts that have happened in the last few years?
Artificial Intelligence, which is Machine Learning (i.e. machines actually learning) has been transformed by a technique called Deep Learning. It has been allowed to abandon some of the constraints previously impeding its progress (such as significant amounts of code) and has made huge leaps in voice, image and text recognition and interpretation.
Of course, these are mostly examples of Deep Learning in which the processes are guided by the data scientist.
The next level application of Deep Learning is along the lines of what was achieved by DeepMind; an algorithm that once unleashed by the data scientist, learns how to perform tasks by observing them rather than being explicitly programmed to perform them.
Our working in this area has seen us leverage AI to make leaps in standard concepts such as Market Mix Modelling. The iterations refine the parameter values within the model to help its accuracy increase by up to 50 percent in some instances. It's like we taught a child to ride a bike and they’ve quickly escalated to Olympic standard on their own.
Now we imagine taking a brief that can be reviewed, refined and reversed instantly so a creative team has five days rather than one to come up with some ideas. And in some instances, could the algorithm even come up with a couple of concepts?
Or, delivering a product to a customer before they realise they need it. Oh right, that’s happening – well done Amazon.
Businesses have embraced it, but advertising has squandered it. We should seize this opportunity instead of thinking of it as a fad. Because the signs are showing that this isn’t one.