Humans have been harnessing the power of machines for aeons. From early agricultural contraptions to modern day manufacturing process lines; we have benefitted from using these tools to make our outputs harder, better, faster, stronger… (I hope the irony of using lyrics from a song by two men dressed as robots isn’t lost on you).
But what about the modern day machine, fuelled by data and computing power as opposed to steam or petrol? In our recent IAB SEA+India New Media Working Group meeting we discussed the application of Artificial Intelligence (AI), machine learning and automation in marketing, with a view to debating some of the harder questions that come with the use of such technologies.
Looking at use cases like automated bidding in SA360, machine learning lookalike modelling or Facebook’s optimisation algorithm we can safely say that machines have improved efficiency in the delivery of digital media.
Machines are finding quicker ways to help us get results, but as so many processes and outcomes are now so easily optimised we need to bring that level of efficiency to delivering business outcomes as opposed to just media success. Modelling value, propensities and building predictive marketing models are the next frontiers to go mainstream; and whilst pockets of excellence exist today there is more that can and will be done in this space.
Yes, optimising in a media channel for sales may see net positive outcomes in revenue for a business that is end to end digital; but what about bridging the online/offline challenge or using machine learning to qualify & drive prospects through the entire marketing funnel (and engage with them in the perfect way at any point in their customer journey)?
There is no doubt that machines have improved our ways of working in the industry, but now the focus seems to be on making machines more effective and delivering impact where it matters most to the bottom line, as opposed to an efficiency in process or workflow.
What about when things go wrong? What if a machine learning model never gets it right or causes overspend, campaign issues and is detrimental to achieving success?
Our group unanimously concluded that there is a growing need for a layer of accountability beyond the machine. Of course, in the agency world if a buying platform AI model doesn’t perform then it is inevitably the agency stakeholders that are accountable, but perhaps on growing occasions there is no identifiable cause for the underperformance. Instances of machines not doing what we intend them to do are common (see examples from Amazon & Microsoft), but we have to ask if the machines were set up for success and given every possible piece of programming to ensure positive and expected outcomes.
We need to always recognise that until there is a truly sentient machine, humans have to take responsibility for the actions of the technology. We are still the ones who set the rules, create the parameters and build the decisioning logic. We must give the machine the appropriate guidance and support in order to execute on its objectives; and thus we are responsible for the outcomes of the technology.
Creativity as it transpires means something different to everyone. Some in our discussion posited that just because something is new or original does not necessarily mean it is creative. For example: if a machine is shown all the paintings of Van Gogh and then creates a new picture in the same exact style, you could argue this isn’t creativity. The machine was operating solely within set parameters and the creativity is calculated and ‘programmed’.
But isn’t this how we operate as humans? We can only use what is in our world, our universe for inspiration, as the genesis of human creativity has to come from the known (otherwise how does the spark, the initial idea come to exist in our minds?). It is indeed true that we often don’t know what has caused the creativity, and the ‘Eureka!’ moment often is hard to trace back to the root instigator or the multiple layers of influence that led to the moment of creativity.
CAN MACHINES LEARN TO BE FUNNY?
One ad creative example was tabled and the notion proposed that a machine could in no way have delivered such reactivity, humour and relevancy. The Marmite ad during the 2019 Ashes series that was all parts tongue-in-cheek, topical and just blatantly funny is a demonstration of the powers of human creativity and how context really is critical for landing something resonant or impactful.
Yes, a machine could make an attempt at humour, but would it be culturally respective, topically relevant and land in the same way that this ad did? How many variables, parameters or other inputs would a machine need in order to pull together something so good? How long would it take? Most pertinently, could a machine ever produce something this good? The group didn’t seem to think so.
AI can now write Search ad copies that are more relevant for user queries, and dynamic creative optimisation can create combinations of words & images matched to data & intent signals from a consumer; but the jury is still out on whether achieving true creativity is possible with machines. After all, it all depends on who you ask; creativity is in the eye of the beholder.