The Pros And Cons Of Using AI In Paid Video Advertising

We are currently living through a time where AI is a regular trending topic. Despite the term Artificial Intelligence first being coined almost 70 years ago in 1955, recently there has been noticeable buzz around the developments in ‘creative’ capabilities of AI. ChatGPT (and potential rivals Sparrow and Bard) has caught media attention for its ability to write convincing dialogue based on user prompts and instructions, while DALL-E and its offshoots (DALL-E 2 and Craiyon, formerly DALL-E mini) made waves last year as an AI that can generate unique images from any text prompts.

These experiments in AI creation beg the question: where does AI fit into the creative landscape of video advertising, and should you be getting your brand on board? The fact is, we are already seeing AI involved in ad campaign creations – its role is wide reaching, from pure data analysis that can help inform a human creator, to script outlines, to full-on video and animation ‘creation’. However, these roles are seeing varying levels of success, bringing a range of pros and cons to the involvement of AI in paid video advertising.

1) Pro: Predicting a customer’s journey

Look for what computers do better than humans and you’ll discover the biggest asset of using AI in paid video advertising – that is, processing data. Huge, brain melting amounts of data. Using data to predict customer behaviour is how we get targeted advertising, and the efficiency of using AI in this task is why it’s become so ubiquitous in modern times. According to Forbes, “Research reveals that the combination of AI and Big Data technologies can automate almost 80% of all physical work, 70% of data processing, and 64% of data collection tasks.” (https://www.forbes.com/sites/anniebrown/2021/04/13/utilizing-ai-and-big-data-to-reduce-costs-and-increase-profits-in-departments-across-an-organization/?sh=6d9981286af7).

This ability to sort through and organize gigantic amounts of data has a number of direct benefits for video marketing. For one: analysis of consumer data (and more specifically, consumer response to advertising) allows brands to optimise their marketing strategies, and to develop campaigns that more accurately target who and where their customers are. Secondly: AI can more accurately track how a customer interacts with paid video advertising, including which parts of the ad they’re focussing on, how long they watch before clicking away, and where they click afterwards. And thirdly: leaving the heavy data crunching to the computers means freeing up more time for the human running the business, which is always a pro.

2) Con: The uncanny valley

Humans have a mysterious ability to tell when something isn’t quite human enough. You may be familiar with the idea of ‘the uncanny valley’ when it comes to CGI animation – that is, unreal characters attempting to look realistic but who don’t quite hit the mark. A 100% real human is natural to us, and a 100% cartoon character is unthreatening, safely in the realm of imagination. But the area in between becomes increasingly unsettling as our brains try to reconcile the image of something that’s trying to be human but is a little off, ending up in the eerie territory we call the uncanny valley.

Similarly, as AI-created content becomes more common we humans are becoming, and will become, better at determining the writing and visual style of things that aren’t quite human-made. This is an ability that develops over time – think of how lifelessly creepy the humans in the first Toy Story movie now look to us, despite not being an issue when the movie first premiered in 1997. It’s inevitable that the uncanny valley will extend beyond CGI characters and into communication styles and artwork, proven by the unsettling nature of AI-created liminal spaces (https://www.iflscience.com/why-are-liminal-spaces-creepy-66157), and the meme-able comedy of AI-generated writing samples (https://www.aiweirdness.com/). To produce effective and moving video advertising content, AI will first have to bridge the gap across the uncanny valley – a gap that is getting wider by the year as we humans become more and more attuned to the unreal.

3) Pro: An unbiased analysis of data

AI systems are able to objectively analyse the data sets they are fed. In 2018, Lexus released a commercial for their new Lexus ES with a script written entirely by an AI (https://youtu.be/l91ehyqFca8). Collaborating with IBM, their AI system ‘Watson’ analysed 15 years of award winning car and luxury brand campaigns to identify common elements of entertainment and emotion, and produced a script outline based on these points (check out the full ad here: https://youtu.be/a2_smFoqQUg). Conversely, a human analysing the same set of commercials will likely create a very different set of results, as we often look out for what we think is important, or what appeals aesthetically to our own eye, rather than processing objective connections in artistic material. Any given human’s idea of what is entertaining, emotionally affecting, or just plain ‘good’, is affected by layers and layers of personal biases, many of which can get in the way of seeing the pure factual results.

4) Con: … But an unbiased analysis of biased data

AI can only base their analyses on the data given – and if that data is coloured by human bias, so too will the results. News outlets widely covered the story a few years back of Amazon doing away with their AI recruiting engine, as it was revealed the engine had learned (from the patterns of previous hiring behaviour) to discriminate against women candidates (https://www.reuters.com/article/us-amazon-com-jobs-automation-insight-idUSKCN1MK08G).

AI will not censor its results without intervention, and is incapable of understanding tact, sensitivity, or the very subtle changes of social justice movements that can affect what is and isn’t socially acceptable to say or do. While this can be a positive in shedding light on hidden biases within an organization, it can also produce unwanted results for brands analysing external sources, such as other advertisements. In the Lexus ES script mentioned earlier, Watson was only able to develop scripts based on a set of award winning commercials – awards that are judged by humans, who may or may not be judging the nominee pool fairly themselves. A panel of judges made up of a single demographic are going to bring cultural expectations of a ‘good’ commercial that may seem unanimous, but are at best localised, and at worst offensive.

An AI trained to analyse data will only regurgitate the issues present in that data. Algorithms fed flawed examples will only create an endless echo chamber of flawed results. Whether with AI assistance or not, this ultimately means humans are still required to do the hard work to make important changes to the status quo within advertising, content creation, and the wider media industry.