An algorithm is only as good as its trainer

AI is the talk of the marketing world, but marketers must learn algorithms are only as good as their trainers, argues Alchemy One's Joel Trethowan.

The difference between a cyborg and an android is that androids are robots made to look like humans, whereas cyborgs are humans augmented by technology. You see cyborgs every day. Listening to podcasts to improve their knowledge, watching YouTube to learn how to cook better – people, today, are cyborgs – conscious beings attached to technology that makes them more intelligent.

In order to truly build connections with consumers, brands need to be more cyborg. They need to drop the data-evangelist, algorithms-will-save-us programming they’ve been given and switch to a human-led, intuition-centric model where data is a foundation for brilliance, rather than an end-to-end approach.

Data is great. Data is essential. No one is saying we don’t need data. But data does some things well, and some things not so well.

Correlation v Connection

Data is amazing at correlation. It is incredible when you need to quickly pull together information from large or disparate sources and compare them against each other. Analytics tools can quickly provide beautiful visualisations showing you (or your clients) key metrics.

Data can’t tell you when a timely post and principled stand will have a massive impact on your brand’s perception and ability to drive a real connection with consumers. It can’t tap into the cultural zeitgeist and explain to you the emotive, social or economic attitudes that provide the canvas on which your campaign sits. And it certainly can’t tell you how to stay relevant within those boundaries.

Insight v Intuition

In his TED Talk from 2015, data scientist Sebastian Wernicke sums this one up nicely. Two companies, Amazon and Netflix, set out to make a TV series based on the data they have. Amazon holds a competition for show ideas, builds eight pilots and screens them to viewers, drawing insights on view counts, durations, play / pause / rewind metrics. Based on this they build a sitcom about four Senators called Alpha House.

Netflix, on the other hand, looks back through its platform data to try to uncover what makes the viewers tick. They pull together a bunch of data on shows, actors, genres, and using the intuition of their producers and content executives, they create a drama series about a single US Senator called Frank Underwood. Can you guess the show?

In problem solving, you have two parts – the first part, where you break the problem down into its components to really see what’s going on, and the second part, where you put everything back together again. As Wernicke says: “The crucial thing is that data and data analysis is only good for the first part…It’s not suited to put those pieces back together again, and then to come to a conclusion.”

Forensics v Feeling

How did Nike’s Kaepernick ad make you feel when you first saw it? Chances are you either felt a swelling in your chest from a mixture of pride and solidarity, or the same feeling driven by anger (as you likely rushed out to burn your trainers).

But you felt something.

This campaign was not about data (though I’m sure they’ve been watching sales pretty closely the last few days). It wasn’t about forensic deconstruction of Nike’s customer habits. It was about invoking an emotive response to the strength of human conviction. It was about Nike’s position as a social commentator, and the campaign speaks as much to Nike’s strength of commitment as Kaepernick’s.

The campaign was built by incredibly talented and dedicated people at Wieden + Kennedy, who likely spent countless hours building and scrapping ideas until finally they looked at that picture and said ‘this feels right’.

But what about AI?

I can already hear the tech evangelists cracking their knuckles and getting ready to unleash a flurry of comments about the ability of machine learning to drive better media outcomes.

An algorithm is only as good as its trainer. While real-time insights and data point analysis are invaluable, human intelligence is necessary to determine which combination of points will drive the required outcomes.

As data scientist Cathy O’Neil said in her book Weapons of Math Destruction, “algorithms are just opinions embedded in code”. And without the right humans imparting those opinions, you can bet your algorithm won’t know what to think.

The need for human-first augmentation

The strength of a cyborg (and the reason they have become an integral part of the scifi canon) is in the ability to take emotional intelligence, intuition, curiosity those things that make us innately human – and augment them with beneficial technology.

Relevant brands plan their marketing activities with an essential combination of mind and machine, and this stretches from top-level strategy to ensuring that each touchpoint is a contextually-considered, human-designed piece of a greater experience.

By using data as a strategic tool, a weapon to augment the specialised human expertise they possess, brands can reach new heights. They must build relevant, connective, and curated interactions on a personal level with consumers. And in order to do this, brands need to ditch the data-only programming and adopt the cyborg power of the human-led approach.

 Joel Trethowan is managing director of Alchemy One.


Get the latest media and marketing industry news (and views) direct to your inbox.

Sign up to the free Mumbrella newsletter now.



Sign up to our free daily update to get the latest in media and marketing.