The AI explainability vs performance paradox
Between the unwarranted attacks from celebrity billionaires and the inaccurate depictions in Hollywood films, public understanding of what artificial intelligence technology is, how it works, and what it means for humankind remains just as muddled as ever.
That’s great news if you’re writing a screenplay about a sassy dishwashing robot with an insatiable desire to join a breakdancing crew (there’s nothing in the rulebook that says a robot can’t breakdance). Sadly, it’s not such great news for those of us tasked with marketing the world’s most effective AI marketing language optimization tool to the masses.
AI that’s easy to explain probably isn’t all that effective
Explaining what our AI does is a piece of cake. Here at Phrasee, our AI generates on-brand language for your marketing campaign, tests the language it generates on your brand’s audience, learns what works and what doesn’t, and optimizes the language it generates for your next campaign based on how the last campaign performed. In this way, it gets better and better at catering to the marketing language it generates for your brand with every campaign.
Simple and easy to understand, right?
The thing is though, explaining what our AI does is rarely enough. Anyone who’s thinking of investing some of their precious marketing budget to experience Phrasee for themselves usually also wants to know how our AI does what it does and this is where things get tricky.
Were our AI tech a little simpler and more basic then explaining the intricacies of its processes and mechanisms might be a little easier to do but it isn’t.
And, to be frank we’re glad it isn’t, because if it were, it wouldn’t be as effective at delivering increased marketing performance and ROI to our awesome customers as it is, which would be a shame.
Effective AI is difficult to explain
Phrasee was built by one of the most innovative and meticulous scientists in the artificial intelligence game, Neil Yager, PhD.
Neil spent a full decade in university learning everything there was to know about AI technology and how it works. He is about as well-versed an AI expert as you’re ever likely to find.
And yet, when speaking about the amazing capabilities and awesome power of AI like Phrasee, Neil can be counted on to do one or both of the following in almost every case:
A) Shrug his shoulders and show you this chart:
B) Say something like this:
“There is an inverse relationship between the accuracy of a machine learning algorithm and our ability to understand and interpret its output. This is an active research area and perhaps advancements will be made in the future. However, for now, we’re in a tough spot. Taken to the extreme, consider the human brain. It is the most powerful computation device in the known universe. The vast majority of our actions (e.g. “should I open this email?”) are not based on careful and deliberate reasoning. They are made quickly and subconsciously, and we have surprisingly little insight into our own thought processes. It isn’t surprising to me that computer intelligence is heading in a similar direction.”
In a nutshell, once an effective AI has been built, data goes in, results come out, and even the world’s top AI experts have a hard time explaining why and how this process happens. If the results are good, the AI’s designer has done their job well. If they aren’t, it’s back to the drawing board.
That’s why here at Phrasee, we prefer to let our results speak for themselves. Can we explain in plain language the gruesome details of exactly how Phrasee generated millions of pounds in additional email marketing revenue for Virgin Holidays, or how it reduced Wowcher’s Facebook ad cost per lead by a full 31%?
But when you think about it, does it really matter?