Yesterday I overheard someone say “the AI on the Bolt app is amazing”, when I asked why they said “it’s because when I go to book my bike to football practice each Monday it knows where I want to go”. Wait… is personalisation AI now?
AI has great marketeers; a lot of the technology that is underlying AI has been around for years, but now simple features utilising recommendation algorithms are being viewed by consumers as “AI”. And much like a Waitrose opening up within a 5 mile radius of your house increasing the value of your property, being able to say your product is “AI-driven” will add perceived value to your offering (whether it actually offers more, or not).
So an app that recommended which TV series to binge next, I’m looking at you Netflix, has gone from being called “personalised” to being called “AI” with very little change.
As product experts we always knew the key to a great product in the last 10 years has been personalisation, meeting the consumer where they are, offering exactly what they need at the moment they need it – even before they realise, if possible. And we knew this would become the expected standard, so any product that didn’t offer personalisation would fall by the wayside.
So is personalisation “AI”?
There’s been understandable pushback, especially by creatives, against the use of AI – and for what it’s worth, I’m in support of their plight. But articles, like this one from the BBC, tell stories of people attempting to label content that has been created using any form of AI involvement.
Is this product ‘human-made’? The race to establish an AI-free logo
But it always comes back to the same question – what actually is “AI”?
On the one hand users are referring to simple recommendation algorithms – and I mean simple, the Bolt example is purely delivering “at 6pm on a Monday night he always wants to go to X” – and on the other hand the general public are calling for all AI use to be labelled.
Even the most prominent argument in the BBC article suggesting that generative AI be labelled has its flaws. Generative AI is often what powers autocorrect and sentence completion. So does this mean that any piece of writing that has been written in Microsoft Word must be labelled? Assuming, of course, that the writer is not perfect and writes ‘teh’ as often as I do.
Perhaps we can define AI as anything probabilistic
The difference between probabilistic and deterministic is whether the output is always the same based on the input. So if we take a x4 function machine from our primary school maths lessons, sending a 3 in will always produce a 12. This is deterministic.
Recommendation engines, like the Netflix example we talked about earlier, are based on probability and the calculations involved in the AI function machine are unknown. This is because it’s based on masses of user data and there are millions of patterns hidden in that data that we, as humans, cannot derive in any sensible time frame.
So – in very simplified terms – what happens is it is trained on the usage data of users 0-99, and patterns show that x% of users that watched movie A and TV show B went on to watch movie C. So there’s a y% chance that user 100, who also watched movie A and TV show B will enjoy movie C. It’s far more complex than that, but you get the gist.
Because the output is based on a probability, that’s known as probabilistic output.
This all sounds plausible until I start talking about expert machines. These are basically like those flowcharts we did as kids in the girl magazines, or perhaps football magazines if that was more your thing. “Figure out which member of Backstreet Boys you should go on holiday with” – you remember them!
This is actually how a lot of customer service chatbots work. If they talk about X then ask them this question, if they answer A then ask them this question – they are scripted, hard-coded rules. It’s deterministic. Yet we call this AI.
So has the term “personalisation” be killed by “AI”?
Ultimately AI is a marketing term, and it has phenomenal marketeers. “AI” doesn’t actually mean anything.
The technologies driving a lot of what we now call “AI” used to be labelled for the purpose it actually served – including personalisation. Ask anyone what “personalisation” means and they’ll likely give some form of “a product or service that adapts to my personal use”. Ask someone what “AI” is then they could give any number of responses depending of their understanding of the technology.
I’d love for “AI” to be better described by what it’s actually doing. That way far fewer people would be tricked into paying extra for capabilities they were already getting, or products they don’t really need.
“Personalisation” did become the expected standard as we always knew it would, but “AI” came along and stole the limelight by delivering exactly the same technology is in a shiny wrapper – a bit like when you do a great piece of work and your manager presents it to the boss as their own.
So if you feel like you don’t really understand what AI is, that’s because it’s a made up thing. Understanding the purpose and function of the tool you’re using is the best way to be sure you’re not being scammed by AI’s brilliant marketing. That’s what we do at AIFI, we remove the hype, dispel the magic and just bring it back to what it always has been – maths.