Will AI agents ever understand what humans really value?

Lucian Camp is a financial services brand consultant, copywriter, author and blogger.

It seems we’ve reached the point – or at least I’ve reached the point – of being unable to imagine how the biggest and most important new tech ideas are actually going to work in real life.  I’m thinking of three in particular.

The first is self-driving cars, and especially in London.  My issue isn’t the usual fretfulness about safety:  it’s very specifically the pair of pedestrian crossings in Goods Way, behind Kings Cross station.  There’s a never-ending stream of pedestrians crossing on both of them, so that the only way you can get across in your car is by sort of nudging forward onto the crossing, very gently and not very legally, until the pedestrians pause for a moment and let you have your turn.   Otherwise, you’ll be waiting there till midnight – and I don’t think driverless cars will be programmed to do it.  And of course there are thousands of other places in London where such unwritten rules apply.

The second is delivery drones, and also especially in London.  Amazon are allegedly keen on these, but I can’t help thinking their enthusiasm is likely to cool when the first few malfunctioning drones accidentally drop their loads onto the heads of innocent passers-by, or crash-land into queues at bus stops.  One of my most recent Amazon purchases was a remarkably heavy pair of bolt-cutters (don’t ask…).  It doesn’t bear thinking about.

Now, I’m experiencing a third imagination failure, and this one’s about AI shopping agents.  I get it up to a point:  having just spent a miserably long time online trying to find 150-watt-equivalent warm-white dimmable LED bayonet-fitting light bulbs, I’d have very happily have handed the whole tedious business over to an AI agent and accepted whatever it offered me.

But that wouldn’t be anything new.  It wouldn’t involve anything much more than the kinds of filters that online retailers have offered for years.  If AI agents are going to take us much further,  they’ll need to work to much looser briefs, and come back with much more unexpected and lateral solutions.  And that will necessarily mean engaging with the emotional and qualitative side of our attitudes and behaviours as consumers – the way we think and feel about brands.

Here, the central problem is that many (most?) of us are wildly inconsistent and irrational.  Some days I’m in the mood for Brand A, but other days are Brand B kind of days.  And when I’ve just received an alarming bank statement I really feel I should switch to that low-cost Brand C.  But then, a week later, my finances are no better but if I’m booking a restaurant for a nice Mothers’ Day lunch, Brand C feels all wrong and I’m back to Brand A again.

A well-made You Tube explainer tells us how all this is going to work, using the purchase of a suitcase as an example.  The customer provides a simple prompt – “I want a sturdy suitcase, well-made. Not too dear” – and Google Gemini comes back with a recommendation.  The customer agrees, and the AI makes the purchase.

Well, there may be customers who’d be happy with that, but not many.  What size should this suitcase be?  Is it going in the hold, or the overhead locker?   If the latter, does it need to meet easyJet’s size criteria?  Or Ryanair’s?  Or BA’s?  Or all of them?   Four wheels, or two?  What colour?  What brand?  (Do Samsonite and Antler feel a bit, well, mass-market?  How about one of those fancy Rimowa aluminium ones?)  How do you feel about a copy of something more famous?   Obviously price is an issue, but is it actually more about value – will you pay more for something a bit special?  If so, how much more?  And there are plenty of secondary issues – weight, security, what the reviews say – that you’d accommodate in your own personal algorithm if you were making the purchase, but I don’t think an AI agent would.

Maybe I’m just a fussy shopper, but I don’t really think I am.  On a given day, I’m just as capable of ignoring all this complexity and making a completely irrational choice – “I had to have it because I just loved the colour!”   But the variability of my decision-making makes it harder, not easier, for the AI.  There are a thousand ways to get me wrong, but far fewer to get me right.

The tech is pretty much there to achieve this new AI-driven of shopping.  (Although not yet to get us to the end-game, which I’m told is where your suitcase tells the AI that it’s getting old and tired, and one of the wheels is wobbly, and the AI goes off on its own and buys you a replacement.)

It’s widely suggested that all this will bring about the death of brands, and that before too long it’ll all be about the way that facts are presented to get that vital AI recommendation.  That may be true in some markets, where brands were never that important anyway – light bulbs are a good example.  But in the end, consumers still have to be happy with what the AI proposes – and in all the many markets where brand is the source of much of that happiness in the old economy, it’ll continue to be so in the new one.

If AI shopping agents can’t cope with this fuzzy, subjective, shape-shifting dimension, it seems to me that a lot of consumers aren’t going to be very impressed by them.  And that’s without the delivery drone accidentally dropping that new suitcase on their head.