People in the ad industry talk like AI shopping agents are just around the corner, but I’m not so sure anymore.
I used to act like I was sure. If I was ever asked about how AI would change advertising, I’d always sagely point out that it would affect consumer behaviour, too, and that brands of the future may have to get past people’s personal AI assistants before reaching an actual human. I’d read about this idea of everyone having their own digital gatekeeper in an interview with computer scientist Pedro Domingos, and it seemed like a clever thing to parrot in 2019.
But now that personal shopping agents are possible, I’ve been swayed by arguments for why their adoption may not be imminent.
First, there isn’t as much reliable and unbiased information about products and services on the internet as you’d think. If you’re just looking to compare specs before making a purchase, no problem — AI agents can crawl company product pages to their clankery heart’s content. But you can’t expect a manufacturer to provide details about their product’s flaws and, as a couple of Andreessen Horowitz partners noted in a blog post last month, ‘most product reviews are noisy, gamed, or overly polarised.’
And even these noisy and overly polarised reviews might be out of shopping agents’ reach. Amazon is stepping up efforts to stop AI bots scraping its data, and I’m not sure why any other platform that hosts reviews would readily give them up, either, since they’re unlikely to see any referral traffic from shopping agents.
So, there’s reason to suspect that the quality of data with which these agents will have to make their recommendations may be lacking.
The other problem is the business model. If companies charge for access to an AI shopping agent, adoption of the technology will be severely hampered. On the other hand, if, say, Google makes its agent free for users, how could people trust that it will serve their interests? If we’ve learned anything from recent history it’s that Google will eventually start taking money from brands in return for preferential treatment, which creates a pretty serious conflict of interest.
These are not insurmountable problems, of course. Licensing deals may give the AI companies access to all the reviews they need, and free shopping agents could make money through sponsored listings. But they are big enough to suggest the rollout of shopping agents will not be the straight road that some people think it is.
And I’m only talking about the issues with shopping agents that make recommendations. AI models that go a step further and make purchases on people’s behalf are a whole other bag of snakes. We have the technology, we just might not have the trust.
