Marketing AI depends on knowing what it’s good for. (Hint: It’s not “Dear Sydney.”)

Google has withdrawn its pathetic “Dear Sydney” commercial, in which a young girl is supposed to use Google Gemini AI to write a fan letter.

You can easily place the blame on Google’s advertising team for creating this embarrassing ad, but I think the problem is rooted far deeper. Nobody at Google actually knows what AI is supposed to be good for — why else would they think it was good for replacing a basic human connection? Of course, it’s not just Google. Microsoft, Meta, and OpenAI don’t appear to know what AI is good for, either. They don’t know how we’ll use it, so they don’t know how to market it.

Generative AI is just sitting there ready to help you do just about anything. And a tool for everything is really not what we’re generally looking for.

This is a common problem with general purpose technologies

When John Mauchly and J. Presper Eckert invented ENIAC, the first general-purpose electronic digital computer, in 1945, nobody knew what computers would be good for. The first purpose of ENIAC was to compute artillery trajectories. It eventually became clear that computers were good for all sorts of things. But nobody could have predicted they’d be good for helping perform music or keeping car engines running efficiently or modeling the US economy, but not so good at writing love letters.

When Vint Cert invented the Internet in 1976, nobody realized what a general network could be could be good for. The first purpose of the Internet was for scientists to share data between universities. Nobody knew we’d use it for streaming TV series or commenting on baby photos or sharing political memes. Or that it wasn’t so great at replacing a walk in the woods with a friend.

When Apple launched the iPhone in 2007, nobody realized what a handheld computing device could be good for. We thought phones were for phone calls, and learned that they could also be good for checking weather reports and stock quotes. Nobody know we’d be using them for traffic-aware driving directions or streaming music on-the-go or connecting to political rallies from anywhere. Or that they’d become so addictive that we’d forget how to talk to each other.

AI and large language models have been around for a long time, but when ChatGPT 3 launched in late 2022, they went mainstream. Like computers, the Internet, and smartphones, Generative AI is a general purpose technology, and we’re still figuring out what it’s good at. It is clearly good at finding patterns in large data sets. It’s clearly good at summarizing masses of information. It’s good at generating unexpected ideas. It’s bad at knowing what’s true and what’s not true. But we’re poking around, just like the computer experts in the 50s and the Internet users in the 80s and the smartphone early adopters in the 2000s. It’s very hard to tell which are the niche uses, which are the uses that need some inventor to make things easier, and which are the massively helpful uses that will emerge in the decades to come.

I am completely unsurprised that corporations can’t yet figure out how to get productivity gains from AI. It took decades to figure out how those other general purpose technologies could best help people, but their rollouts were slow enough that early adopters could feel their way without much in the way of expectations. Gen AI, in contrast, is seemingly instantly available everywhere, well ahead of the usual schedule on which people figure out how best to use such general-purpose technologies in practical and effective ways. (Quick tip: If you want a head start on these ways, read Alex Samuel’s pieces in the Wall Street Journal and get in line to get Charlene Li and Katia Walsh’s book.)

If you gave a caveman a working iPhone, would they know what to do with it? If you gave a peasant in the dark ages a credit card with a chip in it, would they know what to do with it? If you gave a person in the 1980s a PC running Gemini, would they know what to do with it? We need not just knowledge, but a whole infrastructure to surround it, to make these things useful.

A decade or more from now, generative AI will be in nearly everything, just as now there are microprocessors, internet connections, and smartphone apps embedded in nearly everything.

In the meantime, the winners among the tech companies will be those who figure out how to use these technologies to deliver useful ways to improve our work and our lives, and then proselytize those actual uses and methods, not just some massive “productivity enhancer” that sits there mocking us as we do our everyday work. To market that properly, first they need to figure out the easy use cases that will get regular people excited. (Why am I thinking Apple may be the first to get this right?)

And for lord’s sake, stop trying to convince us that AI is the best way to replace the actual human connections that make our lives worthwhile. That’s the lesson of the humiliating failure of “Dear Sydney.” I hope the big tech vendors are learning it.