(Photo by Heidi Kaden on Unsplash)

I'm just going to come out and say it:

You're terrible at AI.

I'm stealing borrowing a line from Meb Faber, founder of Cambria Investment Management, who has repeatedly pointed out that people are bad at investing. It's not an attack on individuals, he explains, but an observation about the collective investing public. They unthinkingly fall into bad practices and their portfolios suffer as a result.

Faber's take is not limited to the financial world. Which brings us back to where I started:

You're terrible at AI.

Collectively, of course.

Individually, some of you are great at it! You've found ways to put AI – and for brevity, I'm using that as my umbrella term for all of data science, machine learning, and generative LLMs – to good use. You're making money. Your business processes have improved by leaps and bounds. Yes.

But collectively … you have a lot in common with the investors Faber describes: you fall into practices that seem reasonable on the surface yet work against your best interests. You eschew simple, practical guidance in favor of flashy approaches, taking on a heap of unintended bets along the way.

You're not doing it on purpose. Every company exploring AI envisions a favorable outcome. But you're thrown off by a cacophony of AI vendor pitch materials, as well as news about other companies' AI successes. You've become convinced that AI is something you can just plug into your business and see benefits. You feel you're getting left behind. So you dive head-first … into the shallow end of the pool.

Over the years I've seen companies trip over matters such as:

• Hiring a data scientist or Chief Data officer as the first step. (How do you know why you need AI in the first place?)
• Treating AI as a foregone conclusion. (Far better to start with your business pain points, then see where AI might help.)
• Excluding the data team from early product planning meetings. (You assume your AI team is only there to build, not to guide. But they are your in-house AI experts, after all…)
• Hacking together a proof-of-concept project from a tutorial and pushing that straight to production. (Just because it runs, doesn't mean it works!)
• Assuming the models will exhibit consistent performance over time. (So you don't monitor them, and you don't keep humans in the loop.)

And that's just off the top of my head.

How do we get better at AI, then?

We first have to recognize the common thread in those harmful practices: a lack of executive AI literacy.

When company leadership waves off AI knowledge as a problem for the data team, they're falling into a trap that will fuel unrealistic expectations and equally unrealistic initiatives. That leaves the company exposed to a lot of downside risk with little upside gain in view. In colloquial language we call that "gambling."

If we want to be less bad at AI, then, we need to start from the top of the org chart. Anyone with a C-level or product leadership role should understand the basic concepts behind what AI can and cannot do. (They don't need to write code to train models, or work through the math behind neural networks. It's cool if they choose that route! But it's not required.) That will help them to make informed decisions on how the company should approach AI, and improve their interactions with the data team. They'll also be prepared to go toe-to-toe with pushy AI vendors. Overall, they'll avoid the all-too-common AI pitfalls that make us collectively terrible at AI.

—

(I highly recommend you check out Meb Faber's podcast and blog. You can also watch his talk, How to Spot Bubbles, Avoid Market Crashes & Earn Big Returns – he goes into detail on the "you suck at investing" idea starting at the 8:44 mark.)