American AI

The Right to a Human v. the American AI Investment Thesis

The most unlikely scenario ever generated.

How much better would life be if we always had the right to a human? Unfortunately, we don’t. Companies are increasingly saving on customer service workers in favor of AI assistants and good advice is expensive, so is therapy. Most of the time, we’ll have to make do with an AI model.

Nikhil Suresh writes a wonderful story about an experience he had with car maker Mitsubishi’s customer chatbot. A very polite voice asked him to describe the problem he was experiencing with and said he’d receive a call back as soon as someone was available. Suresh describes it as the single most competent implementation of a chatbot he had ever seen in the wild. A natural sounding voice, quick responses, and the promise of a quick resolution. And then, nothing. He never received a call back.

“When Mitsubishi did not call me back, what happened? Did that request just go into the void, showing one less incident for the year? Does it appear that the phone bot resolved my query without the need for human intervention? All we know is that it didn’t show up as an error, or I’d have received a call. I’m sure it looks great in all sorts of ways except the one that matters, which is that I was planning to buy a car and decided not to buy another one of theirs.”

I can relate to this story. For example, I receive several e-mails each week from people who want to be interviewed. Many of these mails are upbeat, friendly, and relate what they want to talk about to my posts or prior conversations. The display of effort makes me smile, but I can tell that most of the e-mails are generated, so I seldom reply. But then, some time ago I thought what the heck, let’s respond to a few of these and try to arrange a call, see what happens. And… no reply… so… why did they reach out in the first place? Why go through the bother of generating an e-mail that appears personal and sincere, reaching out to a stranger, and then ignoring the response?

Our lives are full of such mysteries now. The observation that AI makes the easy stuff easier and the hard stuff harder, is only half truth. Easy tasks such as cancelling a subscription, coordinating a call, or receiving help with a request about a product or service is too often a Kafkaesque experience that leaves customers bewildered and frustrated with more questions than they came in with. At the end of the process, they don’t even know whether to ask why or what. Why… what?

The disconnect between what customers and companies need is symptomatic of the AI movement as a whole. Customers may need help with basic requests, but board members are eager to prove that AI is a real, sensible investment before their competitors do. The pressure to adopt AI creates a headless chicken race to the bottom, where companies overpromise and oversell what they can do with AI and cost-cut more on other expenses than what is healthy for their business.

Julie Averill, who worked as the chief information officer of the athletic apparel company Lululemon for eight years, distinguishes between AI wishing, AI washing, and AI layoffs in a kind of flywheel.

AI wishing is the belief that AI is magic; you can wave it towards any hard problem like a wand and skip the work of actually solving it.

Then comes AI washing. Companies are pressured to show results, so they overstate what AI is currently capable of doing and what it may do for them in the future.

This leads to AI layoffs. The companies proclaim that they need fewer people because AI made their operations more efficient. In reality, the efficiency does not yet exist, but the companies need to free up cash, sometimes to spend more on AI.

Then, the cycle continues. While CEOs and board members paraphrase canned lines from inspo-posts on LinkedIn such as “remember, this is the worst the technology will ever be” with dead eyes and the enthusiasm of North Koreans or MAGA republicans praising their Dear Leader.

Paradoxically, AI foundation models can prove and disprove decades-old mathematical conjectures, yet fail to solve basic customer service requests. Here, we could talk about the “jagged frontier”, but I have found a better framework for understanding how AI can solve extremely complex problems and fall flat on easy ones. If this framework holds up, it pokes holes in the already porous American AI investment thesis and underscores the need for further regulation. Much more in the paid section below.