Tableau CEO Ryan Aytay, storytelling skeptic, looks toward AI to narrate data — “It's just like full self driving”

Beyond AI explanations: the importance of downstream storytelling [v1.6]

The annual Tableau Conference took place in San Diego, CA April 29 through May 1.

Is it still good enough for Tableau to help people “see” and understand data? Or should this revered tool now lean toward AI to impart meaning? I asked Tableau CEO Ryan Aytay essentially that question at the recent Tableau conference — only to find that his answer led to more questions I wouldn’t get to ask.

His answer was good as far as it went — which was only to help with data’s first read, the “facts only” version — leaving users to themselves with the inevitable: dramatized data. We’re drama machines, and we render everything in stories.

For Aytay, storytelling comes down to narration, which he says is best done with AI. “The problem with storytelling in an unautomated way is it's a different story every time,” he said. “I think a lot of that can be automated, to be very fair,” he said. “It's just like full self driving [cars.] Is it ready yet or not? We don't know. It's getting better.”

AI probably does fine with narrated data’s first iteration, when the everyday data consumer wants the facts, just the facts. How are sales in this region or that one? Why is that total cost of sales up so much this year? Will I get my bonus? Did I get more sales this quarter than that *&^% in the New York region? Simple questions, quick answers. AI beats charts every time.

But what happens to self-driving data when our grasp of data is about more than just-the-facts? Suddenly, it’s in foreign territory. That’s where we, the dramatizing humans, translate data to our native language, stories. Each layer of stakeholders interprets and reinterprets, tells and retells the stories. Second, third, fourth, or later iterations adapt to new points of view. At each generation, AI’s original just-the-facts rendition dissolves in the collective brainy mist.

Imagine one day in late spring at Amalgamated Poultry headquarters. AI starts the morning with gut-wrenching data: diving gross profit. Guys in the data trenches wince. “Dang! The big boss ain’t gonna like it.” By lunchtime, Big Boss sees it and says, “We expected this. It’s all because of the sales group’s reorg. If nothing else, Thanksgiving’s coming up and, you’ll see, the big birds will ship.” Just before his driver arrives to take him home, the biggest boss of all hears about it. He says, “Well, it better improve quickly or heads will roll, and I don’t mean chicken heads.” Early the next morning, investors say, “Bail!”

But has AI necessarily given up? Couldn’t AI be on the job even now?

AI’s clever boss could have told it to stand by as the humans got carried away — then feed the dramatizing humans new stories. Surely, AI should be able to ingest email, texts, and voicemail the way large-language models do. It could anticipate the stories, then seed and feed antidotes or stimulants.

AI’s seeds sprout and propagate, hoping to head off a panic. But the trouble’s still not over. AI’s intervention is met with an immune response: human reactions to surveillance and attempts at control. Would AI be up for it?

Game on.

---

Do this: What instances have you observed or taken part in that resemble the kind of downstream storytelling described here? Could you tell how it had been altered, if at all? If so, how did you feel about that? If you have questions, email me with your thoughts at tc.dothis@datadoodle.com.