AI is Just Productivity You Don't Understand Yet

When I see someone say or write "I'm using AI for [X] task", it drives me crazy.

Having been in the machine learning domain for over 20 years I've seen the term AI used in many eras and under many guises, but never meaning the same thing in practice. When I was writing the Deep Learning book for O'Reilly in the mid-2010s (with Adam Gibson), I banned the term "AI" from the main book and then set aside a long appendix chapter to explain a short history of the term. Why? Because when you are writing a machine learning book on neural networks you have to be specific in what you mean and how you mean it^[1].

The thing I realized that really bothered me is neither the people marketing "AI" nor the people saying they were using "AI" we're 100% sure what they meant. It was just some general term that was a placeholder for "do more stuff faster". As soon as the technology wave crashed, the marketers moved on to the next thing to call "AI" and it seems like everyone keeps on playing along^[2].

Once the idea of AI reached the sales and marketing departments, the term "AI", or "Artificial Intelligence", has meant whatever the person writing the marketing copy at the moment needed it to mean.

The Shifting Meaning of Artificial Intelligence

For decades, the term artificial intelligence^[3] has been a moving target, never staying in one place for long. What one generation calls AI, the next quietly reclassifies as something ordinary — no longer “intelligent,” just useful. Early benchmarks such as playing chess, recognizing speech, or translating languages were once seen as the pinnacle of machine intelligence.

Yet, as soon as these milestones were reached and integrated into business operations, the techniques lost their aura of AI and were renamed as specific disciplines like “search algorithms,” “speech recognition,” or “machine translation.”

This pattern has repeated itself over time.

Each time machines meet a benchmark that defined AI, the definition of AI slides forward to a new set of tasks — a horizon that has not yet been reached.

In practice, AI is less a single, fixed achievement and more a label we attach to the frontier of what machines cannot yet do. As progress is made, yesterday’s AI becomes today’s software, absorbed into the tools we use to drive productivity and solve problems more efficiently.

For executives, this moving goalpost is important to understand. AI is not one static technology to buy or implement but a stream of evolving capabilities that can be harnessed as they mature. Each wave of progress brings opportunities to improve operations, open new markets, and rethink how work is done — if we are ready to recognize that what was once “intelligent” quickly becomes simply another tool in the business arsenal.

AI is Just Productivity You Don't Understand Yet

The term "Artificial Intelligence", the way it is used colloquially in culture today, just means productivity you don’t yet understand.

Full stop.

In every age, what people call “AI” is nothing more than a collection of methods that transform information faster, more accurately, or at larger scale than before — but in ways that feel unfamiliar at first. To cut through the noise, the simplest mental trick is this: whenever you see “AI” in a proposal or a product, ask:

"how does this method improve a specific part of information transformation?"

If no one can answer clearly, you can dismiss it as hype.

When the answer is clear — when it explains how a task becomes faster, cheaper, or more precise — you no longer need to call it AI.

You can give it its rightful name and put it in the right productivity bucket. Optical character recognition? Automated underwriting? Predictive maintenance? Once the mechanics of value are understood, the label of AI falls away and what remains is a real tool, grounded in the business of making knowledge work more productive.

Companies don’t buy hype, not for long. They buy productivity — even when the productivity wears a fashionable name for a few years. As leaders, your job is to see through the fog early.

Name the productivity gain.

Name the method.

And then buy or build it for what it is: a lever that makes information work cheaper, better, or faster.

AI is nothing more and nothing less than the next lever you don’t yet recognize.

Footnotes

1. There is no room for hyperbole in the explanation of back propagation.
2. Douglas Adams was on to something when he mused that "Technology is a word that describes something that doesn't work yet"
3. Obviously the idea of general intelligence (AGI) in a machine or program is intertwined in this discussion, but that's a sub-topic I am seperating out for now.