AI and Human Inefficiency

In The Four Points, I stated, “People assume AI is competing with human intelligence. I believe it is competing with human inefficiency.”

I believe this is a valuable statement. Looking back, I wish I had spent more time unpacking it in the book. Perhaps this is a good place to explore the idea further and, more importantly, evaluate whether the argument holds up.

When I refer to human inefficiency, I am not suggesting that people themselves are inherently inefficient, incapable, or unwilling to work. In fact, the opposite is often true. Organizations are filled with intelligent and experienced people who spend a remarkable amount of their time performing tasks that require very little of their intelligence or experience.

We search for information we know exists, recreate work that has already been completed, move information from one system to another, work through long email chains to find a relevant fact, and assemble reports from data that is already available. We summarize meetings, prepare notes, format documents, reconcile spreadsheets, search contracts, and repeat administrative processes because those activities have become part of how organizations operate.

These activities are necessary, but they do not necessarily make better use of human intelligence. In many cases, they simply consume the time of intelligent people. That distinction is at the center of my argument.

There is growing research suggesting that artificial intelligence can reduce some of this friction. In a randomized experiment published in Science, researchers Shakked Noy and Whitney Zhang studied 453 college-educated professionals completing occupation-specific writing assignments. Participants with access to ChatGPT completed their work approximately 40 percent faster, while the quality of their output increased by 18 percent.

What interests me about that result is not simply the productivity improvement. The education, professional experience, and underlying intelligence of the participants did not change during the experiment. What changed was their ability to apply those capabilities to a task more efficiently.

A separate study involving 5,179 customer-support agents reached another interesting conclusion. Access to a generative AI assistant increased productivity by approximately 14 percent overall, with considerably larger gains among novice and lower-skilled workers. The researchers found evidence suggesting that the technology was helping less-experienced employees adopt practices associated with more capable and experienced workers.

That finding gets particularly close to the idea I was trying to express in The Four Points. Much of the knowledge already existed within the organization. It resided in experienced employees, previous interactions, established practices, and accumulated organizational information. Technology did not have to invent that knowledge to create value; it helped make existing knowledge more accessible and usable.

Research involving knowledge workers has also found evidence that generative AI can reduce the time devoted to routine communication. In a six-month randomized field experiment involving thousands of workers across dozens of organizations, researchers found meaningful reductions in time spent on email among employees who used the technology. At the same time, they found less evidence that AI fundamentally changed the broader composition of people's work.

That distinction is worth considering because it suggests that AI may initially have its greatest effect not by transforming every aspect of a job, but by removing pieces of friction embedded within it.

There is also an important limitation to the argument. Research involving 758 Boston Consulting Group consultants demonstrated substantial improvements in speed and performance when people used AI for tasks that fell within the technology's capabilities. However, when participants were given a task outside those capabilities, those relying on AI were less likely to reach the correct answer.

That result reinforces something equally important: reducing inefficiency does not eliminate the need for human judgment. If anything, widespread access to tools capable of producing answers quickly may make judgment, experience, context, and accountability more valuable. Efficiency has little value when it allows us to arrive at the wrong answer faster.

This brings me back to the original statement. I do not believe the available research proves that AI is competing with human inefficiency rather than human intelligence. That is my interpretation of what we are beginning to observe. The evidence does, however, demonstrate that AI can allow people to complete certain tasks faster, access and apply existing expertise more readily, and spend less time on some of the administrative friction surrounding knowledge work.

Perhaps, then, the more useful question for organizations today is not whether artificial intelligence will eventually become more intelligent than people. A more immediate question is how much of our people's intelligence is currently being consumed by inefficient work.

If AI can reduce the time spent searching, transferring, summarizing, formatting, compiling, and processing information, it does not have to replace human intelligence to create enormous value. It can simply give people more time and capacity to apply the intelligence, experience, and judgment they already possess.

That is what I meant when I wrote that AI may not be competing with human intelligence at all. It may be competing with the inefficiencies that have prevented us from making full use of it.