The age of artificial intelligence has brought about a basic rule that will change the economy: anything that can be measured will eventually be automated. This isn’t something that will happen in the future; it’s happening right now. AI systems are showing amazing skills in many areas, from creative work to professional services, and they are changing the way we think about human labor and competitive advantage.
The wave of automation is already here
AI models today don’t need big breakthroughs to change whole industries. They are already changing the way creative professionals like writers, designers, photographers, architects, and advertisers do their jobs. At the same time, they are changing the way people in finance, consulting, accounting, and tax preparation do their jobs. As AI systems process huge amounts of data and give personalized advice at much lower costs, even traditionally protected fields like law, medicine, and academia are facing new problems.
Experts are arguing about whether artificial general intelligence is one year away or decades away, but the change has already started. Anthropic’s most recent data shows that 43% of user interactions with AI systems are already direct automation instead of human collaboration. As AI agents get smarter and more integrated into work processes, this number keeps going up.
Knowing the Risks of Automation
There is a clear logic behind which tasks are most likely to be automated: AI systems are most likely to target environments with a lot of data and clear metrics. Highly measured and codified environments, such as legal codes, tax rules, compliance protocols, or streams of sensor data, are the most likely to be automated right away.
People used to think that our ability to make decisions when we don’t know what will happen next was our last advantage. But this point of view is problematic because the meaning of “judgment” is always changing. It seemed like medical diagnoses, legal contract reviews, and creative storytelling that capture cultural moments all needed unique human insight, but AI systems are getting better at doing these things. Research indicates that individuals might favor AI therapists, counselors, and mediators in numerous circumstances, due to their round-the-clock availability, uniform quality, and considerably reduced expenses.
The Plan for AI Success
A big step forward in computer vision led to the current AI revolution. Fei-Fei Li, a computer scientist, realized in the middle of the 2000s that image recognition algorithms were not very good because they didn’t have enough data. Her answer was ImageNet, a huge image database with carefully labeled images that changed research by allowing people to compare their work to others. They changed the field with just two consumer graphics cards in 2012 when Alex Krizhevsky, Ilya Sutskever, and Geoffrey Hinton used this dataset with neural networks.
This breakthrough showed us a simple but powerful formula that is still used in AI today: combine the right data, clear goals for success, and enough computing power. This framework is useful for more than just recognizing images. It works whenever we can set up a task environment, set measurable goals, and give processing power for learning through trial and error.
The Growing Circle of Measurement
This automation engine is speeding up because of two trends. AI systems can now make fake data, which means they can make almost unlimited training examples for situations that are too rare or dangerous to capture naturally. Second, putting AI on different devices and sensors makes a low-cost network of measurement tools that can measure things that couldn’t be measured before.
The economic threshold for automation keeps going down as measurement gets cheaper and more common. Economist Zvi Griliches’s ideas about how hybrid corn is used led businesses to only use AI for high-value tasks where the benefits are clear and the costs are low. But as technology gets better and prices go down, even small use cases become possible.
We’re getting close to a time when “measurement too cheap to meter” is the norm, and continuous, accurate sensing is the norm. Synthetic data fills in gaps in real-world data, and lightweight AI models lower the need for bandwidth and latency. Every increase in the accuracy of measurements leads to better AI-driven decisions, which leads to more automation, which leads to more automation.
The Human Edge in Uncertainty
Humans evolved as generalists who could handle incomplete information and do well in situations where things weren’t clear. Our well-developed prefrontal cortex lets us think about things that aren’t true, like imagining different outcomes and changing our plans as things change. Even as AI systems get better, they still have a hard time copying this ability to plan across domains and without limits.
AI will continue to have trouble in a few important areas: places where measuring something is physically impossible or too expensive, places where privacy and ethics are important, places where people need clear reasoning, and places where people just want to talk to people. But just like the adoption of hybrid corn, future generations may look at these limits again and come to different conclusions about how much human and artificial intelligence should be used together.
The Domain of Genuine Uncertainty
Knightian uncertainty tasks, where the risks themselves are not clear, may be the biggest limit on AI automation. Scaling startups, putting money into risky projects, dealing with new threats, making monetary policy during regime changes, setting AI ethics standards, coming up with new art forms, or starting cultural trends all happen in areas where traditional probability models don’t work.
To solve these problems, we need to be able to imagine new and complicated counterfactual worlds, which is something only humans can do. Some creative work involves putting together things we already know, but truly ambitious projects depend on our ability to think of new possibilities and find our way to them even when the risks are too great to measure.
As technology advances, it turns problems that can’t be measured into problems that can be measured, so this list is always changing. Every change causes problems in the economy and society, and the people who are best at being creative, talented, and allocating capital get the most rewards. AI makes educational tools more accessible to everyone, which is strange because it also acts as a universal copilot. This could help more people reach these high levels of success.
What This Means for Leaders
For leaders of organizations, success is becoming more and more dependent on developing things that can’t be measured. This means concentrating on skills that can’t be measured, solving problems that don’t have clear solutions, and building qualities that can’t be measured, such as trust, taste, and quality of experience. To be successful, you need to have the courage to move forward even when all the signs point to waiting.
Amar Bose is a good example of this principle. While his competitors focused on measurable specifications, Bose focused on how music sounded to people in real life, which no existing metric could capture. Bose Corporation was able to change the audio industry by focusing on quality that couldn’t be measured.
Useful Suggestions
There are a few important parts to the strategic prescription. Organizations should back projects with a lot of uncertainty and no clear return on investment, reward teams that find new ways to look at problems and accept ambiguity, and move talent around to roles that deal with uncertainty in research, new markets, and complicated relationships with stakeholders.
It’s important to make it possible for people to meet by chance and combine ideas. This means adding extra time to schedules, encouraging people from different departments to work together, and seeing planned uncertainty as a strategic advantage rather than a problem.
Leaders who keep an eye on both measurable metrics and qualities that are hard to measure will be in the best position to succeed in the future. As AI keeps pushing the limits of what can be automated, the value of people lies more and more in finding their way through the spaces between the numbers—the areas that can’t be measured where real innovation and a competitive edge come from.
The revolution in measurement is changing every part of business and society. People who know what AI can and can’t do will help shape the future of work in a world where AI is everywhere.
