As AI becomes more capable, we are asking it to do work that used to require a lot of human knowledge and experience. It can write software, analyze data, research complex topics, develop plans, create documents, and make recommendations.
It is easy to measure progress by how much work we can hand over to AI. But I think there is another measure that is just as important: Do we still understand what AI is producing for us?
One idea that has stuck with me from listening to people talk about human responsibility and AI is that human oversight can’t simply mean putting a person at the end of the process to approve the result. If AI researches the problem, makes the assumptions, develops the solution, and presents a finished answer that nobody really understands, having a human click Approve doesn’t mean the human is really in control.
Delegating Work Doesn’t Remove Our Responsibility
We already delegate complicated work to people every day. A leader doesn’t personally do every task their team performs. But when the work matters, someone still needs to understand what was done, why important decisions were made, and whether the result is acceptable.
AI shouldn’t change that responsibility.
That doesn’t mean a human needs to redo everything AI does. If AI writes 10,000 lines of software, requiring someone to manually recreate or inspect every line would remove much of the benefit.
Instead, we need enough understanding to make good decisions. We should understand what AI was trying to accomplish, the approach it took, the important assumptions it made, and the evidence supporting the result. We should also know where AI is unsure and what could happen if it is wrong.
The goal isn’t to understand every detail. It is to understand the parts that matter to the decisions and outcomes we’re responsible for.
AI Should Help Us Understand Its Work
There is an interesting risk as AI gets better. When we do difficult work ourselves, we naturally learn while doing it. We run into problems, discover tradeoffs, and find things we hadn’t considered. That experience gives us a deeper understanding of the final result.
AI can now do much of that work for us. That’s incredibly useful, but it can create a gap: the quality of what AI produces can go up while our understanding of it goes down.
I don’t think the answer is to use less AI. Instead, AI should help us close that gap.
Along with producing an outcome, AI can explain the important decisions it made, point out its assumptions, tell us what it isn’t sure about, show us the evidence behind the result, and highlight what deserves our attention.
We shouldn’t have to review everything AI does. That would defeat much of the value AI provides. But we should understand enough to know whether the outcome is one we’re willing to accept and stand behind.
Our Role Is Changing
As AI takes on more of the work, our role changes too. We can spend less time directing every step and more time making sure the goal is clear, deciding what AI should be allowed to do, defining what a good outcome looks like, and understanding the important decisions that were made along the way.
That doesn’t make the human less important. It changes where human attention provides the most value.
AI can take on more responsibility for doing the work without removing our responsibility for the outcome.
The future of working with AI shouldn’t be measured only by how much work we can give it. It should also be measured by whether we can give AI more responsibility while still understanding enough to make good decisions about what it produces.