Human Judgment in the Age of AI

Artificial intelligence is an agent of leverage – a tool that can extend what people are able to do, but not replace the experience, intuition, and discernment that shape human judgment. In fact, as AI becomes more capable, it may help us see more clearly where human judgment matters most.

That was my takeaway from a recent AI Weekly article examining 159 real-world artificial intelligence projects across 21 industries. Some of the results are remarkable. AI is helping doctors read X-rays faster. Robots are assembling food orders at speeds humans simply cannot match. Artificial intelligence is sorting packages, helping drive cars and producing measurable gains for businesses. But then there are the failures; those may be the more interesting stories. Ford reportedly brought back 350 human quality inspectors after automated systems failed to catch defects experienced workers could see. A school district abandoned AI-writing detectors after a student was wrongly accused of cheating.Ajudge overturned roughly $100 million in federal grant cancellations after an AI-assisted review produced serious mistakes.

There seems to be a pattern here. Give artificial intelligence a narrow task with measurable results, and it can be astonishingly useful. Ask it to exercise judgment, i.e., to understand context, recognize a subtly nuanced circumstance, or determine what simply doesn’t look right, and suddenly the human being standing beside it becomes much more important.

I have discovered something similar in my own use of AI. Recently, while researching a historical question, I had gathered several seemingly disparate pieces of evidence. Intuitively, I believed they were connected, but proving it meant finding the one record that could tie them together. WithAI helping me navigate the layers of an online historical database, traced the evidence to a military record collection on Fold3 where I hoped to find a letter that could confirm my theory.

The trail didn’t end there. The actual letter wasn’t online. The next step was the National Archives and Records Administration. AI helped me identify exactly what I needed, determine where the record should be held, and formulate a precise request to the archivist. After sending the email, I had an answer within 24 hours. That is the kind of time-saving assistance AI can provide.

AI didn’t develop the theory. It didn’t look at those disparate historical clues and sense that there was something worth pursuing. I did. But once I knew the answer I wanted, AI helped me navigate a complicated research trail that might otherwise have left me staring at fifty possible directions and wondering which one to take.

Perhaps that is where the conversation about artificial intelligence ought to be headed. Not, “Will AI replace us?” but something more useful.

What decisions are we willing to turn over to a machine? Which ones should remain human-generated? How do we know when AI has made a mistake if we’ve already stopped checking its work?

And perhaps the most important question of all: As artificial intelligence gets better at thinking like us, will we get better, or worse, at thinking for ourselves?

Lisa Musick of Praha welcomes feedback and questions via email at: lisa@lisamusick. com.