Can AI Make Better Decisions than Employees?

Before I get started, I’m aware that discussions around AI making decisions can easily get into ethical territory. For this article, I’m going to cravenly avoid that angle and stick to the technical implications that I think drove the original question.

The short answer is: yes. However, if you subtly change the question to “Does AI Make Better Decisions Than Employees”, I’d change my answer to “sometimes, at the moment.”

There was a revealing study by Stanford University published in October where GPT-4 (the model that powers ChatGPT) was tested against doctors for diagnostic reasoning (i.e. a patient’s symptoms, medical history and relevant information is provided and a diagnosis is requested). The AI alone outperformed all the doctors, including the doctors who were given AI to help them! Those doctors, it turned out, would ignore or overrule the AI’s recommendations and stick to their guns.

This is a thought-provoking point about what being 'better than humans' means. We hold the human standard up as the best, when we are in fact incredibly hampered by irrational biases, heuristics, emotional influences and other human shortcomings. AIs – even GPT-4 which was superseded months ago - have surpassed us in many areas already.

The aim of the AI companies is absolutely for AIs and AI-powered tools to make decisions for us and do so with higher accuracy, greater speed and lower cost than humans. Aside from our abovementioned human fallibility, AIs can parse vastly more data than us, hold it all in mind at once, and calculate it at superhuman speeds. We really don’t compare.

However, there are many types of decisions to be made in business. For some, AI is not the answer and won’t be for the foreseeable future. Here’s my attempt at a simple breakdown:

Where AI is not best to decide things:

In matters of taste, such as art, AI decision making would be a detriment to the production. It may speed things up or make things cheaper and that trade-off may still be worth it for some approaches to production, but editorial/artistic decisions made by AI are inferior to those made by skilled, experienced humans. Secondly, when it comes to interpreting humans, assessing the meaning of what a person said or did, managing relationships and so on is going to be an area that’s difficult for AI for a long time yet. It’s getting to the point it can assist when it’s trained on large data sets, but with no lived experience, AI can’t interpret humans in the way we can.

Where AI is best advising, but not deciding:

There are lots of places where AI is excellent at advice. The GP example above, being one. Yes, we still may not listen to it and that may be a problem in some fields, but we are also currently at a point where the AI does not have all the information needed to make a decision and the final call should therefore be human. AI is a very powerful tool for checking our thinking, offering alternative ideas that we wouldn’t consider, and acting as a muse. But for many fields, particularly where mistakes are costly, it should not be deciding things. However, this category is shrinking every week as the technology improves.

Where AI is best to decide things:

This is a bigger area than people think. Partly because algorithms have been invisibly deciding stuff for years now but also because we overlook and take for granted many of the little decisions we’ve automated away already. AIs already decide when a transaction looks suspicious and prevents fraud, it decides what counts as “noise” and cancels it in our headphones, it blocks spam from our inboxes and corrects our typos. And on and on. The more capable the AIs get, the more data they are trained on, and the more tasks they are taught, the more decisions they will make. They are superlative pattern-recognition machines and as such, can decide when something is significant or not far better and faster than us, as well as in huge datasets that humans can’t practically look into.

I’m aware I’m at risk of depicting a world where someone sits down at 9am and just does task after task or decision after decision until 5pm. In reality, work is different. It’s full of other forms of work and social elements, with outputs that are harder to grade in efficiency terms or calculate the value of. AI changes all that too but that’s a subject for another time. 

In summary, we are now at the beginning of the age of AI, which among other things, is seeing more and more decisions being handed to the machines. While they will be better than us in most fields, that doesn’t mean they will be perfect. But who is?