How Much Can You Trust Your AI Assistant? As Much As The Rest Of Your Team

This article was originally posted on Forbes.com, March 2024.

The world’s leaders recently met to address urgent global issues at the World Economic Forum in Davos-Klosters, Switzerland. The overarching theme was rebuilding trust.

One area of focus was the use of artificial intelligence (AI) as a driving force for the economy and society. The topic was also explored in depth in the Chief Economists Outlook in January 2024. The report states that chief economists expect AI to increase efficiency (79%), accelerate innovation (74%) and help increase standards of living (57%) in high-income economies. At the same time, 56% believe there is a decline in trust in AI, and only 23% believe it has a net positive impact on employment.

Trust what you know.

As the CEO of a company that delivers AI-driven solutions for autonomous finance, customers often ask me how we ensure that AI is not learning the wrong things. As I’ve said before, they fear not being able to control what AI learns, how it learns and what the outcome of this learning might be. People can’t trust AI because they don’t know how it works.

There are many definitions of trust, but I believe two of the fundamentals include humanity and transparency.

When it comes to AI, people probably trust its capability, but the elements of humanity and transparency are missing. Humans largely trust other humans because we share the same human experience, but this doesn’t extend to artificial intelligence, even though humans created it.

In reality, AI learns what we humans teach it. Sometimes clients complain that an AI tool did something wrong, or did not deliver what it was asked to do. I explain that happens because the instructions were not precise enough, or did not follow a logical flow to create the desired result. I ask them: What information did you give the AI tool, and what exactly did you ask it to do with the information?—and the key word here is exactly.

I always make a point to review the input with the clients.

Give clear instructions.

For example, I once gave a client a model to test with his own data. The client gave the tool a list of a thousand different documents and asked it to categorize them into different types. He told me the tool was not able to perform the task. Why not? During the review, he revealed that he did not specify the categories, but assumed that the tool already knew them. I explained to him that an AI tool is like an assistant; it’s like a new member of the team. You wouldn’t expect a new employee to automatically know the categories of documentation you work with, so you would give specific instructions on how you expect the person to complete the task.

It’s the same with an AI tool. It’s a matter of giving exact instructions in the form of prompts and providing a logical, step-by-step process flow. The other key element is to provide the correct data. Companies that are processing hundreds or thousands of invoices manually every month often experience data entry errors, omissions and common accounting mistakes due to incorrect data.

Humans will start trusting AI once they understand that they can trust the machine as much as they can trust themselves and their teams. It all depends on the transparency of the information you share. If you haven’t told the tool that one of your vendors has a new address, how will it know? It will continue issuing the invoice to the wrong address until the human in charge makes the change.

Learn the basics.

Regarding the fear of AI taking away jobs, another big issue with AI, I recently asked one of my team members what she would recommend to a young person who wants to go into accounting.

She said, “First learn the basics—the required mathematics, the processes, the principles that underpin accounting practices and the preparation of financial documents. Once you are well versed in the basics, let the AI tools do the drudgery and the boring bit. Your job will be to catch the errors. You’ll know an error was not made due to a wrong calculation made by the tool but due to incorrect data in the system. Your job will be to find that error and fix it.”

I couldn’t agree more. In the case of AI, trust requires transparency and humans to ensure that it all runs properly. Only when people start trusting AI can it fulfill its true potential as a positive force in the economy and society.

The information provided here is not investment, tax or financial advice. You should consult with a licensed professional for advice concerning your specific situation.

Uli Erxleben for Forbes Council, February 2024.