Separating Fact From Fiction on AI: A Q&A With Deloitte’s Will Bible

The realities of AI's role in the audit today and what we can expect to see in the future.


At the June AI, Technology Innovation & Compliance conference presented by Compliance Week, Deloitte’s Will Bible discussed how AI and other emerging technologies will transform regulation and compliance in the finance industry. FEI Daily spoke with Bible about some key takeaways from the conference and how he believes AI will affect the audit and compliance in the future.

FEI Daily: You recently presented at Compliance Week’s AI, Technology Innovation & Compliance conference. What were some of the highlights from your session?

Will Bible: Our session was about Artificial Intelligence (AI) and Machine Learning and some of the conceptual impacts to compliance functions within companies. The audience was senior compliance officers and some internal audit folks as well.

We talked a lot about separating fact from fiction. We talked some about the challenges with deploying AI in practice, in particular the need for large amounts of data to be able to actually develop algorithms that are useful in automating business processes.


We also spoke quite a bit about some of the limitations and how the robots are not necessarily coming for all of us. This is a tool in the tool kit that will help you keep up with the vast amount of information and requirements that are kind of being put on people every day. Machine Learning and AI are tools that let you sift through a lot of data and make sense of it all, as opposed to one that in the short term is going to automate the judgments that people make in their professional careers.

FEI Daily: How have you seen AI affect the audit so far?

Bible: Today it’s being used in a limited fashion, almost as a utility to execute audit procedures. What I mean by that is it's an extension of a person. For example, within Deloitte, if we have an auditor that is reading contracts and trying to extract information from those contracts, we have an AI tool that they can use and train to do that extraction. They can train the tool to do it or a subset of those and then the tool will start to do it on their behalf and they stay involved and watch over it to make sure it stays on path, but it becomes an extension of what they're actually doing, not a replacement.

A lot of the automation that happens right now, I think of it as automating slivers of people's day, little fractions, at least in the professional services world. And part of the message of the conference was whether or not the Chief Compliance Officer has something to be worried about in the short term. I think the answer to that is no. If it’s someone’s sole job function to read contracts and extract data, then you would have replacement. But that's not been the job function for many people in a long time.

FEI Daily: What about compliance? Is AI making an impact today?

Bible: Where you see AI really taking hold is where there are vast amounts of data, and easily accessible data. One panelist was actually talking about how FICO monitors payment information across the world's payment networks. In order for that to happen, in a compliance function, a company would have to have access to large amounts of information and be able to train an algorithm to do things. That's not where we see a ton of AI development right now. Most AI development's in using public data sets, things like picture recognition, converting text to voice and voice to text, etc., really things that are driven by digital depositories on the internet.

FEI Daily: When you’re sitting down with a CFO or with a compliance officer, how do you describe AI to them in relation to what they do?

Bible: The first thing that I think is important is distinguishing AI from Machine Learning. AI is a broad field of training computers to replicate human intelligence. What we really see in practice right now is the development of Machine Learning, using Machine Learning to develop algorithms to replicate certain processes that people are involved with. It becomes a lot easier to understand when you say ‘You're going to use lots of examples to essentially teach a machine how to do something that someone already does.’

I think most executives understand that concept because it's not dissimilar to when we took manufacturing assembly lines and we mechanized them. We went out and studied what people did and then we built machines that probably did it a little bit differently than the people did. This is the same kind of mechanization but for digital jobs. And I think that's pretty easy to understand.

Where AI gets really fascinating and different than the assembly line is that those machine learning algorithms can be much broader, can encompass much more data than we would even conceive of as people, and so they can find things that you wouldn't expect. They can pick out patterns and pick out needles in the haystack that as a human, we really can't see because we don't have the capacity to absorb all the information.

The thing to emphasize if people are starting to think about how to use Machine Learning to automate processes, is that it's first a data problem. How do you get a source of ground truth that you can use to then train an algorithm to do something? And even if you're using unsupervised Machine Learning where you don't have that source of ground truth, you still need a vast quantity of data.

I often point people to the reason why we see development of AI-type tools in internet land. It’s because that's where all the data is. People upload lots of pictures and video and having access to data let's AI developers use that information to build the algorithms. Digitization is really the first step to get there. You walk into a building where you’ve got everything on paper in filing cabinets, you're not doing Machine Learning anytime soon.

To read from Will Bible on innovation in accounting and auditing, click here.