Mind blowing

At the 34th Annual ACFE Global Fraud Conference in June, I gave a presentation, in two identical sessions, on ChatGPT and generative artificial intelligence (AI). Both sessions were standing-room-only with a combined attendance of over 1,600 in-person and online. Trust me, it wasn’t me they crowded the conference room for — it was the topic. The presentation’s popularity sent me a clear message: Fraud examiners are eager to learn about this exciting, emerging area of technological innovation.

So, for those who were unable to attend, I’d like to share the important takeaways in this column. Like many of my anti-fraud technology colleagues, I find myself immersed in a whirlwind of excitement surrounding generative AI and how it can impact fraud prevention and detection. The sheer potential of generative AI technologies has ignited a frenzy of curiosity and anticipation in all industries. The hype isn’t unfounded. It stems from the transformative capabilities that generative AI vehicles like ChatGPT bring to the table. Let’s examine what ChatGPT is, what it isn’t, and how it will indeed transform how you look and interact with your data in the very near future.

What are ChatGPT and generative AI?

ChatGPT is an AI chatbot developed by OpenAI and released last November. “Chat” is a reference to it being a chatbot, and “GPT” stands for generative pre-trained transformer — a type of large language model (LLM). Think of an LLM not as an advanced law degree, but as a super-large body of content that’s indexed for search and training of the model. In ChatGPT’s case, the LLM consists of over 300 billion words, mostly scraped from the internet.

So then, what’s this term “generative AI”? It’s a type of artificial intelligence technology that can produce various types of content, including text, imagery, audio and synthetic data. ChatGPT is a popular form of generative AI, as are Google Bard and many other generative AI tools available online either for free or at a price.

So many use cases, but this one will blow your mind

ChatGPT is, essentially, a content creation, “ask me anything” type of tool. People use ChatGPT for first draft product descriptions, blog posts, social media posts, drafts of new business ideas or even drafting of entire articles. It also works well for language translation, conversational AI chatbots, writing code (even malware), policy writing, courses and educational content creation, help with your kid’s homework, research assistance, customer support and even surveys. Have a look at what the free version of ChatGPT came up with when I asked it to write me a fraud risk management survey. (See Figure 1 below.)

Figure 1
Source: OpenAI

Interestingly, you typically won’t get the exact same answer twice to the same question, and I’m told that the paid version of ChatGPT produces better, or more, results than the free version.

All of these “unstructured” data use cases are cool and rather impressive. But being focused on innovation, I wanted to take it a step further. What happens if we point generative AI tools to a structured data source like accounting data? Could we use it to help us identify potentially improper, corrupt or fraudulent payments to vendors? Could it identify high-risk employees abusing travel-and-entertainment expenses, or high-risk customers or distributors via a company’s sales transactions?

Instead of clicking around in your Tableau or PowerBI dashboard, could an AI chatbot do the navigation for you — seamlessly directing you to high-risk transactions, such as those listed in Figure 2 below, and provide recommendations on where to go next?

The answer is a mind-blowing “yes!”

Figure 2
Source: Kona AI

By now, all of us are familiar with data visualization tools like Tableau, Qlik, Spotfire or PowerBI. Visualizing data, instead of looking at a spreadsheet via rows and columns, has changed the game in fraud prevention and detection analytics. Users click around dashboards to explore and filter through data to find high-risk vendors, customers or employees in their fraud-fighting and compliance monitoring efforts. But now, we can predict that the future will be an overlay of such dashboards using generative AI tools in the form of easy-to-use, yet sophisticated, chatbots. These chatbots will suggest high-risk activities to the end user and drive the dashboard toward where they need to go, then make intelligent (or intuitive) recommendations on what the user should do next. Users will simply give the chatbot tasks or questions such as:

• Show me my highest-risk vendors, customers or employees.
• Show me any invoice with the word “special payment” in the description, in Argentina, within the last 60 days.
• Find me similar transactions to X (those selected transactions).
• Which employees are at the highest risk of paying a bribe, and why?

See Figure 3 below as an example. You can also visit the “How It Works” webpage on my company’s website for a short narrative video (click the second video screen).

Figure 3: How generative AI chatbots will soon be navigating the dashboards for you. In this example, over 177,000 invoices are visually displayed in a dashboard, and the chatbot recommends the three highest-risk transactions to drill down into, offering suggestions on next steps.

Source: Kona AI

The dark side of generative AI and risks to avoid

With all the promises and excitement around generative AI, there are some sinister, and sometimes just plain negligent, uses of this technology that we should consider from a risk-management perspective. In preparing this column, I spoke with Andy Gandhi, global practice leader in data insights and forensics at Kroll. He advises his clients that any responsible use of generative AI tools should prevent employees from disclosing sensitive company information. He also says that the results of ChatGPT or any other generative AI should only be considered a first draft and never as an authoritative source. One recent case he points to involves a plaintiff’s attorney representing a man in a personal injury lawsuit in Manhattan. The lawyer found himself at the mercy of the court when he submitted a federal court filing that cited at least six cases that never existed. He’d used ChatGPT to generate the filing and didn’t understand its limitations. The cases cited were all bogus and generated from the bot. (See “Lawyer Uses ChatGPT In Federal Court And It Goes Horribly Wrong,” by Matt Novak, Forbes, May 27, 2023.)

“Clearly, you (not ChatGPT) are responsible for the content you create,” Gandhi explains. “ChatGPT may help get the creative juices flowing with quick and easy content generation, but you are indeed responsible for fact checking the results and adding your expertise and research if your name is on that paper. Don’t be like that plaintiff’s attorney who did not fact check.”

Just as generative AI accelerates content creation for creative and good use cases, it also has sinister uses, too. For example, ChatGPT can help fraudsters with little or no coding experience generate malware and provide advice on hacking techniques. (See “Is ChatGPT the newest gateway to fraud?” by Mary Breslin, CFE, Fraud Magazine, May/June 2023.)

It’s an exciting time for fraud examiners

Just over a decade ago, we were blown away when software tools like Tableau became popular for use in anti-fraud and anti-corruption investigations and compliance monitoring. These tools transformed spreadsheets and databases into beautiful, interactive dashboards to help us find fraud and corruption in our vendor, customer and employee datasets. Now, with generative AI tools, our minds are about to be blown away again. Instead of clicking around dashboards, chatbots will use generative AI to help us navigate, find potentially improper payments, make recommendations on next steps and continuously learn.

In my view, generative AI will soon help investigators with about 70% to 80% of the high-level review process. Just like an alarm on a smoke detector starts ringing at the slightest sign of a potential fire, this new technology will quickly raise red flags for CFEs in search of any signs of fraud. But it won’t remove humans from the equation. The judgment of competent CFEs is now, and will continue to be, the most important part of the fraud risk management and investigative process.

Vincent M. Walden, CFE, CPA, is the CEO of Kona AI, an AI-driven anti-fraud and compliance technology company providing easy-to-use, cost-effective payment and transaction analytics software around corruption, investigations, fraud prevention and compliance monitoring. He welcomes your feedback and ideas. Contact Walden at vwalden@konaai.com.