Managing AI Risks
Artificial Intelligence (AI) is the “next best thing” in technology – and for good reason. Simply put, AI promises to allow team members to work more efficiently and effectively. Further, AI can help relieve team members from performing tedious, time-consuming tasks that do not challenge or fulfill them. However, despite these advantages, AI can pose significant risks, too. Read on, and you will learn about managing AI risks in this article.
A Primer On Generative AI
Although multiple forms of AI exist, generative AI is the tool that most professionals focus on today. As its name implies, generative AI can generate responses to questions known as “prompts” that you pose. For example, suppose you wanted a quick summary of recent tax law changes. In that case, you could “prompt” a tool such as ChatGPT to “prepare a summary of recent tax law changes.” Upon issuing the prompt, the platform creates a customized response. Further, you could add additional prompts such as “provide more details.” When you do, the system responds with – you guessed it – a more detailed analysis.
The simple example provided above only scratches the surface of all the ways you can interact with your AI platform to get the information you need. Notably, the prompts you issue can ask for content perhaps beyond your wildest imagination. For example, you can use generative AI tools to write speeches, draft emails, generate drawings, review tax returns for errors and irregularities, conduct fraud tests, and even analyze a company’s financial statements for opportunities to improve performance. Unfortunately, as you start to use AI for some of these tasks, you potentially expose your data – or that of a client – to risks you may not have realized exist.
What Are AI’s Biggest Risks?
Despite their amazing capabilities, let’s be clear: generative AI tools are not perfect and perhaps never will be. Whenever you issue a prompt to your AI platform, you assume some risks, including those listed below.
· Hallucinations, the risk that your AI platform makes a mistake and responds incorrectly to your prompt, often for reasons you may never understand.
· Biased outputs, the risk that a lack of relevant data or insufficient training of the data in your AI platform leads to a faulty
answer.
· Copyright infringement issues may arise when AI users fail to provide appropriate citations and attribution for using AI-generated content.
· Data privacy and security issues can arise if you fail to understand your license agreements or your AI platform
fails to adequately and appropriately secure your data.
Notably, to be effective with AI, you will also need to be effective at managing AI risk. Following are some examples of managing AI risks.
Hallucinations
Let’s face it…you and I are not perfect. We make mistakes, likely daily. Although we would like to think that our AI tools are immune from errors, they are not. Like us, AI tools make mistakes, known as hallucinations. Hallucinations occur for a variety of reasons. Among these are misinterpreting prompts, insufficient training, and relying on outdated data. Regardless of the cause, you are left with an incorrect response.
To guard against relying on a hallucinated response, you may consider issuing prompts to multiple AI platforms. For example, consider issuing the same prompt to ChatGPT and Gemini. If you receive similar responses from both platforms (which use different LLMs), then you likely can rely on the responses as being accurate. However, if the responses are different, you likely need to dig further to find which – if either – is correct. Remember, your credibility is on the line! Do your due diligence, and don’t become a victim of a hallucination!
Biased Outputs
In some respects, biased outputs resemble hallucinations – that is, they represent faulty responses. However, the cause of biased outputs differs significantly from the cause of hallucinations. Specifically, biased outputs occur when humans fail to train AI data adequately. Additionally, biased outputs can arise when the AI platform does not contain sufficient representative data – another form of human error. To overcome bias – regardless of cause – those responsible must ensure that human bias or error does not creep into the platform.
Copyright Infringement Issues
Intellectual property rights have been under siege for years, and AI only increases these risks. It is far too easy for an AI user to prompt a platform for a response. Then, the person who issued the prompt uses the response without appropriate citations, attribution to the original author(s), or compensation. Not only does this issue represent an AI risk, but it also represents a legal risk. Specifically, failing to cite sources could trigger a lawsuit. Therefore, perhaps the best way of addressing this risk is to follow the motto “Better safe than sorry.” Specifically, err on the side of caution and cite your sources – even if they are AI-based – to minimize the risk that you are named a defendant in a copyright infringement lawsuit.
Data Privacy And Security Risks
The fourth and final risk to cover in this article is data privacy and security risk. While these items are technically separate, we will treat them as one. Ask yourself the simple question: is my AI provider “looking” at my data and prompts? Today’s reality is that most AI providers are tracking and retaining a detailed history of your prompts, just as most search engine providers are doing the same. Are you comfortable with that? If not, at a minimum, carefully consider the types of prompts you issue. Likewise, never upload private data – bank account numbers, Social Security numbers, etc. – to an AI platform. Both these actions can go a long way when managing AI risks.
To be more proactive, consider deleting your prior interaction history. For example, in ChatGPT, log in to your account. Then click your picture near the upper right corner of the window. Next, click Settings. Then locate Delete all chats near the bottom of the window and click Delete all on the window’s right side. Doing so permanently removes all your previous history. Additionally, note that most other AI platforms provide similar options.
Summary
AI offers so much promise and will be an essential productivity tool for almost all business professionals soon, if not already! However, like most technologies, AI brings risks. Specifically, hallucinations, biased outputs, copyright infringement issues, and data privacy and security risks are here already. And that’s just for starters… other risks will almost certainly materialize. Like with malware, ransomware, and other technology-related risks, tools, techniques, and procedures will no doubt evolve to help us when managing AI risks.
At K2 Enterprises, our commitment lies in providing unwavering support and expert instruction to CPAs. Explore the wealth of resources on our website, where you’ll find valuable insights on selecting the most suitable accounting software, ensuring your firm is equipped with the right tools for the journey ahead. If you work in accounting or finance, K2 Enterprises provides continuing education programs to enhance your skills and credentials. Need help learning how to solve your business’s accounting technology needs and selecting the right software for accounting or CPA Firms? Visit us at k2e.com, where we make sophisticated technology understandable to anyone through our conferences, seminars, or on-demand courses.
