Home > Risk > Auditors, don’t rush to use AI!

Auditors, don’t rush to use AI!

February 19, 2024 Leave a comment Go to comments

I find it frustrating to see so many so-called experts calling for internal auditors to grab the latest technology, the latest being generative artificial intelligence (AI), and find a way to use it.

Let me tell you this: artificial intelligence is not always that intelligent!

Let me also tell you that you should only use technology when you know how it will be used and that there will be a positive Return on your Investment (ROI).

Taking that last point first, several years ago I was invited to meet the leaders of a large internal audit department in Detroit. (I will let you guess who.) They introduced the two leaders of their new Data Analytics team, who explained that they had purchased some great tools, hired several experts in their use, and were now starting to build analytics routines. They started with Accounts Payable and then the General Ledger. But they had no idea how the rest of IA would use their reports. It was a matter of buying a great hammer and then looking for nails without thinking about what needed to be built.

Now for the less-than-intelligent generative AI.

When ChatGPT first came out, like many others I decided to download and test it. One of the first things I did was to ask it who I was. It returned a decent description of my professional history, but nothing on my personal life (and my mother[1], brother[2], and son[3] are all prestigious individuals). OK. So I tried the same question again to see if the results would be different. They were. It created a fictional version of me.

Recently, Microsoft came out with their AI, CoPilot. I tried the same test to see what I would tell me about myself. This is what it said:

Norman David Marks is a retired executive with expertise in corporate governancerisk managementinternal auditenterprise performance, and the value of informationHe has also served as a coach and mentor to individuals and organizations worldwide1. His contributions span various domains, including:

  1. Auditing: Norman authored the book “Auditing that Matters: Case Studies,” which provides internal audit practitioners and students with real-life case studies for discussion and debate. These cases help upgrade understanding of world-class internal auditing practices12.
  2. Chief Audit Executive (CAE): With experience overseeing companies with up to $28 billionin annual revenue, Norman has implemented risk management, ethics programs, and disclosure processes3.
  3. Businessman: There is also an English management consultant and businessman named David Mark Norman, who hails from a prominent City dynasty and is now retired from executive positions4.

In summary, Norman David Marks has made significant contributions to the fields of auditing, risk management, and corporate governance, leaving a lasting impact on the industry.

While it only mentions what I would call one of my lesser books, it’s not too bad.  It misses a lot of information about me that it should have been able to find easily, but at least it is me. One thing I like is that it provided references for the information it provided.

Would I recommend the tool to anybody looking for information on a potential new hire? No. It is incomplete and what is missing is, in my case, material to the decision.

I decided to test CoPilot a second time. This time, instead of asking the same question, I asked about my wife, Diana Bee Har Cheah (using her maiden name).

This is what it had to say about her:

Diana Bee Har Cheah is a distinguished Professor Dr. specializing in Breast Surgery at the University of Malaya in Kuala Lumpur, Malaysia. Her expertise lies in breast cancer management, research, and clinical outcomes. Let’s explore some of her notable contributions:

  1. Association Between Cardiometabolic Comorbidities and Breast Cancer Profiles:
  2. Clinical Outcomes in Early Breast Cancer Patients Treated with Intraoperative Radiotherapy:
  3. Impact of COVID-19 on Breast Cancer Management:
  4. Metastatic Malignant Thymoma Presenting as a Breast Lump:
  5. Exploring Health Information-Seeking Practices of Breast Cancer Patients:

Her work contributes significantly to breast cancer research and patient care, making her a respected figure in the field of breast surgery and oncology.

This is not my wife!

I showed it to her, and she was very surprised that there was someone with the same name as her.

But there isn’t!

The AI found a university in Malaysia where some on the faculty have part of her name, and others have the rest.

It generated a profile by combining the resumes of different people for a fictional professor with my wife’s name.

Now, in some situations her name is shown in a different order, with Diana as a middle rather than a first name: Bee Har Diana Cheah instead of Diana Bee Har Cheah.

CoPilot gave us this:

Bee Har Diana Cheah, also known as Madam Cheah Bee Chew, is the remarkable founder of Bee Choo Origin. Established in 2000 in Singapore, Bee Choo Origin is renowned for its herbal treatment for hair. Let’s delve into her inspiring journey:

  • Mission and Vision:
    • Madam Cheah was determined to find an affordable 100% natural remedyfor common hair issues such as grey hairhair loss, and scalp-related problems.
    • Guided by Traditional Chinese Medicine (TCM) physicians, she embarked on a journey of learning, experimentation, and concoction.
  • Signature Herbal Cream:
    • After extensive research, Madam Cheah developed what is now known as the “Bee Choo Origin Signature Herbal Cream.”
    • This herbal treatment, made from premium quality herbs, has become the cornerstone of Bee Choo Origin’s success.
  • Growth and Transformation:
    • From humble beginnings as a single woman working from home with a stand-alone steamer, Bee Choo Origin has expanded globally.
    • With over 170 storesworldwide, the brand continues to focus on continuous improvement, blending modern technologies with two decades of expertise.
  • Core Values and Beliefs:
    • The brand’s logo reflects its core values:
      • Genuineness: Authenticity builds trust with customers and partners.
      • Excellence: Dedication to delivering the highest quality hair care solutions.
      • Care: Customers are at the heart of everything Bee Choo Origin does.
    • The flower bud in the logo symbolizes new beginnings, growth, and the power of nature to renew and restore.

Madam Cheah Bee Chew’s commitment to transformative hair care has made Bee Choo Origin a global brand, providing natural solutions for confident, healthy hair since 200012.

Again, MS CoPilot made a serious mistake. It found someone with a similar name and assumed (and demonstrated what ‘assume’ means) it was the same person.

So, don’t feel bad if you haven’t found a use for AI in your practice.

If you have:

  1. Make sure you can trust it.
  2. Make sure it will return a positive ROI.

I welcome your thoughts.

POSTSCRIPT:

Alexander Ruehle challenged me, as he should, saying that auditors should rush to AI and learn how to use it. With that in mind, I searched for pieces on the form of AI called LLM, large language models, and found this: https://deepchecks.com/top-5-risks-of-large-language-models/

This is how it starts:

While they offer great promise in terms of providing instant and accessible information, there are inherent risks associated with their potential to disseminate incorrect or false information.

One of the primary concerns surrounding LLMs is their susceptibility to generating incorrect information. These models generate responses based on patterns and associations learned from training data but lack true comprehension and fact-checking capabilities. Consequently, they may produce inaccurate or misleading information, especially when confronted with ambiguous or contextually nuanced queries. Users who rely solely on these responses without independent verification are at risk of being misled and propagating misinformation.

Furthermore, LLM and ChatGPT are only as reliable as the data they are trained on. If the training data contains incorrect or false information, the models may inadvertently perpetuate those inaccuracies. This highlights the need for rigorous data selection and preprocessing to mitigate the risks associated with training language models.

I encourage you to read and consider the rest of the article. If you want, use your favlorite AI tool to seek information on AI-related risks.

====================================================================

[1] She has an MBE for her professional and charitable works.

[2] He is a recently retired professor emeritus at Northwestern University with multiple awards and honors.

[3] He is a partner with a major law firm.

  1. djallc
    February 19, 2024 at 10:22 AM

    Like with all new technology, I have never believed living on the “bleeding edge” is the best place to be. It is rarely an efficient use of effort and frequently is an exercise of self-promotion (“look how great we are using the shiny new toy”). Planned, measured, long-term investment is a better path.

    P.S. While I can guess the Detroit company (knowing the culture of most of the big ones in Detroit) this story probably applies to way too many companies…

  2. Anonymous
    February 19, 2024 at 10:59 PM

    Excellent examples, as always. I think you are absolutely right, and also the ones with the opposite view are also right in a lot of what they say. Essentially I think both sides are right (but you are more right). Matter of emphasis. I do prefer cautious realistic sober sensible cognisant embracing but on the other hand not too hesitant and slow. The revolution is happening. We ignore it for too long at our peril. But we dive in superficially also at our – perhaps potentially much bigger – risk.

    • Norman Marks
      February 20, 2024 at 9:07 AM

      You are right

  3. Anonymous
    February 20, 2024 at 8:29 AM

    Great article, as always. I can see both sides of the topic. While I am old-school and prefer the older methods of sampling and feeling in control of that sample, I do realize one person is only human and cannot possibly sample all data manually. There needs to be technical assistance for testing. That being said, the technology being used needs to be understood by the person using it. AI, generative AI, LLM, robotic process automation, etc. are all the buzz words now. How can those tools assist the internal audit function? I am an auditor. I perform testing based upon sampling methods. I draw a conclusion and discuss with management. Now comes the subjective portion of the IA function. What level of risk do the test results fall under? What mitigants exist to lessen the risk level? Can AI help with this process? I do not think so. In the examples given, AI could not provide the correct description of persons input and the question was one basically built upon fact. Drawing a conclusion from audit results is sometimes highly subjective. Technology cannot be subjective. It only knows facts. It cannot see the grey of any situation. So, why should an auditor rely upon AI? Only if you know the results for questions are accurate should you rely on AI. But, there again, what auditor has the time to validate everything that ChatCPT and other tools spit out?

  4. Anonymous
    February 21, 2024 at 5:52 AM

    Another good assessment of risk. But while AI may not be ready for implementation in our audit processes, we should be reading and studying the technology so we can discuss the risks with the business, as they are more likely to implement AI before auditors get on board.

  1. No trackbacks yet.

Leave a comment

This site uses Akismet to reduce spam. Learn how your comment data is processed.