ChatGPT and Generative AI in Accounting
ChatGPT and Generative AI in Accounting
t
os
rP
XU LI
RAMÉE LIU
yo
The release of ChatGPT, a powerful, generative artificial intelligence (AI) model, heralded the
start of a revolution in the way people worked across different sectors. 1 Through an easily
accessible chatbot interface, this breakthrough in AI technology allowed ChatGPT and other
generative AI models to automate a range of work processes, saving valuable time and
resources. These powerful AI technologies could generate output such as text, audio, computer
code, images, simulations, and videos. In 2023, generative AI was starting to transform the way
professionals worked in the accounting industry, enabling them to streamline auditing and
op
accounting processes, leverage natural language processing (NLP), and search information
more efficiently.2
The big-four accounting firms had been investing heavily in developing their own AI solutions
to improve efficiency and to provide innovative solutions to their clients. In early 2023, with
the widespread adoption of ChatGPT, some of the big-four accounting offices chose to prevent
their staff from using generative AI models for their clients’ work, while other professional
tC
firms embraced using generative AI models and provided guidance and training to their staff to
help them learn the necessary skills.3 As the use of generative AI continued to grow quickly in
2023, businesses and organizations that were unprepared for this rapid automation of new skills
were still weighing the risks and benefits of using AI to determine how to best utilize this
technology.
Joesy Loh, an audit partner of a CPA firm for SMEs in Singapore, was in favor of incorporating
generative AI to improve efficiency in audit work and the preparation of financial statements.4
No
Prior to an in-house meeting with other partners and senior staff, she was creating a list of
advantages and strategies to avoid the potential drawbacks of using generative AI. How could
1 Sabrina Ortiz, “What is ChatGPT and why does it matter? Here's what you need to know,” ZDNET, 21 July 2023,
https://www.zdnet.com/article/what-is-chatgpt-and-why-does-it-matter-heres-everything-you-need-to-know/ accessed 21 July
2023.
2 Sabrina Ortiz, “What is ChatGPT and why does it matter? Here's what you need to know,” ZDNET.
3 Edmund Tadros, “PwC warns staff against using ChatGPT for client work,” afr.com, 6 February 2023,
https://www.afr.com/companies/professional-services/pwc-warns-against-using-chatgpt-for-client-work-20230205-p5chzi,
accessed 23 August 2023.
4 Joesy Loh is a fictitious name.
Do
Ramée Liu prepared this case under the supervision of Dr. Xu Li for class discussion. This case is not intended to show effective
or ineffective handling of decision or business processes. The authors might have disguised certain information to protect
confidentiality. Cases are written in the past tense, this is not meant to imply that all practices, organizations, people, places or
fact mentioned in the case no longer occur, exist or apply.
© 2023 by The Asia Case Research Centre, The University of Hong Kong. No part of this publication may be digitized, photocopied
or otherwise reproduced, posted or transmitted in any form or by any means without the permission of The University of Hong
Kong.
Ref. 23/774C
Joesy persuade other partners, who were significantly worried about the accuracy of the
t
generated output and the data security of their firm and clients, to employ this new technology?
os
Introduction of Generative AI
AI was the branch of computer science that created programs and machines that could perform
tasks that normally required human intelligence.5 Machine learning was a subfield of AI that
enabled computers to learn from data without being explicitly programmed.6 It had been applied
rP
in various domains, including office work, where it could automate business and accounting
processes.7
Generative AI
Google Cloud Tech explained that generative AI was a type of AI that created new content
based on its learning from existing content.8 Generative AI was considered a subset of deep
learning models within the field of machine learning. 9 Unlike traditional machine learning
yo
models that provided output in the form of numbers, discrete answers, classifications, and
probabilities, generative AI had the ability to process more complex patterns and generate
output such as natural language, images, computer code, videos, and audio.1011 This facilitated
direct interaction between humans and machines.
Users were able to communicate with generative AI chatbots to search for general information
and receive responses in simple or complex formats, eliminating the need to refer to a list of
websites on search engines such as Google or Bing. They could chat and communicate directly
op
with generative AI models using text and commands without requiring any software or
programming skills.12 In more advanced versions, users could upload text files, images, or other
formats to engage with chatbots.
Generative AI models provided immediate responses and interactions with users through their
NLP abilities.13 These models were trained to process natural language inputs and produce text
tC
outputs for various functions such as translation, summarization, content generation, and
classification. Generative AI models with NLP capabilities were often available for free or
could offer faster responses through paid subscriptions or through certain platforms. This
offered cost savings compared to purchasing AI software and saved time and labor for
individuals or businesses across various tasks.
Between 2020 and early 2023, venture capital firms invested over USD1.7bn in generative AI
solutions, with a significant focus on AI-enabled drug discovery and AI computer software
No
5 Xu Li and Ramée Liu, “Artificial Intelligence in Accounting,” ACRC, The University of Hong Kong Business School, 2020,
https://www.acrc.hku.hk/Case/Detail/1058, accessed 31 July 2023.
6 Google Cloud Tech, “Introduction to Generative AI,” YouTube, 9 May 2023,
This document is authorized for educator review use only by Dr.K.Hema divya Kantamaneni, KL University until Mar 2024. Copying or posting is an infringement of copyright.
[email protected] or 617.783.7860
23/774C ChatGPT and Generative AI in Accounting
t
chip and manufacturing parts design, material science, and synthetic data generation.15
os
Generative AI Models for Business-Related Use
Use of Publicly Available Generative AI Models
Business users frequently utilized publicly available generative AI models to submit their own
rP
data or queries. These models were often equipped with NLP capabilities, enabling them to
extract and summarize information from contracts, documents, and websites. This functionality
proved particularly valuable for text writing, enhancing professional writing styles, and offering
translation services. Furthermore, these generative AI models could assist in the creation of
PowerPoint presentation slides by “copying and pasting” data within chatbot interfaces.
However, a potential concern arose when various business units shared their data with these
generative AI models, as their data could possibly be incorporated into the training data sets,
yo
potentially leading to corporate knowledge leaks and data security issues for the companies and
their clients [see Figure 1]. Although some developers of publicly available generative AI
models claimed to prioritize data security and assured users that the received data were not used
for training their models (such as ChatGPT), the lack of strict government regulations on AI
developers in many countries made it challenging for users to verify the data security practices
of these AI developers.
Figure 1: Commonly Used Generative AI Solutions
op
tC
No
ChatGPT
OpenAI, an AI research company founded in 2015 and led by Sam Altman, developed
ChatGPT.16 OpenAI was also responsible for creating Dall-E, a system that generated images
based on text descriptions. Microsoft was an important investor and partner of OpenAI.17
Do
14 Jackie Wiles, “Beyond ChatGPT: The Future of Generative AI for Enterprises,” Gartner, 26 January 2023,
https://www.gartner.com/en/articles/beyond-chatgpt-the-future-of-generative-ai-for-enterprises, accessed 1 August 2023.
15 Jackie Wiles, “Beyond ChatGPT: The Future of Generative AI for Enterprises,” Gartner.
16 “OpenAI,” Wikipedia, https://en.wikipedia.org/wiki/OpenAI, accessed 13 September 2023.
17 Sabrina Ortiz, “What is ChatGPT and why does it matter? Here's what you need to know,” ZDNET.
This document is authorized for educator review use only by Dr.K.Hema divya Kantamaneni, KL University until Mar 2024. Copying or posting is an infringement of copyright.
[email protected] or 617.783.7860
23/774C ChatGPT and Generative AI in Accounting
The name ChatGPT was “generative pre-trained transformer.” GPT referred to a large-scale
t
NLP model known as a large language model (LLM), which utilized deep learning techniques
os
to generate text that resembled human language.18
On 30 November 2022, OpenAI released a powerful version of generative AI called ChatGPT-
3.5 (version 3.5), which was available to the public for free.19 As a spearhead for generative AI,
ChatGPT-3.5 amassed 100 million sign-ups within two months.20 ChatGPT-3.5 could search
for general information and generate detailed responses to complex user queries. However, its
rP
database was limited to events that occurred before September 2021.21 Additionally, OpenAI
did not provide service coverage directly to users in mainland China, Hong Kong, Macau, and
a few other regions or countries.22
In February 2023, ChatGPT gained the ability to connect with a range of plug-in apps and other
software using application programming interfaces (APIs) or computer coding. 2324 This
allowed integration with applications such as Microsoft Excel spreadsheets, email, search
engines, and messaging platforms such as WhatsApp.
yo
In March 2023, a more powerful version of ChatGPT, called ChatGPT-4, was introduced.25 It
offered higher accuracy, the capability to process lengthy text, and image recognition
capabilities.26 Similar to its predecessors, ChatGPT-4 was trained on data up until September
2021 and did not have direct access to the internet for updated information. 27 Notably,
ChatGPT-4 demonstrated an improvement in understanding accounting concepts, as it achieved
a 78% score on the US Certified Public Accountant (CPA) exam, an achievement that
ChatGPT-3 had failed.28 In the same month, a paid subscription option called ChatGPT Plus
op
was launched, offering faster response times and internet access at a cost of USD20 per month.29
OpenAI also revealed that a more advanced AI model, ChatGPT-5, was currently in
development. On 29 August 2023, OpenAI introduced ChatGPT Enterprise, a version
specifically designed for businesses. Companies needed approval from OpenAI to utilize this
enterprise version of ChatGPT.30
tC
18 Jackie Wiles, “Beyond ChatGPT: The Future of Generative AI for Enterprises,” Gartner.
No
19 Sabrina Ortiz, “What is ChatGPT and why does it matter? Here's what you need to know,” ZDNET.
20 Sabrina Ortiz, “What is ChatGPT and why does it matter? Here's what you need to know,” ZDNET.
21 Sabrina Ortiz, “What is ChatGPT and why does it matter? Here's what you need to know,” ZDNET.
22 Jon Martindale, “These are the countries where ChatGPT is currently banned,” digitaltrends, 12 April 2023,
27 Sabrina Ortiz, “What is ChatGPT and why does it matter? Here's what you need to know,” ZDNET.
28 “ChatGPT and accountancy-know its limits-ICAEW,” ICAEW Insights, 22 March 2023,
https://www.icaew.com/insights/viewpoints-on-the-news/2023/mar-2023/chatgpt-and-accountancy-know-its-limits, accessed
23 March 2023.
29 Sabrina Ortiz, “What is ChatGPT and why does it matter? Here's what you need to know,” ZDNET.
30 Steve Hook, “ChatGPT Enterprise – What is it?,” PC Guide, 29 August 2023, https://www.pcguide.com/ai/chatgpt-enterprise/,
This document is authorized for educator review use only by Dr.K.Hema divya Kantamaneni, KL University until Mar 2024. Copying or posting is an infringement of copyright.
[email protected] or 617.783.7860
23/774C ChatGPT and Generative AI in Accounting
ChatGPT-4 to create Bing Chat, a chatbot. Starting in May 2023, users worldwide gained access
t
to Bing Chat using their Microsoft accounts, with certain limitations on access.31
os
Meanwhile, Google, after years of developing various generative AI models, introduced Bard
in March 2023.32 Bard was a chatbot that integrated Google’s search function and was made
available for public use in the US. Google continued to experiment with Bard’s release in
different countries.33 Furthermore, Google embarked on a new AI project called Gemini, which
aimed to redefine the AI industry with its multimodal capabilities, and posed a challenge to
rP
ChatGPT-4 in the future.34
In the first half of 2023, the generative AI landscape saw the emergence of over 100 chatbots
developed by various companies, predominantly from the US. 35 Some notable examples
included Falcon, PaLM (Pathways Language Model), Claude II, and Sage.36
Notably, in March 2023, Baidu, a prominent Chinese tech company, introduced its chatbot
system called Ernie Bot.37 Initially, Ernie Bot was made available to selected business clients
yo
in mainland China. It had a more localized approach, utilizing the Chinese language and
incorporating local data. On 31 August 2023, Ernie Bot became open to the public in mainland
China after successful registration as Baidu users.38 Alongside Baidu’s offering, other Chinese
tech giants such as Alibaba, Tencent, and Huawei also provided their own generative AI chatbot
solutions for domestic use. 39 However, the use of generative AI models developed outside
mainland China was not permitted in the mainland. Similarly, overseas search engines were
also subject to restrictions and were not accessible in mainland China.
op
Copilot Generative AI for Business Use
Microsoft 365 Copilot was an innovative generative AI tool to command AI software. 40 It
functioned as a chatbot and had the capability to extract information directly from files stored
on personal computers or servers, generating the desired output based on user requests. Users
of Copilot had the advantage of accessing data from various applications, including Microsoft
tC
Word, Outlook, Excel, PowerPoint, and MS Teams. This extensive integration allowed Copilot
to automate tasks and generate content, potentially saving users time and boosting productivity.
By accessing office files, it could assist users in analyzing data, identifying patterns, and
predicting trends.
31 Maria Diaz, “How to use Bing Chat (and how it's different from ChatGPT),” ZDNET, 21 July 2023,
No
https://www.youtube.com/watch?v=gwEuvrI4fx4&list=PLVsxvnj4Med4JKzao96hKoOdjn6TW2mZi&index=15, accessed 8
July 2023.
35 Google Cloud Tech, “Introduction to Generative AI,” YouTube.
36 Google Cloud Tech, “Introduction to Generative AI,” YouTube.
37 Cheyenne Dong, “Baidu launches ChatGPT rival ERNIE Bot, sees 30,000 companies apply to use it in first hour,” technode,
38 “百度 AI「文心一言」今日起向公眾開 (Baidu AI “Ernie Bot” is open to the public today),” 881903.com, 31 August 2023,
https://edition.cnn.com/videos/business/2023/08/01/exp-china-ai-regulation-lu-stout-pkg-08012aseg2-cnni-business.cnn,
accessed 1 August 2023.
40 Jared Spataro, “Introducing Microsoft 365 Copilot – your copilot for work,” Microsoft Blogs, 16 March 2023,
https://blogs.microsoft.com/blog/2023/07/18/furthering-our-ai-ambitions-announcing-bing-chat-enterprise-and-microsoft-365-
copilot-pricing/, accessed 20 August 2023.
This document is authorized for educator review use only by Dr.K.Hema divya Kantamaneni, KL University until Mar 2024. Copying or posting is an infringement of copyright.
[email protected] or 617.783.7860
23/774C ChatGPT and Generative AI in Accounting
On 17 March 2023, Microsoft launched the Microsoft 365 Copilot for a list of corporate clients
t
and claimed it had the potential to “revamp how office work was done.” 41 In July 2023,
os
Microsoft 365 Copilot announced a pricing structure for business clients using Microsoft 365.42
The service was priced at USD30 per month per user.43 But Gartner, a leading IT research firm,
reported in June 2023 that Microsoft 365 Copilot was not yet ready for enterprise use.44
rP
created models specifically tailored for business use. These models implemented stringent
measures to ensure the protection of sensitive data. One of the key practices was the strict
separation between the data uploaded to generative AI models and the data used to train those
models. This separation ensured that the uploaded data did not contribute to the training process,
thereby preserving the confidentiality of the company’s information. Also, developers
incorporated advanced technologies such as data encryption and data anonymization to enhance
data security [see Figure 2].
yo
Looking ahead, the trend indicated a shift toward localized generative AI models within specific
business units. This localization approach aimed to further strengthen data security and privacy,
aligning the AI models more closely with the unique needs and requirements of individual
businesses. For generative AI models available for business and accounting use in August 2023,
see [Exhibit 1, Generative AI Models for Business and Accounting Use].
41 Microsoft, “Introducing Microsoft 365 Copilot with Outlook, PowerPoint, Excel, and OneNote,” YouTube, 17 March 2023,
Do
https://www.youtube.com/watch?v=ebls5x-gb0s&list=PLVsxvnj4Med4JKzao96hKoOdjn6TW2mZi&index=33, accessed 8
July 2023.
42 Yusuf Mehdi, “Furthering our AI ambitions – Announcing Bing Chat Enterprise and Microsoft 365 Copilot pricing,”
Microsoft Blogs.
44 Jared Spataro, “Introducing Microsoft 365 Copilot – your copilot for work,” Microsoft Blogs.
This document is authorized for educator review use only by Dr.K.Hema divya Kantamaneni, KL University until Mar 2024. Copying or posting is an infringement of copyright.
[email protected] or 617.783.7860
23/774C ChatGPT and Generative AI in Accounting
t
os
Before generative AI became available to the public, many accountants and auditors had been
using different forms of AI in their work, such as robotic process automation (RPA)
applications and optical character recognition (OCR) technology, and also accounting software
that had built-in AI functions, see [Exhibit 2, using AI in accounting before generative AI].45
[For some of the examples in accounting tasks using AI, see Exhibit 3, AI for accounting tasks.
rP
Navigating Accounting Standards with Generative AI
When auditors and accountants utilized the commands in ChatGPT effectively, while providing
sufficient background and context, the model had the potential to assist in locating answers
related to accounting standards, guidelines, tax regulations, and even relevant accounting
entries. While ChatGPT could offer valuable insights, accountants had to be diligent in
verifying the accuracy and reliability of the answers provided. It was advisable to approach the
responses as a starting point or guide for further investigation and research, rather than
yo
accepting them blindly.
ChatGPT-4 had successfully passed the US CPA exam with a score of 75%, indicating that the
results it generated were not flawless. While ChatGPT-4 provided source citations as part of its
output, it was essential to exercise caution and independently verify the information from those
citations. In addition, generative AI models such as ChatGPT-4 were not specifically trained to
comprehend complicated accounting and tax scenarios tailored to a specific company or client.
op
Therefore, utilizing ChatGPT-4 for client- or customer-facing activities without considering the
potential risks involved was not advised.46
reports, and website information for potential or existing customers, identifying any red flags.
On 30 March 2023, Bloomberg announced the development of BloombergGPT, a LLM,
boasting decades of collected data to enhance financial research capabilities. 47 For more
comprehensive client information in the future, BloombergGPT could be utilized.
Sheets, offering assistance in generating formulas, functions, and features such as pivot tables
and VBA coding.48 Additionally, the chatbot system had the capability to extract data from PDF
files or websites and generate tables for easy integration into spreadsheets. This functionality
was useful in summarizing data from financial results of listed companies, saving time for users.
To further streamline the process, ChatGPT could be integrated into Excel spreadsheets,
facilitating faster responses and workflow efficiency. 49
45 Xu Li and Ramée Liu, “Artificial Intelligence in Accounting,” ACRC, The University of Hong Kong Business School, 2020.
Do
This document is authorized for educator review use only by Dr.K.Hema divya Kantamaneni, KL University until Mar 2024. Copying or posting is an infringement of copyright.
[email protected] or 617.783.7860
23/774C ChatGPT and Generative AI in Accounting
US government regulators, such as the Securities and Exchange Commission (SEC) and the
t
Internal Revenue Service (IRS), had been using AI technologies, including data analytics, to
os
detect irregularities in financial transactions and discover fraudulent activities. The IRS
deployed AI to uncover tax evasion and was “opening examinations into large hedge funds,
private equity groups, real estate investors, and law firms.”50
Generative AI also assisted auditors and accountants in data analysis and fraud detection in
financial transactions. By analyzing data and identifying patterns, it enabled the prediction of
rP
future financial trends, aiding businesses in making informed decisions, such as in investments
and pricing strategies. In addition, it assisted in identifying unusual or suspicious behavior, such
as unauthorized access or abnormal spending patterns, contributing to compliance efforts and
preventing fraud or policy violations before they escalated.
yo
information by automating time-consuming tasks associated with record-keeping. This
innovative technology could assist in streamlining processes such as data entry, reconciliation,
and reporting.
ChatGPT-4 could also recognize images; therefore accountants could train the model to retrieve
the data on invoices, purchase orders, documents, and contracts to process in their enterprise
resource planning (ERP) and accounting software. By automating these tasks, generative AI
freed humans of the burden of tedious manual work and reduced the risk of errors.
op
Financial Reporting
The financial reporting process within ChatGPT comprised various steps, such as identifying
reporting requirements, collecting and preprocessing financial data, training the model,
generating the report, and conducting a thorough review and necessary editing. To explore
ChatGPT’s potential as an automation tool for financial reporting, see [Exhibit 4, ChatGPT in
tC
50 Alan Rappeport, “I.R.S. Deploys Artificial Intelligence to Catch Tax Evasion,” New York Times, 9 September 2023,
https://www.nytimes.com/2023/09/08/us/politics/irs-deploys-artificial-intelligence-to-target-rich-partnerships.html, accessed
13 September 2023.
51
Edmund Tadros, “PwC warns staff against using ChatGPT for client work,” afr.com.
52 Edmund Tadros, “PwC warns staff against using ChatGPT for client work,” afr.com.
This document is authorized for educator review use only by Dr.K.Hema divya Kantamaneni, KL University until Mar 2024. Copying or posting is an infringement of copyright.
[email protected] or 617.783.7860
23/774C ChatGPT and Generative AI in Accounting
generative IT tools within its teams, restricting wider usage. However, Deloitte was actively
t
testing these new technologies and allowed staff to access them on their work computers.
os
In the first half of 2023, small to medium-sized CPA firms were dealing with the risks and
limitations of adopting LLM for automation. Compliance concerns hindered widespread
adoption, and some partners, overwhelmed with their daily workload or the firm’s lack of
technical expertise, were reluctant to embrace new technologies. In addition, affordable and
easily accessible accredited generative AI systems that were equipped with advance data
rP
security were not readily available in the market until July 2023, when LLaMa 2 was launched,
which was free to the public [see Exhibit 1].
Accountants working with listed companies also expressed significant concerns regarding the
potential leakage of financial data. Given the sensitivity of their financial results to share prices
and compliance requirements, many accountants hesitated to utilize generative AI models. By
mid-2023, many listed companies had gradually adopted proper protocols and guidance
concerning data privacy and security issues, and started to use generative AI.53
yo
Generative AI–Equipped Accounting Software
In August 2023, “SAP AI’s” generative AI functions were being developed but had already laid
out a road map.54 In the meantime, some accounting and ERP software firms were working
with AI developers to integrate more AI or generative AI functions in their software.
Accounting software for SMEs, such as QuickBooks, Xero, and Zoho Books, had already added
chatbots that used NLP to provide personalized support and assistance to users.55 The chatbots
op
could help with tasks such as invoicing, expense tracking, and payment processing. [For
generative AI applications that could integrate with accounting software, see Exhibit 1.]
Other Professionals Using Generative AI
In March 2023, PwC announced a global partnership with AI startup Harvey. This partnership
granted exclusive access to the professionals at PwC’s Legal Business Solutions, enabling them
tC
to utilize an AI platform that combined NLP, machine learning, and data analytics to automate
legal work. 56 Alongside PwC, other established firms such as Allen & Overy (a law firm) and
Reach (a newspaper publisher) were also actively experimenting with this technology.57
By August 2023, over 80% of Fortune 500 companies had adopted ChatGPT in management,
communication, and creative work, according to Altman. 58 These users included selected
companies that were able to access ChatGPT Enterprise. Meanwhile, banks in Hong Kong
No
tightly controlled the use of generative AI or other software on work PCs, requiring prior
approval except for projects as part of “actual functionality.”59
56 “PwC announces strategic alliance with Harvey, positioning PwC’s Legal Business Solutions at the forefront of legal
This document is authorized for educator review use only by Dr.K.Hema divya Kantamaneni, KL University until Mar 2024. Copying or posting is an infringement of copyright.
[email protected] or 617.783.7860
23/774C ChatGPT and Generative AI in Accounting
t
os
The professional body of Chartered Accountants, ICAEW, stated that “generative AI such as
ChatGPT offers plenty of opportunities for accountants to automate certain tasks, but it’s
important to understand its issues and limitations before putting the technology to use.”60 The
potential drawbacks included lack of transparency, limited data availability, cybersecurity risks,
and ethical considerations.
rP
Data Security and Privacy
The transfer and handling of data in relation to AI models presented potential risks in terms of
data privacy and security. There was a concern that the data could be misused, stolen, or
exploited for criminal activities. The rapidly evolving technological landscape required
businesses to proactively understand the benefits and risks associated with these new
technologies. Although ChatGPT and other generative AI vendors claimed to have encrypted
and safeguarded users’ data, government regulations and supervision were still in the process
yo
of catching up with the fast-paced advancements in technology. Consequently, many businesses
faced difficulties in ensuring adequate cybersecurity measures.
In 2021, the European Commission proposed the “EU AI Act” to regulate AI, addressing these
concerns.61 In June 2023, EU lawmakers agreed on a draft version of the act, which would
undergo negotiations with the Council of the European Union and EU member states before
being enacted into law.62 In the US, in July 2023, the Biden administration collaborated with
op
leading AI companies to enhance the draft of the AI Bill of Rights, aimed at managing risks
associated with AI.63
The mainland Chinese government closely monitored the development and content within its
AI industry. On 15 August 2023, the Cyberspace Administration of China and six other state
departments jointly enacted the “Interim Measures for the Management of Generative Artificial
Intelligence Services《生成式人工智能服务管理暂行办法》 ” to regulate generative AI,
tC
Data Accuracy
The user’s experience when asking the same question of the same or different generative AI
models was unpredictable and varied. The results obtained were stochastic, meaning they were
subject to random variation.65 Google Cloud Tech said, “[H]allucinations are words or phrases
that are generated by the model that are often nonsensical or grammatically incorrect.”66 The
No
generative AI models had the potential to provide incorrect or incomplete data, which could
result in costly errors if relied on without proper verification.
Accountants expressed concerns over the potential inaccuracy and biased outcomes of
information generated by generative AI, stemming from the lack of transparency in the AI
60 “ChatGPT and accountancy-know its limits-ICAEW,” ICAEW Insights.
61 “What is the EU AI Act?” The AI Act, https://artificialintelligenceact.eu/, accessed 24 August 2023.
62
Hanna Ziady, “Europe is leading the race to regulate AI. Here’s what you need to know,” CNN Business, 15 June 2023,
https://edition.cnn.com/2023/06/15/tech/ai-act-europe-key-takeaways/index.html, accessed 26 August 2023.
Do
63 “FACT SHEET: Biden-Harris Administration Secures Voluntary Commitments from Leading Artificial Intelligence
Companies to Manage the Risks Posed by AI,” The White House, 21 July 2023, https://www.whitehouse.gov/briefing-
room/statements-releases/2023/07/21/fact-sheet-biden-harris-administration-secures-voluntary-commitments-from-leading-
artificial-intelligence-companies-to-manage-the-risks-posed-by-ai/, accessed 24 August 2023.
64 “China sets AI regulations with "core socialist values,"” CNN Business.
65 Stochastic means “a stochastic process or system is connected with random probability,” Cambridge Dictionary,
10
This document is authorized for educator review use only by Dr.K.Hema divya Kantamaneni, KL University until Mar 2024. Copying or posting is an infringement of copyright.
[email protected] or 617.783.7860
23/774C ChatGPT and Generative AI in Accounting
models. Additionally, challenges arose when the models were not adequately trained on
t
updated accounting and tax data or when the provided data by the user were unclear.
Insufficient context, constraints, and background information also hindered the effective
os
utilization of the AI tools if users did not receive proper training. 67
rP
developers were working on authenticity verification tools to challenge the notion of “seeing is
believing.” 68
yo
property rights.
70 Bryce Elder, “Surrender your desk job to the AI productivity miracle, says Goldman Sachs,” Financial Times, 27 March 2023,
11
This document is authorized for educator review use only by Dr.K.Hema divya Kantamaneni, KL University until Mar 2024. Copying or posting is an infringement of copyright.
[email protected] or 617.783.7860
23/774C ChatGPT and Generative AI in Accounting
regulating in the form of an “insight committee.”75 In May 2023, Musk said during the Wall
t
Street Journal’s CEO Summit, “[T]here is a risk that advanced AI either eliminates or
os
constrains humanity’s growth.”76
Musk also predicted AI’s positive impact, foreseeing an “era of abundance” with low-cost
production of food and various goods and services.77 But he cautioned that risks arise when
technology was utilized for advanced tasks, which could endanger humanity.
rP
The Way Forward
Generative AI offered advantages for businesses and accounting, evolving at its early stages. It
had the potential to transform the way businesses approached financial management and
decision-making. However, no technology could provide complete data privacy or security on
its own. It required management and experienced specialists to choose the right AI tools and
provide training, guidelines, and governance to implement and monitor the use of new
yo
technologies. The goal was to oversee work results, correctly use new technologies, and
safeguard company and clients’ data. While the regulations in AI, biases and inaccuracy, and
social and ethical issues remained to be solved, how would generative AI reshape accounting
in the future?
What protocols should Joesy employ at her CPA firm to adopt and implement the appropriate
generative AI tool equipped with data security? How should Joesy initiate staff members in
adopting new technologies to free them from repetitive work and focus on reviewing the results?
op
tC
No
Do
75 Pras Subramanian, “Elon Musk worries about an advanced AI that ‘eliminates or constrains humanity's growth,’” Yahoo
Finance, 24 May 2023, https://finance.yahoo.com/news/elon-musk-worries-about-an-advanced-ai-that-eliminates-or-
constrains-humanitys-growth-221316566.html#:~:text=Speaking, accessed 7 June 2023.
76 Pras Subramanian, “Elon Musk worries about an advanced AI that ‘eliminates or constrains humanity's growth,’” Yahoo
Finance.
77 Pras Subramanian, “Elon Musk worries about an advanced AI that ‘eliminates or constrains humanity's growth,’” Yahoo
Finance.
12
This document is authorized for educator review use only by Dr.K.Hema divya Kantamaneni, KL University until Mar 2024. Copying or posting is an infringement of copyright.
[email protected] or 617.783.7860
23/774C ChatGPT and Generative AI in Accounting
t
os
In August 2023, these generative AI models were available in the global market for business
and also for accounting use. Some were still in the testing phase and had limitations for usage.
This list is not inclusive, as new products and versions were released continuously.
rP
Microsoft Azure OpenAI Service
In March 2023, Microsoft announced ChatGPT was available through Microsoft Azure OpenAI
Service with added governance and compliance features.78 Microsoft Azure OpenAI Service
was only available for business clients that were already “Microsoft managed customers and
partners,” or MS Azure Cloud service users.79
yo
Microsoft’s business services.80 It also announced plans to sell a stand-alone subscription for
USD5 per user per month in the future.81 Furthermore, Microsoft mentioned that Bing Chat
Enterprise would be accessible through Microsoft 365 Copilot at a later date.
LLaMa
In February 2023, Meta AI partnered with Microsoft to develop LLaMa (Large Language
Model Meta AI), a chatbot system limited to selected business users when it was launched.82
op
On 19 July 2023, Meta AI launched the LLaMa 2 that was free for research and commercial
use upon request.83
ChatGPT Enterprise
When ChatGPT Enterprise was introduced at the end of August 2023, OpenAI emphasized that
ChatGPT Enterprise was not allowed to “train on your business data or conversations, and our
models don’t learn from your usage, so your proprietary and confidential information is safe.”84
tC
ChatGPT Enterprise also claimed it had “enterprise-grade security and privacy, unlimited
higher-speed GPT-4 access, longer context windows for processing longer inputs, advanced
data analysis capabilities, customization options, and much more.”85
No
78 Kyle Wiggers, “ChatGPT comes to Microsoft’s enterprise-focused, Azure-powered managed service,” Techcrunch, 10 March
2023, https://techcrunch.com/2023/03/09/chatgpt-comes-to-microsofts-enterprise-focused-azure-powered-managed-service/,
accessed 20 August 2023.
79 Kyle Wiggers, “ChatGPT comes to Microsoft’s enterprise-focused, Azure-powered managed service,” Techcrunch.
80 Jared Spataro, “Introducing Microsoft 365 Copilot – your copilot for work,” Microsoft Blogs.
81 Jared Spataro, “Introducing Microsoft 365 Copilot – your copilot for work,” Microsoft Blogs.
Do
82 “Introducing LLaMA: A foundational, 65-billion-parameter large language model,” Meta AI, 24 February 2023,
13
This document is authorized for educator review use only by Dr.K.Hema divya Kantamaneni, KL University until Mar 2024. Copying or posting is an infringement of copyright.
[email protected] or 617.783.7860
23/774C ChatGPT and Generative AI in Accounting
t
(CONT’D)
os
Google AI Cloud Platform
Google added generative AI features to its services, such as Google Cloud AI Platform, which
used AI to help businesses automate tasks and make predictions.86 It had also made investments
in Anthropic, a rival AI startup to OpenAI.87
rP
Salesforce Einstein
Salesforce added generative AI features to its AI application, Salesforce Einstein, to help
businesses automate tasks and make predictions in areas such as sales, marketing, and data.88
yo
Localize Data
Companies could also employ “differential privacy” for “federated learning” IT technologies
to localize data and to provide data privacy and anonymization in their AI systems.91
July 2023.
90 Friea Berg, “Snorkel AI x Hugging Face: unlock foundation models for enterprises,” Snorkel AI, 6 April 2023,
Do
14
This document is authorized for educator review use only by Dr.K.Hema divya Kantamaneni, KL University until Mar 2024. Copying or posting is an infringement of copyright.
[email protected] or 617.783.7860
23/774C ChatGPT and Generative AI in Accounting
t
(CONT’D)
os
Fraud Detection
AI platforms such as Ayasdi used unsupervised machine learning to detect irregular patterns in
financial transactions, helping to identify potential fraud.95 By learning how to generate “normal”
patterns, the AI could flag anomalies that could indicate fraudulent activity.
rP
Invoice Processing
Generative AI could help to automate invoice processing. Tools such as Rossum used AI to
understand and generate information from invoices, thus reducing the need for manual data
entry.96
Source: Prepared by coauthor of the case.
yo
op
tC
No
Do
15
This document is authorized for educator review use only by Dr.K.Hema divya Kantamaneni, KL University until Mar 2024. Copying or posting is an infringement of copyright.
[email protected] or 617.783.7860
23/774C ChatGPT and Generative AI in Accounting
t
os
AI could assist accountants in different tasks, such as reconciling receipts, generating and
sending invoices, and auditing accounting transactions [see Exhibit 3]. 97 According to a
research paper quoted by the CMA Exam Academy in the US, “[T]he application of artificial
intelligence could solve the pain points of inefficiency and low added value in the accounting
field, making the accountants turn to more creative work and bring greater value to the
company.”98
rP
Since 2016, the big-four CPA firms had been developing and using in-house-developed AI
software to improve efficiency, and to provide services to their clients, targeting large
companies. 99 In the same year, Deloitte announced it would introduce AI in accounting,
auditing, and taxation.
In 2020, the use of AI in office work was still in its early stage, mainly relying on basic AI
yo
applications, such as robotic process automation (RPA) and optical character recognition (OCR)
technology.100 These AI applications, such as UiPath, Automation Anywhere, and Blue Dot,
could handle repetitive tasks and save labor hours by interfacing with a company’s ERP and
accounting software. But they also required capital investment, implementation, and
customization, which limited their usage to large companies with high volumes of transactions.
After implementation, users could find it challenging to customize the AI applications without
training or IT assistance. In addition, advanced AI applications required higher investment and
training compared to RPA for office task.
op
In early 2023, popular accounting software such as SAP, Oracle, Freshbooks, and QuickBooks
had already integrated AI elements.101 SAP, a leading accounting and ERP software for large
companies, offered “SAP AI” with built-in AI for core business processes. It claimed to
enhance accounting activities such as payment, expense management, period end financial
close, invoice management, auditing, forecasting, and budgeting, improving overall
tC
performance.102
Source: Prepared by coauthor of the case.
No
Do
97 Nathan Liao, “Artificial Intelligence in Accounting: What Will Happen to Accounting Jobs?” CMA Exam Academy.
98 Nathan Liao, “Artificial Intelligence in Accounting: What Will Happen to Accounting Jobs?” CMA Exam Academy.
99 Nathan Liao, “Artificial Intelligence in Accounting: What Will Happen to Accounting Jobs?” CMA Exam Academy.
100 Xu Li and Ramée Liu, “Artificial Intelligence in Accounting,” ACRC, The University of Hong Kong Business School, 2020.
101 Nathan Liao, “Artificial Intelligence in Accounting: What Will Happen to Accounting Jobs?” CMA Exam Academy.
102 “Finance with SAP AI,” SAP, https://www.sap.com/products/artificial-intelligence/finance.html#accounting, accessed 8 July
2023.
16
This document is authorized for educator review use only by Dr.K.Hema divya Kantamaneni, KL University until Mar 2024. Copying or posting is an infringement of copyright.
[email protected] or 617.783.7860
23/774C ChatGPT and Generative AI in Accounting
t
os
“Benefits of AI for Accounting
The most common use for AI in accounting is to take care of repetitive tasks. For example, AI
systems can capably, and with minimal error, complete tasks that include the following:
Inputting and matching data
rP
Receipt reconciliation
Creating and sending invoices
Expense reports
Tracking price changes
yo
Account reconciliation
Sorting transactions
Data recording and reporting
AI is built on algorithms, which improve over time as they are fed more data. In addition to
continuous improvement, AI isn’t susceptible to human error and has around-the-clock capacity
to work.
op
Three of the primary benefits that experts see AI will provide to accountants are:
1. Invisible accounting: A behind-the-scenes function that frees up human accountants to
engage more in strategic decision-making
2. Continuous auditing: Auditing without the need for breaks and with no gaps in reason
tC
or precision
3. Active insight: Accounting managers and leaders can get real-time visibility into their
finances.”
Source: Nathan Liao, “Artificial Intelligence in Accounting: What Will Happen to Accounting
Jobs?” CMA Exam Academy, 2 February 2023, https://cmaexamacademy.com/artificial-
intelligence-in-accounting/, accessed 1 August 2023.
No
Do
17
This document is authorized for educator review use only by Dr.K.Hema divya Kantamaneni, KL University until Mar 2024. Copying or posting is an infringement of copyright.
[email protected] or 617.783.7860
23/774C ChatGPT and Generative AI in Accounting
t
“ChatGPT is a language model that can generate natural language text based on input prompts.
os
While it can be useful in many applications, it may not be the best tool to automate the financial
reporting process on its own. However, it can be used in conjunction with other tools and
technologies to automate certain aspects of financial reporting.
Here are some potential steps to use ChatGPT to automate the financial reporting process:
rP
1. Identify the Reporting Requirements: The first step is to identify the reporting requirements
for the financial report. This may involve reviewing relevant regulations, standards, and
guidelines to determine what information needs to be included in the report.
2. Gather the Financial Data: Once the reporting requirements have been identified, the next
step is to gather the financial data needed to generate the report. This may involve extracting
data from financial statements, accounting databases, or other sources.
3. Preprocess the Data: Before using ChatGPT to generate the report, the data may need to be
yo
preprocessed to ensure that it is in a format that can be easily analyzed by the model. This
may involve cleaning the data, removing duplicates, or applying other data cleaning
techniques.
4. Train the Model: Once the data has been preprocessed, the next step is to train the ChatGPT
model on the financial data. This may involve using supervised or unsupervised learning
techniques to teach the model how to generate the report.
5. Generate the Report: Once the model has been trained, it can be used to generate the
op
financial report automatically. This may involve inputting the financial data into the model
and allowing it to generate the report based on the reporting requirements.
6. Review and Edit the Report: While ChatGPT can generate the report automatically, it is
still important to review and edit the report to ensure that it is accurate and meets the
reporting requirements. This may involve human review and editing to correct errors or
ensure that the report is presented in a clear and understandable format.”
tC
18
This document is authorized for educator review use only by Dr.K.Hema divya Kantamaneni, KL University until Mar 2024. Copying or posting is an infringement of copyright.
[email protected] or 617.783.7860
23/774C ChatGPT and Generative AI in Accounting
t
“Here are some examples of how ChatGPT can be used in the financial reporting process:
os
1. Income Statement Generation: ChatGPT can be used to automatically generate income
statements based on financial data. For example, you could input the revenue and expense
data for a given period into ChatGPT, and it could generate an income statement with the
relevant subtotals and net income figure.
rP
2. Balance Sheet Generation: ChatGPT can also be used to generate balance sheets
automatically. For example, you could input the asset and liability information for a given
period, and ChatGPT could generate a balance sheet with the relevant subtotals and total
equity figure.
3. Financial Analysis: ChatGPT can be used to analyze financial data and provide insights
into trends and patterns. For example, you could input financial data for multiple periods,
and ChatGPT could generate a report that highlights changes in revenue, expenses, and
other key financial metrics over time.
yo
4. Budgeting: ChatGPT can be used to assist with budgeting by generating projections and
forecasts based on historical financial data. For example, you could input revenue and
expense data for the past several periods, and ChatGPT could generate a forecast for the
upcoming period based on trends and patterns in the data.
5. Financial Reporting Compliance: ChatGPT can be used to help ensure compliance with
financial reporting regulations and standards. For example, you could input the financial
op
data for a given period, and ChatGPT could generate a report that meets the relevant
reporting requirements for your industry or jurisdiction.
It's worth noting that while ChatGPT can be a useful tool in the financial reporting process, it's
important to review and verify the generated reports to ensure accuracy and compliance with
relevant regulations and standards. Additionally, it's important to ensure that the data used by
ChatGPT is accurate and complete to avoid potential errors or biases in the generated content.”
tC
19
This document is authorized for educator review use only by Dr.K.Hema divya Kantamaneni, KL University until Mar 2024. Copying or posting is an infringement of copyright.
[email protected] or 617.783.7860