How to make ChatGPT work for you

Artificial intelligence is improving, and thought is turning to its potential applications.

ChatGPT is essentially a turbo-charged chatbot that can provide answers to questions no matter how big or small. It can write essays, poems and even cover letters for job applications (not that you should) with a confident and fluent writing style. It does this by drawing on the vast amount of human knowledge used in building it, with careful guidance from human experts, who helped train and refine its responses. 

That’s where the significant step forward has been made. The initial language model was fine-tuned by feeding the chatbot an enormous number of questions and answers provided by human AI trainers. Then, the programme was asked to produce several different responses to a wide variety of questions, which human experts ranked from best to worst. This human-guided fine-tuning means ChatGPT is often extremely adept at interpreting questions, finding the information sought and framing the answer in a natural manner. 

“I’ve seen this movie before. If you replace 1995 business web with 2023 AI, many of the same questions are being asked.”

– Gary Turner, Co-Founder and former Managing Director of Xero 

AI and accounting 

As with many professions, accountants have been exploring how AI and chatbots, including ChatGPT, can be used. 

When ChatGPT emerged, one accountant, Stuart Cobbe, decided to put it to the test, running it through an ICAEW first-stage sample paper.  

“You can breathe easy – ChatGPT still fails a first-stage ACA exam,” Cobbe posted on LinkedIn. “And yes, I did spend an hour pasting in questions from an accounting exam paper into ChatGPT.” 

Cobbe entered questions from a sample exam into the chatbot, then extrapolated its answers back into the online test. ChatGPT got 42% on the exam in the end, short of the 55% pass mark set by ICAEW. 

The development of ChatGPT and other similar AI is reminiscent of the early days of the internet, says Gary Turner, Co-Founder and former Managing Director of Xero. 

“Lots of people knew in the mid-1990s that the web was big and important, but people hadn’t worked out the use cases for it yet, other than having a website,” Turner says. “That pattern is recognisable.”

Use cases

Thus far, AI has been primarily used in accounting to spot patterns. Last year, KPMG invested in AI by partnering with Blue J, which “predicts tax scenario outcomes with 90%+ accuracy”. The programme will “dramatically reduce the time spent searching for and analysing tax legislation and case law”, the Big Four firm said. 

For accountants, Turner suggests that, rather than clients interacting directly with super-smart chatbots, it could well be accountants doing so on their behalf. Finance teams within organisations could use chatbots in the same way. 

“Imagine on a phone or a video call, the client is describing their query to you in real time, and you have a chatbot listening to what the client is asking. This is enabling you as the practitioner to advise them appropriately, almost as a co-pilot,” Turner says. 

“It feels like we’re on the cusp of something huge,” he adds. “The things we eventually use chatbots for in 15 to 20 years will be three or four levels beyond what we’re currently considering.”   

Straight from the horse’s keyboard

Here’s what ChatGPT had to say when we asked it about accounting and tax matters.

AAT: Hi ChatGPT. Are you able to give tax advice? 

ChatGPT: “As a language model, I can provide general information and answer questions about tax laws and regulations, but I am not a licensed tax professional and cannot provide specific advice on how to file taxes or handle specific tax-related situations. It’s always a good idea to consult with a licensed tax professional or accountant when it comes to taxes and financial matters.”  

AAT: If AI gave incorrect tax advice, who is responsible? 

ChatGPT: “As AI technology continues to advance, the question of legal responsibility for AI-generated advice and actions is becoming an increasingly important one. In general, the legal responsibility for the consequences of AI-generated advice would depend on the specific circumstances of the case and the laws in question. 

“In general, the legal responsibility for the consequences of AI-generated advice would likely fall on the entity that created or deployed the AI system. If a company creates an AI system that provides tax advice to users, and the advice turns out to be incorrect, the company would likely be held responsible for any negative consequences that result from the use of that advice.” 

AAT: If AI is able to provide tax advice in the future, will there still be a role for humans? 

ChatGPT:  “Even if AI becomes capable of providing accurate and comprehensive tax advice in the future, there will likely still be a role for human professionals in the field of accounting and tax. 

“AI systems can be used to automate repetitive tasks, such as data entry and analysis, but human professionals will still be needed to interpret and apply the results of those tasks. This includes providing context-specific advice and making judgement calls based on the unique circumstances of each case. Humans will also be needed to oversee the development, deployment and maintenance of AI systems to ensure they are operating as intended. 

“Additionally, AI can be a powerful tool to help human professionals to work more efficiently and effectively, but it is not able to replace the human touch and emotional intelligence that is needed in some aspects of the job. 

“It’s also worth noting that laws and regulations around taxes and financial matters are constantly changing, and AI systems would need to be continuously updated to keep up with those changes. Human professionals will be needed to stay current on the latest laws and regulations and ensure the AI is providing accurate and up-to-date advice.” 

AAT Comment offers news and opinion on the world of business and finance from the Association of Accounting Technicians.