Need information? An AI chatbot can provide it in seconds. From summarising complex topics to identifying sources, generative AI is transforming how professionals access information.

But faster is not always better. AI tools can miss context, misinterpret information and confidently present inaccurate results or hallucinations. Librarians offer something different: the ability to assess credible sources, understand nuance and apply professional judgement to deliver information that is comprehensive, current and correct.

Carolyn White, CA Library manager, says AI is a useful research starting point but it shouldn't be the destination.

“AI output should always be checked against an authoritative source before being relied upon, especially when being used to advise clients or to make decisions,” she says. “This is where human expertise is invaluable to ensure research is risk free.”

White says the legal profession is leading the way when it comes to AI-generated hallucinations and inaccuracies – to the extent there is a growing database of examples.

“We have also seen hallucinated and inaccurate outputs in finance and accounting, which is why the CA Library is intent on raising awareness of the potential risks involved and supporting our members that use AI for research.”

Why human research still matters

The CA Library offers a range of services, including AI output verification and augmentation, brief queries and in-depth research.

While generative AI produces outputs using a prediction model based on the content in its large language model (LLM), White says the CA Library approach is vastly different.

“Research librarians have excellent knowledge of authoritative authors, publishers and sources. We use this knowledge to craft and execute a comprehensive research plan, offering sources that are inaccessible to AI, including texts, newspaper and journal articles, CA ANZ-published content and other quality material behind paywalls.”

She says unlike AI outputs, the sources are clearly referenced, right down to pinpointing a page or paragraph. “This approach makes verification and trust easy. And, if we cannot find an answer, we don’t make one up!”

White says research skills are becoming essential in the AI-enabled workforce.

“Critical thinking about the use of AI for research and AI output is essential. Asking what the limits of the AI tool are, whether your prompt could be better, what knowledge gaps exist in the LLM and verifying information is critical.”

When AI has created the wrong kind of headlines in professional services, White adds, “it is almost always the case that the AI output has not been verified.”

“Members should always ask, ‘Does the reference or source exist? Does the AI output reflect the facts in the source?’” White says. “This is where research librarians can help. We will verify references in AI output and undertake additional research using licensed, professional resources.”

Finding answers AI couldn’t

Kathryn Tarr FCA is an income tax professional working as a sole trader with a small group of clients. She is also heavily engaged in learning and education, working with several professional bodies and their learning programs, including CA ANZ.

Tarr used the library research service most recently to obtain resources on the operation of Section 100A of the Income Tax Assessment Act 1936 and related analysis of TR 2022/4 and PCG 2022/2.

“I was blown away by what was provided,” she says. “Within a day, I had an extensive list of over 30 relevant, referenced resources, broken down into four distinct categories for easy access.”

Tarr uses AI extensively in the learning and education services she provides. However, she says that while AI could have generated much of the content, it wouldn’t have been able to access the paid resources or present the information in such a useful way.

“I’m all for the use of AI but in this case, for the time I had to spend, the content produced by the librarian was better and more efficient than what I would have received from an AI request,” Tarr says.

Luke Boswarva CA, founder and director of a boutique risk management and operational resilience consultancy, has likewise used the research service for enquiries ranging from the application of international standards to strategies to grow his practice and niche topics not well covered in open-source resources.

“I also use Copilot Researcher but the responses are often very lengthy and I am not always confident they are grounded in reliable sources,” he says.

Boswarva believes the CA Library research service is more targeted and aligns more closely with the original research request.

“I also value that it directs me to the relevant sections of source material, rather than providing an AI-generated summary of multiple documents. The output is richer, more transparent.

“The library also provides an upfront assessment of the relevance of its research outputs and considers queries from multiple perspectives. It’s a fantastic and responsive service.”

CA Library research service

Need help finding information? Take advantage of the complimentary research service and simply get in touch with the library team with your query. This enquiry page lists some examples of recent research topics. Learn more here.

CA ANZ has compiled a webpage – ‘Using AI for Research’ – with suggestions and checklists. It’s prudent to read this before doing AI research. Learn more here.