"There is a lot of potential for AI, but there are also many risks and complexities that are new and evolving," said Orly Mazur, a tax professor who co-authored a forthcoming Tennessee Law Review article on the potential benefits of tax agencies using generative AI to communicate with taxpayers. (Alamy Stock Photo)
"We wanted to start to build a foundation to help everyone understand, in context, what these tools might mean for tax administration," Minnick said in an interview.
An FTA working group is expected to finish drafting the paper this year. Then it will be finalized and released to FTA members — tax agencies in all 50 states plus the District of Columbia, Philadelphia and New York City. The organization will hold panels to discuss elements of the paper at its conferences this year to facilitate discussions about the technology, Minnick said.
The FTA's work on the paper, which began last year, coincides with several tax administrators across the country expressing interest in using AI tools to improve their customer service operations and identify noncompliance. Michigan Gov. Gretchen Whitmer, for example, announced in her February budget proposal that she is seeking to provide the state Department of Treasury with $1.3 million to use AI tools to assist the agency with "data analytics and fraud prevention."
The heads of tax agencies in New Jersey and Louisiana have previously told Law360 that they'd be interested in exploring how to use AI-based chat options for customer service purposes. And New York's acting tax commissioner, Amanda Hiller, said in December that advancements in data analytics can help auditors better scrutinize tax returns from large, complex partnerships.
Tax agencies have long used AI and automated processes to flag issues with returns and crack down on possible fraud. But the novel considerations that come along with using more modern generative AI tools, such as concerns about accuracy and data privacy, can also make tax administrators hesitant to use the programs, according to Orly Mazur, a tax professor at Southern Methodist University's Dedman School of Law.
"There is a lot of potential for AI, but there are also many risks and complexities that are new and evolving," said Mazur, who co-authored a forthcoming Tennessee Law Review article on the potential benefits of tax agencies using generative AI to communicate with taxpayers.
Minnick said he and the FTA's general counsel, Brian Oliner, have had continuous discussions about how generative AI tools can intersect with a state's laws. Using a chatbot as an example, Minnick said states may need to consider how to address a hallucination, or an incorrect answer, that an AI program may provide to a taxpayer.
He said many states have statutes that specify that a tax agency doesn't need to honor certain inaccurate statements that a department employee gives to a taxpayer. However, it may not be clear how those laws would apply to an AI program, he said.
"From a governance perspective, how do you mitigate that risk or how do you handle it?" Minnick asked.
Questions about how new AI technologies can be used appropriately in tax administration are also present at the federal level. The Treasury Inspector General for Tax Administration said in a November report that the Internal Revenue Service has nearly 70 AI tools in use or in development, but that the agency needs to speed up its development of governance procedures to make sure AI is being used safely and securely.
The IRS already has created a group tasked with governing AI projects under the umbrella of its chief data and analytics officer, but it should accelerate the creation and use of oversight structures to ensure accountability and responsible use of AI, TIGTA said. The IRS agreed with this recommendation, according to the report.
For state and local tax purposes, Mazur called the FTA's project "a fantastic idea" to educate tax administrators about AI and encourage discussions about how the tools can be used and implemented with proper safeguards.
"By having just a central framework, I think it could get more states on board in adopting AI efficiently, and maybe give them a push to start looking into this," Mazur said.
Beyond employing AI tools that a taxpayer can interact with, or that help identify audit candidates, tax agencies could use them to streamline certain routine internal operations, such as generating revenue reports, she added.
While some attorneys have expressed worries that AI might lead to a leaner workforce in the private sector, Nicolas Maduros, director of the California Department of Tax and Fee Administration, told Law360 that he views the technology as a tool that can increase efficiency in tax agencies that are struggling with hiring and retention.
"We've got more than enough work to do across the department," Maduros said. "To the extent we can help our people both be more efficient and provide better services, that is the frame through which I view all of this, not as any sort of effort to reduce our headcount."
Maduros is also the president of the FTA's board of trustees but hasn't been directly involved in its white paper project. He spoke to Law360 before Gov. Gavin Newsom announced on Feb. 28 that he appointed Maduros as secretary of California's Government Operations Agency. Tamma Adamek, a spokesperson for the CDTFA, said Maduros is slated to begin making the transition later this month.
Maduros had been at the helm of the CDTFA since 2017 and has engaged in several AI initiatives since then, including a recent pilot project for using a generative AI program to assist employees who field taxpayer inquiries over the phone, he said.
The initiative was conducted in a testing environment that didn't field any phone calls from actual taxpayers, but Maduros said the pilot's results appeared to shorten the length of a standard call by about 20%. The agency is next considering whether to go into the next phase of implementing the technology, he added.
In the pilot, the AI tool listened along on a call and would search the agency's databases and guidance to provide the department employee with suggested responses to the caller's questions, Maduros said. It would also generate a summary of the call afterward that the employee would review for accuracy before putting it into a taxpayer's background in the system, which employees currently do manually, he said. Additionally, he said the tool was also trained to redact sensitive information, such as a Social Security number, from the call's transcription.
Minnick, speaking generally about AI and not specifically about California's pilot, said that while data privacy concerns are legitimate considerations that tax agencies must keep in mind when looking at using AI tools, protecting taxpayer information is already a paramount requirement for any initiative that tax administrators are considering.
"State and local tax agencies, probably more so than other counterpart government agencies, have data security and privacy as kind of a top directive," he said. "Fortunately, there's not a ton that we would recommend to implement in order to ensure data security and privacy when using generative AI because states are already doing that."
The CDTFA also uses data analytics to help automate outreach to taxpayers when certain indicators show they may have a "lower dollar" amount of noncompliance, Maduros said. He said that initiative was an offshoot of an AI project that sought to better identify noncompliance that could have been missed from the agency's traditional algorithms. That project was determined not to be accurate enough for the department to adopt, but he said the agency is committed to continually enhancing its data analytics tools.
"To the extent we can focus our enforcement efforts on individuals and businesses who are actually noncompliant, it's a win-win situation, both for us and the taxpayers," Maduros said, calling identifying noncompliance "the holy grail" of AI for tax administrators.
The FTA's forthcoming white paper should assist tax agencies that are at various stages of studying for adopting AI tools by furthering their understanding of the technology's benefits and risks and what to consider when rolling them out, Maduros said.
"One of the key roles of the FTA is to bring all the states together and share best practices," he said. "I think this is a natural area for us to have that effort because all of the states are navigating this new technology and how to use it."
The FTA plans to routinely update the white paper because a static product would fail to account for advancements in the technology, Minnick said.
"My expectation is that we would continue to revise it over time," he said. "Because if we had written it six months ago, we'd be revising it right now because so much has changed in terms of the application of the productivity of the tools."
--Additional reporting by Jack McLoone and Michael Nunes. Editing by Tim Ruel and Roy LeBlanc.
For a reprint of this article, please contact reprints@law360.com.