Thursday thoughts: Research
The National Association of Tax Professionals (NATP) announced last week it is replacing its tax research member benefit with access to Ask Blue J, an AI tax research program.
NATP explains that usage of the research service has steadily declined, roughly 80 percent over the past few years, prompting the shutdown at some point in the near future. The Association also claims that internal research indicates members want to do their own research without waiting for a response.
A team vetted Ask Blue J, using questions actually submitted to the human research service, compared the results, and were satisfied with the AI-generated responses. (I’ve seen what NATP staff allows to pass as reasonable answers to questions in its Facebook group though, so this statement doesn’t boost my confidence that much.)
In its announcement of the change, NATP says “We believe Ask Blue J will enhance your research capabilities and support your needs in today’s fast-paced tax environment.” Besides sounding like AI wrote the line, it presents an amateur marketer’s view of this change, and not that of an experienced practitioner.
I understand the marketer’s appeal of a phrase like “fast-paced tax environment,” but it sends the entirely wrong message to practitioners. As Tom Gorczynski wrote in his reaction to the announcement, “Tax research cannot be done correctly and completely with a simple prompt in a software program. Deciding the correct position for a tax matter takes judgment and discernment, comparing the law and authority to the facts.” Judgment and discernment require contemplation, expertise, and experience. Those are not “fast-paced.”
There are two broader considerations here too that I’m not sure factored into NATP’s decision or, more broadly, how tax research programs like Ask Blue J are being built:
While many basic tax questions can be answered with a definitive response, the potentially high-value questions result is a range of positions that may be taken along a spectrum of authority. Not all reasonable positions have a clear basis in the text of the IRC or Regulations. Can an AI model trained solely on written tax authority discern between substantial authority and a reasonable position, perhaps one requiring disclosure?
Many times a tax question indicates a broader financial question. A common and basic example is the decision to elect S corporation treatment for a business. Tax law gives us a clear set of criteria and conditions for whether the business can make the election, but it offers no guidance on whether the owner(s) should make that election. (We’ve all dealt with clients who made valid S elections despite it being a terrible decision for non-technical reasons.)
I’m obviously still forming my thoughts around this decision. My instinct tells me it won’t play out well, if for no other reason than it replaces humans vetted, hired, and paid by the Association with a black box program sold by a third party.
Anyway, if you haven’t taken the Compass Tax Research Process masterclass, you absolutely should. Just as electric tools do not replace the necessary expertise of a trained, experienced carpenter, neither does an AI tax research program replace the need for tax professionals to understand how to perform their own research.
And if, like me, you prefer to collaborate with a community of tax professionals committed to accurate discussion and respectful dialogue rather than help train an AI program, join InCite.
And now for your thoughts…
Have you used Ask Blue J or another tax-specific AI tool? What do you think of NATP’s decision to replace its human research service with an AI tool?