We all agree. AI is changing many, many business practices forever. The AI revolution, like the first .com wave, is fueling far-reaching business transformation — but also a fair amount of misguided and misleading hype.
In itself, this hype is pretty harmless; no more than window dressing by marketing departments. But it can also lead employers to miss the real practical AI opportunities for ben admin or, far worse, lead them into costly pitfalls and even unsafe misapplications of this valuable new technology.
Ben admin regulatory compliance demands total accuracy and a lot more rigor than a gen AI created blog post. This one isn’t BTW!
_________________________
BENEFITSCAPE was founded to apply the ‘magic’ of ever advancing regtech to HCM-based ben admin and regulatory compliance. Regtech is closely related to, and in many respects a subset of, the fintech revolution in banking and financial services: the use of advanced tech to manage complex procedures, analyze complex databases, and action business outcomes beyond the capacity of human intelligence.
Even before the AI explosion, much of the most advanced regtech and fintech could be categorized as intelligent. It identified new, emerging patterns and learnt from those patterns to modify its autonomous actions. Now more generalist AI is being incorporated into aspects of specialist regtech.
Below is PART 1 of a high-level framework to help employers understand better how this emerging AI landscape can benefit ben admin.
_________________________
First things first: AI RMF
AI is itself the subject of a rapidly emerging regulation and legal challenges — from copyright law through data privacy issues to ethical practices and risk liability. Ben admin, including health benefits, handles the highly sensitive personal data of employees. All an employer’s use of AI should take into account AI’s own governance, risk, and compliance — and this applies to employee data every bit as much as customer data privacy.
The US Department of Commerce’s National Institute of Standards and Technology [NIST] has issued a AI Risk Management Framework [AI RMF]. This AI RMF is intended for voluntary use and to improve the incorporation of trustworthiness considerations into the design, development, use, and evaluation of AI products, services, and systems. If outsourcing ben admin services to an external vendor using AI to support its operations, you should make sure that like BENEFITSCAPE they are fully conversant and compliant with the AI RMF.
BENEFITSCAPE also provides employers with a service reviewing the NIST compliance of their AI use.
_________________________
Regtech, AI, and the IRS
The second important point to make is that AI can not only benefits ben admin compliance — it has also raised the compliance bar for employers and posed new challenges. The IRS makes use of ever more powerful in-house regtech and army of AI-savvy data scientists to interrogate submissions. In short, the IRS has a hugely transformed capacity to identify suspect noise, inaccuracies and anomalies in tax returns. Complexity is no longer a smokescreen.
You can read about the IRS regtech revolution here: Regtech Transforming IRS Powers.
_________________________
AI Has Many Guises
In many ways just like human intelligence, AI has its own version of neurodiversity! Different AI engines, different versions of those engines, different training regimes for those engines, and differently configured deployments all process data and tasks and deliverables in very different ways. But all tend to get bucketed together into the catch-all AI moniker.
For many innovations Mark Zuckerberg’s dictum ‘move fast and break things’ works just fine. Elon Musk has borrowed a similar mantra: ‘Take risks. Learn by blowing things up. Revise. Repeat.’ This approach works great for many projects. Even for very well-funded unmanned rocket research. Maybe not so well when passengers are on-board a driverless car. AI is similar. Fast and furious experimentation is fine for some applications. Not for others where greater risk is involved and zero-error tolerance required.
In short, is your choice of AI engine, its training, configuration, deployment and on-going oversight fit for purpose?
_________________________
Part 1: Conclusion
In many areas, consideration of these AI variations is not high risk.
AI hallucinations [also known as errors!], for example, and other AI failures or shortfalls don’t do much harm. AI served up a good enough answer to my customer’s question. AI generated a convincing enough fake picture, post, video, voice to use — if you ignore the 6 fingers [see ‘Kate Middleton family photo recalled,’ New York Post, March 19, 2024].
But in the area of ben admin, regulatory compliance, and tax reporting, these considerations of AI engine choice, training, and deployment along with AI RMF guidelines are mission critical. Good enough in a compliance context demands 100% accuracy and fit for purpose. Not ‘move fast and break things.’
_________________________
In Part 2 of ‘AI & Regtech: A Framework for Employers,’ BENEFITSCAPE will dig deeper into the different types of AI such as Generative vs Discriminative AI or Weak vs Strong — and, more specifically, which to use and which not for different tasks within ben admin and compliance. And how to train, deploy, and oversee these uses for maximum benefit and zero risk.
If you have any questions concerning your company’s use of AI for any aspect of ben admin and/or its compliance with the NIST AI RMF, please do not hesitate to contact BENEFITSCAPE. We are best-in-class specialists and pioneers in the successful application of advanced new technologies to benefits administration and compliance.
