AI’s limitations became pretty apparent after yet another delightful and “robust,” shall we say, conversation with an old friend, an economist, about intriguing ideas on how to universally effectuate, accumulate and port lifetime income guarantees from derived from employment. That conversation got granular, as it often does, with voices raising over details of certain innovative infrastructure tweaks, in progress with some clients, which may serve to make these ideas more usable and welcoming to employers and employees.

AI’s value is no doubt apparent through these sorts of conversations, as it really is a very useful tool we use in sorting out all manner of chaff. Yet my small corner of the world entails a unique weaving together of seemingly unrelated details between what conventional wisdom considers completely different structures to make the sorts of policy initiatives under discussion actually work. This is where AI has a more limitied value.

The need for these conversations isn’t going away, either, because of the value of the policy initiatives underlying them. As noted in Brookings Retirement Security Project’s important 2019 paper by David John, William Gale and Mark Iwry“ From Saving to Spending: A proposal to convert retirement account balances into automatic and flexible income” summarized that

In our view, as discussed above, the right goal should be for retirement savings plans to offer automatic mechanisms that would make it easy for participants to convert saving balances into income. Properly structured and regulated, automatic retirement income structures could help new retirees in much the same way that automatic enrollment and escalation help savers.”

Getting to this “properly structured and regulated” system of converting savings to income does require a bit of Zigging Down the Zag (Dr. Suess), especially when we are trying to do it in a system designed to simply accumulate savings. It is a challenge faced in asset accumulation systems worldwide, and has had its fits and starts here. We in the U.S. , however, actually have an advantage of an extensive “kit” in the DC system from which to work. Deconstructing things like CITs, target date funds and annuities and reconfiguring them-within regulatory bonds, of course- to try to effectively pull off the policy initiatives is a fascinating (and workable) task, which will serve us well going forward.

These sorts of efforts are well beyond the ken of existing AI systems. And in the future? Perhaps, but it may require integration with the ability to sit down in a pub and draw on a napkin with us…. (see the beer napkin annuity).