AI, the wisdom of the crowd and DC default design  

With the launch of generative AI models such as ChatGPT, businesses and governments around the world have clamoured to understand their implications and how not to be left behind. FOMO has been in full effect.

What got me thinking was the idea that the output from these generative AI models is effectively an extraction of the patterns it has learned from the masses of pre-existing information and opinions available on a given topic. There are no doubt nuances to the different types of generative AI models that exist but, in many respects, they are the epitome of the “wisdom of the crowd” principle.

The argument that this will ultimately (if it hasn’t already) reset the base level across the knowledge economy seems very plausible.

Burrowing further down this rabbit hole got me thinking about investments and more specifically the design of default strategies in the defined contribution (DC) pensions sphere. 

Passive (or index tracking) investment is in many respects analogous to this wisdom of crowds principle with the weighted market cap index reflecting the securities that prove most popular with “the crowd” (at least in the near term). Benjamin Graham’s reference to the stock market being a voting machine in the short-run and a weighing machine in the long-run certainly rings true in this context.

There’s also evidence of some of these forces at play in different DC markets around the world. For example, we can see commercial, competitive or regulatory factors having an increasing influence on the evolution of how the typical DC default is designed in a given region.

With the increasing scale and importance of DC pension systems around the world, there’s a natural desire for governments and regulators to focus on ensuring appropriate checks and balances are in place to safeguard members’ interests. Increasing “standardisation” for how DC plans are designed and monitored might provide some level of comfort for regulators but this can't be at the expense of members' long term interests. 

One example is the introduction of an annual performance test by Australian regulators to gauge if a given superannuation fund’s returns are in line with reasonable expectations over a set time period (c. 6 - 9 years). Providers whose funds fail the performance test risk being forced to shut up shop completely, with members’ savings being moved to an alternative default option that has “better returns” (at least over the period being considered against a set benchmark portfolio). The premise of having a system in place to weed out persistent under-performers has a lot of merit but the potential unintended consequences (e.g. herding mentality in default design) should give UK authorities pause for thought given a similar performance test is being mooted for this market

In a world where the output from generative AI models becomes the norm or base output, a higher value or premium will likely be attached to well-considered views that differ from the “mass average”. It could be argued this has always been the case but it certainly feels like this is accelerating for the knowledge economy. If this proves to be the case, it seems the skill we will increasingly need to hone is being able to ask more pertinent and nuanced questions (or 'prompts' in AI parlance!).

Perhaps we will see something similar unfold with the design of DC default strategies. Driven by a variety of forces, we’ll continue to see the emergence of what the masses in a given region deem to be an appropriately designed default. The challenge for those with responsibility for the ongoing design of DC default strategies will be to ask more probing questions about what’s likely to deliver the best outcome for their members in retirement and thereby potentially look beyond what the crowd, in its wisdom, is doing.