In 2017 I posted an article about how the future for actuaries was starting to look, with particular reference to a Society of Actuaries paper by Dodzi Attimu and Bryon Robidoux, which has since been moved to here.

I summarised their paper as follows at the time:

Focusing on…a paper produced by Dodzi Attimu and Bryon Robidoux for the Society of Actuaries in July 2016 explored the theme of robo actuaries, by which they meant software that can perform the role of an actuary. They went on to elaborate as follows:

Though many actuaries would agree certain tasks can and should be automated, we are talking about more than that here. We mean a software system that can more or less autonomously perform the following activities: develop products, set assumptions, build models based on product and general risk specifications, develop and recommend investment and hedging strategies, generate memos to senior management, etc.

They then went on to define a robo actuarial analyst as:

A system that has limited cognitive abilities but can undertake specialized activities, e.g. perform the heavy lifting in model building (once the specification/configuration is created), perform portfolio optimization, generate reports including narratives (e.g. memos) based on data analysis, etc. When it comes to introducing AI to the actuarial profession, we believe the robo actuarial analyst would constitute the first wave and the robo actuary the second wave.

They estimate that the first wave is 5 to 10 years away and the second 15 to 20 years away. We have been warned.

So 9 years on from their paper, how are things looking? Well the robo actuarial analyst wave certainly seems to be pretty much here, particularly now that large language models like ChatGPT are being increasingly used to generate reports. It suddenly looks a lot less fanciful to assume that the full robo actuary is less than 11 years away.

But now the debate on AI appears to be shifting to an argument between whether we are heading for Vernor Vinge’s “Singularity” where the increasingly capable systems

would not be humankind’s “tool” — any more than humans are the tools of rabbits or robins or chimpanzees

on the one hand, and, on the other, the idea that “it is going to take a long time for us to really use AI properly…, because of how hard it is to regear processes and organizations around new tech”.

In his article on Understanding AI as a social technology, Henry Farrell suggests that neither of these positions allow a proper understanding of the impact AI is likely to have, instead proposing the really interesting idea that we are already part way through a “slow singularity”, which began with the industrial revolution. As he puts it:

Under this understanding, great technological changes and great social changes are inseparable from each other. The reason why implementing normal technology is that so slow is that it requires sometimes profound social and economic transformations, and involves enormous political struggle over which kinds of transformation ought happen, which ought not, and to whose benefit.

This chimes with what I was saying recently about AI possibly not being the best place to look for the next industrial revolution. Farrell plausibly describes the current period using the words of Herbert Simon. As Farrell says: “Human beings have quite limited internal ability to process information, and confront an unpredictable and complex world. Hence, they rely on a variety of external arrangements that do much of their information processing for them.” So Simon says of markets, for instance, which:

appear to conserve information and calculation by assigning decisions to actors who can make them on the basis of information that is available to them locally – that is, without knowing much about the rest of the economy apart from the prices and properties of the goods they are purchasing and the costs of the goods they are producing.

And bureaucracies and business organisations, similarly:

like markets, are vast distributed computers whose decision processes are substantially decentralized. … [although none] of the theories of optimality in resource allocation that are provable for ideal competitive markets can be proved for hierarchy, … this does not mean that real organizations operate inefficiently as compared to real markets. … Uncertainty often persuades social systems to use hierarchy rather than markets in making decisions.

Large language models by this analysis are then just another form of complex information processing, “likely to reshape the ways in which human beings construct shared knowledge and act upon it, with their own particular advantages and disadvantages. However, they act on different kinds of knowledge than markets and hierarchies”. As an Economist article Farrell co-wrote with Cosma Shalizi says:

We now have a technology that does for written and pictured culture what largescale markets do for the economy, what large-scale bureaucracy does for society, and perhaps even comparable with what print once did for language. What happens next?

Some suggestions follow and I strongly recommend you read the whole thing. However, if we return to what I and others were saying in 2016 and 2017, it may be that we were asking the wrong question. Perhaps the big changes of behaviour required of us to operate as economic beings have already happened (the start of the “slow singularity” of the industrial revolution) and the removal of alternatives that required us to spend increasing proportions of our time within and interacting with bureacracies and other large organisations were the logical appendage to that process. These processes are merely becoming more advanced rather than changing fundamentally in form.

And the third part, ie language? What started with the emergence of Late Modern English in the 1800s looks like it is now being accelerated via a new way of complex information processing applied to written, pictured (and I would say also heard) culture.

So the future then becomes something not driven by technology, but by our decisions about which processes we want to allow or even encourage and which we don’t, whether those are market processes, organisational processes or large language processes. We don’t have to have robo actuaries or even robo actuarial analysts, but we do have to make some decisions.

And students entering this arena need to prepare themselves to be participants in those decisions rather than just victims of them. A subject I will be returning to.