Author: Anna Maria Schuller, CUSAP Secretary
In recent years, Artificial Intelligence has been silently making history and breaking records, to the point where we have grown desensitised to its incredible capacity. It is thus not surprising that when three AI LLMs placed in the top five at the Metaculus Cup, a competition for forecasters, with one even snagging the top spot, no one really took notice.
Forecasts are a tricky thing, difficult to assemble, highly dynamic, and reliant on vast amounts of data. Perhaps more importantly, they are incredibly lucrative. Corporations shell out ridiculous sums of money every day to glimpse into the proverbial crystal ball. But the crux of the forecast is the same as that of most AI-generated work; It relies exclusively on previously gathered information and draws sweeping assumptions. Whether stated explicitly or not, some degree of “business as usual” must be assumed to generate predictions of the future.
Statistically, this is a safe-ish bet, likely to be somewhat correct, most of the time. But systems are at their most vulnerable when events do not follow predictions based on historical patterns. In the age of disruption, relying on forecasts alone is to risk them failing when we need them the most.
Military strategists were among the first to understand this. They realised that they needed to consider many possible futures, including unlikely but potentially consequential events, to prepare their forces for unexpected attacks. One of the most influential pioneers of this method wasHerman Kahn, whose book “On Thermonuclear War” (1960) helped establish scenario-based thinking as a systematic approach to anticipating radically different futures.
In the early 1970s, Pierre Wack, then an analyst at Shell, recognised the potential of scenario planning in the field of corporate strategy. In the lead-up to the 1973 oil crisis, Mr. Wack was searching for ways to prepare for the unforeseeable, recognizing that the global economy was on a trajectory that was becoming more and more unpredictable every year.
At its core, scenario planning identifies certain pre-determined elements as well as critical uncertainties in the status quo, isolates trends and drivers that are likely to have a significant impact on future outcomes, and then builds theoretically plausible scenarios around these findings. The scenarios are intended to span a great range of possibility, and largely neglect probabilistic weightings of events, in order to challenge the participants to engage with a greater field of risk and opportunity. It is a new way of thinking: rather than committing to forecasts, the strategist is able to test their plans against all of these scenarios to prepare for uncertainty. Scholarly interest in the methodology has transformed it into a tool for explorative studies in the fields of climate change mitigation, public health and epidemiology, political science, and many others. While many institutes, think tanks, and corporations have developed their own detailed methods, they share a general workflow:
Initially, the surrounding information landscape must be mapped, to provide context for the subsequent analysis. This scoping is often guided by asking participants to identify significant “drivers of change” in the ecosystem, challenging them to think multidimensionally. Instead of focusing on the plausibility of each of these drivers, they are evaluated based on their potential impact and urgency. This method focuses the discussion on the factors with the potential to radically reshape the future, without dismissing seemingly low-impact, low-urgency drivers .
The next stage involves the construction of multiple logical chains of geopolitical, economic, and societal events that lead to specific, distinct versions of the future – the different “scenarios”. Methods vary in how this stage is approached: Some choose to work backwards from a certain outcome, while others choose to develop scenarios from a set of pre-defined drivers. Most commonly, three to six significantly different scenarios are developed to achieve sufficient variability to challenge the strategist, without creating convergent or redundant scenarios.
As alluded to previously, the strength of the scenario-based approach lies not in the information it synthesizes, but in its ability to help organizations and clients adapt their thinking and consider many trajectories simultaneously. To achieve this, the scenarios need to be crafted into compelling and communicable narratives. This includes the careful design of a title, storyline, and selective detailing to make the scenario tangible, while ensuring that the emerging narrative is consistent with the audience’s understanding of the world.
As an association focused on protecting and scrutinizing the scientific method, we must consider the challenges of this methodology when utilized for scientific publications. Firstly, as the method deliberately deprioritises probability and quantification to enable creativity, the process is largely conversational and interpretive. Scenarios are developed in dialogue with experts, clients, and other stakeholders, who are likely to favor considering scenarios they deem plausible or familiar, and cannot be expected to differentiate critical uncertainties and pre-determined elements without expert guidance. This places a remarkable amount of responsibility on the skill and ability of the administrator to direct discussion and remain in line with the underlying principles of the method.
It is also apparent that considering many futures requires more effort than analyzing a single prediction. The more detailed the analysis, the more potent it may be, however, beyond a certain point, additional analytical depth may become counterproductive. Excessive detail may slow or paralyze the process, ultimately resulting in a collection of many attempted forecasts that do not carry sufficient narrative weight.
Finally, the most dangerous challenge lies in the communication of the resultant work. If the storytelling behind the scenario does not reach the strategic decision makers in a way that encourages them to expand their own foresight, the whole method falls flat. Scenarios are inherently not actionable: they are not trajectories to be followed or predictions to be acted upon. Rather, they represent potential futures against which strategies can be tested, but they must be crafted in a way that enables this.
Scenario planning has had a good run. It has established itself as common practice in military strategy, shepherded Shell through multiple financial crises and turbulence in the oil and gas industry, became the standard in many C-suite strategy meetings, and finally found its way onto the desks and into the publications of numerous scholars.
But with AI outperforming human predictions, is it possible that the sun is setting on scenario planning? Is AI likely to reduce the uncertainty in the world to an extent that makes scenario planning redundant? Philosophically, I am not equipped to answer this question. Intuitively, I want to conclude that there is too much future for anyone (or anything) to predict with sufficient precision to make it universally useful. Scenario planning, in my view, does not replace forecasting, but clearly puts its limitations into perspective. If you are prepared for even the improbable, you can weather almost any disruption.
In scholarship, scenario planning is an exercise, a mental experiment, essentially a conversation starter. I have had the great pleasure of sitting in on a scenario planning workshop focused on geopolitics in the context of climate change and industrialization. While initially skeptical of the narrative nature of the method, I can now confirm that not only do the scenarios allow scholars to think more creatively – they instil a sense of optimism that I believe is much needed to face the oncoming uncertainty.
