We have written before about the question of how often we should expect pandemics — the simple model we applied then was this: is Covid-19 an example of a Spanish flu like event and so likely 1/100 years or should we consider it a sars-virus spill over event, of which we roughly have 1/10 year events (and then saying that perhaps the severity of such spillovers is distributed roughly as we have seen so far so that we get a 1/30 years or 1/40 years frequency of a Covid-sized pandemic.
Now, a couple of scientists have looked at the patterns and evidence and they are suggesting that although the Spanish flu was a 1/400 year event, a Covid-sized pandemic might well be a 1/59 years event. So another 60 years and we may very well have another.
That means that anyone below the age of 20 is likely to experience another pandemic in their lifetimes.
This could, of course, be recency and availability bias – but it does seem likely knowing not only the changing demographics but also the change in spillover patterns. More people in cities, aging populations, animal reservoirs moving closer to people as forests decline — together these factors could easily be modeled to lead to an increasing incidence.
If we also add antibiotic resistance the pool of eligible pathogens expands radically, and so this makes pandemics an issue that might deserve a lot more attention even as we scale back restrictions. It will be interesting to see if we manage to really mobilize efforts here or if we will consider the pandemic just “bad weather and bad luck” and then proceed to promptly forget it.
What kind of data sets can be helpful for predictions? One category of data that often are ignored are the data that you can get by simply asking people – or, as it is sometimes called, “walking about”. In a recent article about predicting unemployment, the authors note that this kind of often qualitative data can be very efficient in forecasting, and that even if you just ask people if they are worried about something that index can be useful to understand:
However, in Blanchflower and Bryson (2021) we indicate that fear of unemployment predicts both – that is, it can predict the real GDP change (post revisions) and unemployment change. We know, therefore, that things are getting bad when the fear numbers turn seriously negative in consecutive months. It is this – the turning point in fear of unemployment – that should really be the focus of our modelling efforts if we are to predict turning points in the business cycle.
The potential for further forecasting research using these kinds of data seems high.
Bryson A & Blancheflower, D “How the economics of walking about helps predict unemployment”VoxEU 24th August 2021
The insight here – that aggregated, deeply subjective views catch deeper trends – seems right to me, and somewhat underutilized. It also seems to indicate that the best indices in complex feedback systems with independent agents may be the sentiments of the agents themselves.
How can we deal with ten year forecasting and planning in an interesting way? The naive approach would be to simply guess what state of the world will be probable in ten years, and then describe that as well as we can. This is akin to a kind of science fiction writing, and can add depth and richness to a view of the future, but also seems a very brittle way to try to plan.
Looking to the stars for insights.
There are other interesting methods here that deserve to be explored.
The first is to look at the current day debates, and ask which ones will likely be resolved in the coming ten years. The trick here is to find the hottest issues of the day and then assess if we think that this will be a hot issue ten years from now – and if it is not, how it has changed.
This builds on the realization that human affairs usually do not revolve around the same hot issues for decades – something decides these debates, and changes the playing field.
A key notion here is the idea of the stability of controversy. What things that are controversial today will still be controversial in ten, hundred or even thousand years? Some issues are 100 year controversial – you could argue that same-sex marriage in the US is an example of this – but most issues are not. The immigration issue in Sweden is one example – 15 years ago this was still very controversial in terms of if there even was a problem, and today there seems to be (right or wrong) a significant consensus on something having to be done, and that we did create a problem with the volumes that we thought manageable.
A person interested in the future back then could perhaps not have predicted that the evolution would lead to a consensus around the issue – but the lack of stability in this controversy (it was rapidly changing) suggests that they should have been able to predict that it would have been resolved.
Applying this to ten year planning today we can ask first which controversies are less than ten year stable, and then the next step is to sketch out how they can resolve. Usually this can be framed as “either X or Y”.
The similarity and analogy with scenario building here becomes obvious. In scenario building we go through a lot of extra steps and think through driving forces and other variables, but what we then aim to do is to identify those driving forces that are unstable and uncertain and ideally will resolve in an “either this or that”-fashion. They then become the axes of our scenarios.
And this is how we can then use the identified controversies that are likely to be resolved: they become the possible variables in a n-dimensional set of scenarios for the future – and we can then look at subset scenarios and pick those that seem most relevant to what we are trying to do.
The second method that I think merits more thought is to consider industry scandals as naturally occuring events. Just like earth quakes of different magnitudes have a 1/10 or 1/100 year probability to occur, such scandals – and not just industry, but political – seem to follow a similar pattern. That means that if we want to plan for the next ten years we should be able to say with some certainty that there will have been major “normal accidents” to use Charles Perrow’s term during that period and as these will have occured in the world as we know it now we can ask what a major technical, economic or political scandal will mean – because one will have happened with near-certainty.
Together these methods are interesting tools to play with for any larger planning exercise that allows it self to dial back to a resolution that shows the world in decades and not quarters.
Another observation about this “planning resolution” is that it puts the focus on other pace layers – like the legal, regulatory and political pace layers – since these have major impact in decades, in a way that is sometimes missed if you have too high resolution in your planning efforts.
Niels Bohr proposed that one fundamental insight of quantum physics was that some phenomena or systems could be described in two or more mutually exclusive ways and that it would be a mistake to pick one description as the “right one” – both could be accurate.
This violates the logical dictum of the excluded middle, in a sense, since it suggests that when we ask if a system is A or B, the answer is “both”. It is a radical view, and just how radical is illustrated by physicist Frank Wilczek’s use of another example: legal liability.
Wilczek suggests that humans can be described as physical systems wholly determined by physical laws, or acting intentional agents with motives and responsibility – and that both descriptions are accurate representations. It would be a mistake, then, to argue that we have no legal responsibility because we can be described as physical systems, because we can equally well be described as acting, intentional agents.
Bohr’s coat of arms — opposites are complementary…
This feels like cheating – and at least my inner high-school philosopher wants to know “but which is it?” — the challenge here is if we accept the lack of a singular ontological ground truth we suddenly seem to be on a slipper slope to Feyerabend-land, where everything is permissable.
This is an interesting problem – it echoes of Dostoyevsky’s observation that if God is dead, then all is permitted — the lack of a single foundational layer of reality, the lack of a hierarchy of reality, seems to unmoor us from truth. But maybe the answer to that is in the “mutually exclusive”? If two descriptions overlap in any way, then they do not count — what we need to have developed a complementarity is to have a description that is excludes the others.
We then end up with a new meta-layer of ontology: all the mutually exclusive descriptions of a phenomenon that we can devise. This is intriguing, because it suggests that the idea of the model has a much more foundational role in describing reality than we may have guessed; rather than just abstract things away, complementarity suggests that the world is made of mutually exclusive models.
One could also imagine a pragmatic version of a complementarity metaphysics: the time spent reducing sufficiently mutually exclusive descriptions to a single one is not well-invested; it is much better to seek ways of using these mutually exclusive descriptions and judge them on their usefulness.
A lot of work has gone into what is sometimes called “technology forecasting” – attempting to understand how semiconductors, artificial intelligence, quantum computing, proteomics etc will evolve over the coming years. Such research is valuable and interesting – and can be interestingly contrasted with something we could call “capability forecasting”.
Capability forecasting is focusing not on the technology as much on what we will be able to do. Now, you may argue that capability forecasting is dependent on technology – and you would be partly right – but the point of forecasting capabilities is to try to approach the problem of forecasting generally from a new angle.
The question that interests us is this: what will we be able to do routinely and well in the coming ten years?
As we start generating capabilities we can figure out what they would require, and doing so allows us to see how technology can be used with organizational innovations — allowing us to forecast technology in context rather than as an isolated roadmap.
Here are a few future capabilities that could be interesting to explore – and think about. When do we think we will be able to do the following ?
Cure the top 10 killer cancers at 80% efficiency
Mine asteroids cost effectively
Generate 80% of our energy without any environmental impact
Educate 80% of the world’s kids
Extend our life spans to an average 120 years in the top 10 life span economies today
Reduce sum total prison time overall by 50% in the US
For some of these the answer maybe “never”, and that is not just a viable answer – it is a very interesting answer! It suggests that that path in the capability tree is closed off to us, and then we can ask why that is — eliminating possible future capabilities is something that allows us to back out scenarios more credibly; whenever a scenario depends on a capability that is ruled out that scenario falls.
If we, for example, say that humanity will never be able to terraform planets to become more hospitable to life, well, then interstellar expansion of humanity will depend on finding already inhabitable planets or some other means (large ships, space stations etc etc). Or to take a less sensational example: if we believe we will never be able to reduce overall time spent in prison, how do we then approach prison design to ensure that it prison time can be used productively, say?
Lists of things we will never be able to do tend to age badly, though, so one way to explore capability forecasting is to survey people on exactly this – what do you think we will absolutely not be able to do in the coming years? And then forecast a world where we do at least 50% of those things…
The idea of an index in economy is simple: find a way to measure a change in an ensemble of values in a single value, and then track that single value over time.
The challenges are many: how do you pick the values in your basket, and do you weight them differently? Do you update them, and if so with what periodicity? And then – of course – how do you interpret the index? What does it mean? And how do you make sure that it is not interpreted in a way that overloads it with meaning?
Indices can have an emotional, almost visceral, effect on us. Days the stock index is up some of us feel more elated and happy – and when it is down we feel less successful and confident, and that is not just us amateurs. Even someone like Warren Buffett admits to buying different breakfasts depending on how bouyant the market makes him feel!
The challenge of building an index is that you need to find good data sets – or produce them – and then refine the index over time. This takes hard, honest econometric work, so when new indices are presented is almost always worthwhile to examine them more clearly.
One such new index recently presented is the North American Container index. The authors state that:
this index can be incorporated into a structural vector autoregressive model of the US economy that includes, in addition, a measure of real personal consumption and US manufacturing output. The model facilitates the identification of shocks to domestic US demand as well as foreign demand for US manufactured goods, while accounting for unexpected frictions in North American container trade associated with shipping delays, port congestion, labour strife, and foreign supply chain bottlenecks. The model shows that, on average, shocks related to frictions in the container shipping market have a nontrivial effect on the US economy. They account for 29% of the variation in US manufacturing output relative to trend and 38% of the variation in detrended real personal consumption.
And here is a chart:
Notes: Based on the number of twenty-foot equivalent (TEU) containers processed in the major North American ports. Seasonally adjusted and log-linearly detrended.
Certainly an index worth thinking about, and one peaking these days!
But more interestingly – what are some indices that you would like to be able to construct? Which data sets would you need? Imagining indices a fine summer day may sound like the nerdiest thing you could imagine, but it is a good way to think through problems.
Looking back, it is fairly easy to see that the race to the moon was a geopolitical competition, an attempt to use a technological task as a proxy for answering the question of which political system was the most robust, innovative and effective. But was we enter an age of new geopolitical races, it seems much less clear what this would look like.
The first question is what the new geopolitical race between the US and China will be a race to. There are plenty of candidates: AI, quantum computing, central bank digital currency…and there is no clear leader. Even if we knew that say AI was the field of competition, it seems unclear what would be the equivalent of landing on the moon.
Producing General Artificial Intelligence is unlikely to be something that is achieved in a single push from a single country or actor, even if we imagine that the effort was organized in a Manhattan-project like way. What would the test be? That a piece of software passes the Turing test?
The Turing test sounds like a conclusive test that could rival the landing on the moon, but it has far too many subjective components to be a really interesting test of technological capability – it is, in a sense, much more a test of shallow intelligence design; the design of an intelligence that can mimic human intelligence in limited experimential contexts. Some would argue that the Turing test already has been passed in numerous cases depending on your contextual requirements, and so using this to determine the outcome of a geopolitical race seems less helpful.
Even in a highly technical area like quantum computing we find no clear test of the much discussed “quantum supremacy” (horrible term). The current holder of this title seems to be China, after a recent experiment with a problem set that was solved successfully in an hour, whereas a traditional computer would have required 8 years.
But here, as well, we are demonstrating a narrow capability that is hardly generalizable into geopolitical power. The point, or one of the points, of the space race was that it was a proxy test for organizational, technical and innovative capabilities of a nation — AI, quantum computing and similar initiatives do not qualify as such. Better technology in any of these fields are incremental geopolitical advantages, not decisive superior strategic positions.
The country that dominated space could literally also use that advantage to control what happened on earth.
Interestingly we lived through a mini-geopolitical race during the pandemic – the race to a vaccine – and here we see something else: when it comes to medicine and to vaccines Russia and China were first, China in June, and Russia in August – but the test was determined not by a documented technological feat, as much as government approval of the new vaccine.
The mRNA-vaccines that the US managed to produced were instead hailed as the greatest medical innovation and accomplishment that we have seen in a long time (and arguably rightly so, although it did not happen in a year, the science had been around and the companies working on mRNA had looked at the vaccine problem for a lot longer) – and so the US seems to have, in the eyes of the world, won that race. It was not just being first that counted, but also the world admitting that you were the first.
The geopolitical nature of the race were clear – and if anyone missed it, Russia called their vaccine Sputnik – but the outcomes were not as obvious. Who won?
So, what does this imply for the future? It could evolve in a number of ways – clear tests for any of the existing technologies could emerge, or the field of competition might shift into areas where such tests exist – such as fighting cancer. The ability to cure the more common forms of cancer would be a feat close to the moon landing. Another area that could be interesting is energy technology – an order of magnitude more efficient solar power technology could be another field that would give a lasting geopolitical advantage to the country able to produce it.
But maybe the geopolitical competition we are now seeing will not converge on the mental model of a “race” where you need to get there first, but instead be organized more like frontier where the winner is the one who advances the entire field of science the most; it is the sum total state of technological capability that counts.
One thing that would support this hypothesis is that the technologies we are looking at here support each-other. Quantum computing could change AI and revolutionize medical science. Instead of a race, what we see looks more like the strengthening of a network of technical capacity – a general technological capability.
Winning that is much more complex than winning a race, and requires a much more thoughtful approach to R&D. The good thing is that it can be a distributed effort between different countries. The idea that R&D strategies and innovation programmes should be national may have served out its usefulness.
One possible scenario would be a NATO for technological innovation and science emerging to balance the Chinese efforts – sharing innovation across geopolitical lines rather than national ones.
Here is a narrative about the Internet that is getting more and more common:
It certainly seems that the Internet is now the realm of a small number of enterprises that dominate this space. This is no longer a diverse, vibrant environment where new entrants compete on equal terms with incumbents, where the pace of innovation and change is relentless, and users benefit from having affordable access to an incredibly rich environment of goods and services that is continually evolving. Instead, today’s Internet appears to be re-living the telco nightmare where a small clique of massive incumbent operators imposes overarching control on the entire service domain, repressing any form of competition, repressing innovation, and extracting large profits from their central role. The only difference between today and the world of the 1970s is that in the telco era, these industry giants had a national footprint. In contrast, these days, their dominance is expressed globally.
The core elements of this theory is that the Internet started decentralized and that it then grew more and more centralized around a couple of really big companies. The challenge, this narrative implies, is finding a way back to a more decentralized Internet.
The quote above comes from a post on CircleID by Geoff Huston, who is chief scientist at APNIC, and it goes on to discuss the causes of this centralization. Huston suggests that maybe the economies of scale in advertising are to blame:
It appears that there are very real dynamics of scale in advertising. Advertisers want their message to be seen by the greatest number of potential consumers, and advertising platforms want to provide a service to the greatest number of potential advertisers. The larger the platform, the greater the potential of the platform to meet both of these requirements. The result is intense pressure to consolidate in this market. And that is what we are seeing. And, of course, the Internet is now inexorably entwined in this situation. We wanted the Internet to be “free,” and today, it certainly is. I can use a search engine to query a massive compendium of our accumulated knowledge and more without paying a cent! I can access tools and services for free. I can store all my digital data without cost to me. It sure looks like a “free” Internet to me! But is the inevitable cost of all of this one of overarching centrality and dominance by a small collection of global megalithic enterprises that distorts much of the rest of the global economy?
Ibid.
It is worthwhile teasing out the mental model here — advertising benefits from scale, and so is driven to consolidation and this consolidation also centralizes the Internet as a second order effect.
This is an important and interesting line of thought that deserves to be pursued further, but it also needs to be deepened. There are a number of questions here that I think need to be explored:
How do we measure centralization? Do we get different results if we look at traffic flows, economic flows and time flows? In what dimensions and under what definition is the Internet centralized?
If advertising is driving the consolidation, what drives advertising? Is it the efficient allocation of attention?
Are there other possible models of what causes centralization? A certain centralization is observable in many phenomena – and we can think of this as examples of the 80/20 percent split across nodes and activity / traffic / time in any network.
What is an ideal state? Was the Internet really decentralized or was it just not yet cohesive? An alternative description could be something like: “the early Internet condensed into the current one”.
What other networks have effectively been decentralized? The telcos are not decentralized today, so that is an interesting example.
Do networks decentralize or do they become less central in larger context?
What is the average degree of centralization of human networks?
These questions, and many more would present an opportunity to examine the centralization thesis more in detail. It does seem as if this axis could be an interesting analytical category to look more closely at. We could start by looking at things like the centralization of cities.
The concept of deep uncertainty is intriguing and important – as defined it is:
Deep uncertainty exists when parties to a decision do not know, or cannot agree on, the system model that relates action to consequences, the probability distributions to place over the inputs to these models, which consequences to consider and their relative importance. Deep uncertainty often involves decisions that are made over time in dynamic interaction with the system.
The thing that really is exciting is that the concept of deep uncertainty starts from the assumption that we will not get to a shared model of what is going on here, and it then asks the question: well, can we still distill principles for decision making?
All problems that allow for a shared model can be played out, those that don’t require other means of attack.
My instinctive answer is that this would be impossible – but as usual instincts cannot be trusted.
Turns out that there is a whole research field here – led by organizations like The Society for Decision Making Under Deep Uncertainty (DMDU). The DMDU has, among other things, published an interesting checklist for making investments or planning decisions, that is available here, but more than anything they are systematically trying to study how we can make progress on issues where we do not share a model of reality.
The thing that makes this interesting in my mind is the idea of classes of uncertainty – and different kinds of uncertainty. The two major classes I think are most helpful are the following:
Uncertainty over probability distributions – we agree on a model of the world and on possible outcomes, we just lack any ability to assess or distribute probabilities – and refuse to succumb to the short-hand of 50/50. This is a kind of “outcome uncertainty” – we know the outcomes, but lack methods of determining their probabilities.
Uncertainty over what is going on here, the inability to even list – partially or exhaustively – possible outcomes. This is more akin to some kind of “model uncertainty”, where we simply seem impossible to meaningfully model the situation.
This second kind of uncertainty is intriguing. It seems, intuitively, that we can always model a situation, if nothing else as a very, very simple system, but perhaps what we want to say here in the second case is that our model does not reach the George Box-threshold?
Box famously said that “all models are wrong but some are useful”. Maybe what we have in “model uncertainty” are models that are not useful – or better: models that are not even wrong. That would then lead us to ask why that is — what could possibly lead to model uncertainty?
One example would be highly particularistic events – like the encounter with an alien intelligence – where there are no patterns to draw on for making any kind of model (but…octopuses?). Or perhaps a situation with too many moving pieces? Maybe there are two versions of model uncertainty – one would be where we cannot cross the Box-threshold, and one would be where any model would be useless by the time it was formulated because the object of modeling would have changed so that the model slips below the Box-threshold again?
This general question – are there classes of problems where the modeling of the problem always takes more time than it takes for the problem to morph beyond the usefulness of the model so achieved – seems overly theoretical but is probably much more common than we think.
Now, back to the real issue though. What do we do when we lack the ability to model a system? How do we elicit a pattern from the problem? By interaction. That is another notable item in the definition. But is interaction and alternative to modeling or a variant on modeling? How are modeling and interaction different?
This is probably a trickier question than we think – interaction can happen without a model of the system you interact with, it is a way to probe the system and elicit a pattern from it. The pattern is not a model of the system an sich, but a model of its behavior.
Again – is this a distinction without a difference? Are any models of systems not just models of behavior? I think there is a case here for arguing that modeling a system allows us to more accurately predict any range of behavioral patterns that the system can produce, but modeling the behavior only gives us access to that specific behavior and so does not say anything meaningful about the other possible states that the system can take or behaviors it can produce.
Predicting a boxer in the ring will not tell you how they vote.
Behavioral models and system models differ. That seems almost mundane, but when we deal with really complex systems it matters — and interaction is how we elicit behavioral patterns. Dialogue, experiment become key, rather than abstract description. Complexity is best explored through interaction.
So much more here, and a lot of conceptual confusion on my part. Need to dig deeper – but the usefulness of the notion of deep uncertainty alone makes it worthwhile to pursue more.
A recent study in PNAS suggests that we can at least start thinking about that through inversion – the study of what intolerance is. By looking at the areas of the brain that activate during polarized responses etc a group of researchers are now arguing that intolerance is strongly correlated with a need for certainty.
If polarization is the response to uncertainty in our society, we can readily see how the increasing complexity of societies, and the weaker ability we have to predict the future, contribute to dividing us.
But why do we react this way?
If we take an evolutionary stance to the problem it is not hard to see situations in which the best reply to uncertainty is not Julia Galef’s scout mind, but rather the solider mind that narrows the options and favors belief over doubt – any kind of threat, still unidentified, from the forest or, say, a sound in the middle of the night, should not awake curiosity as much as a strong set of beliefs about that threat – wrong or right.
What is intriguing is that it does not matter what the belief is. If you meet social uncertainty with a strong national-conservative set of beliefs or if you meet it with socialist convictions; nature does not care what you believe, it is only interested in making sure you believe something, because doubt will kill you.
Beliefs will kill you too, if you believe the wrong thing, but in many cases it does not matter if you believe it is a sabre tooth tiger in the forest or an enemy tribe, it only matters that your response is the same. The content of the belief is immaterial to the evolutionary outcome – you leave and survive.
If you choose to believe it is just a branch snapping, and stay – then you run the risk of being eaten, of course, but here is the beauty of the fact that both left and right-wing polarization exists – could evolution have favored randomized belief as to divide groups up in the face of uncertainty?
Whoa! This is speculation across many dimensions – not least suggesting that there is a group selection element here – but entertain the thought for a bit, acknowledging that it is fanciful. What if evolution divides groups in the face of uncertainty to ensure that survival is not dependent on a single belief?
And you really do not need to imagine that there is group selection going on; you could just assume that we have a threshold to beliefs in the face of uncertainty that triggers if too many people seem to believe the same thing or not many people enough seem to believe it. It would be the opposite of peer pressure — peer opposition, built into our natures – an evolve tendency to disagree to a degree.
If nothing else it suggests that uncertainty – not probability – is still dominating our societies, now through polarization, even if the economist profession has done away with uncertainty as a concept and replaced it with calculable risk.
