We are putting together a technical paper. It starts off by listing the problems we are interested in:
Problems that are complex and/or multidisciplinary, and where data is peripheral or at least not central to the decision-making, is unavailable for a new problem, or is out of date for a problem that is following an underlying phenomenon that can only be inferred (not easily extrapolated)
Climate Change – a problem which is large and looming. The current data does not reflect the scale of the problem, and instead one or more climate models need to be used to estimate the scope of the problem over many years. There are many cross-specialties – atmospheric modelling, oceanographic modelling, agriculture in a changing climate, natural disasters.
Complex Specifications – a specification for a complex piece of technology can run to a thousand pages, have a glossary and terms defined inline, and be full of bullets referencing other bullets within clauses. In other words, a description that is hard to understand, even by experts. It will typically have multiple areas of expertise – for a plane, that might be wings, fuselage, engines, control surfaces, avionics. If the technology is “new”, there is no body of text to support it.
Complex Legislation – similar to a complex specification, but with an important increase in complexity. It will often reference the psychological state of the person whose actions it seeks to control.
Anti-Money Laundering – a problem where large and cumbersome organizations (banks) are being used for criminal purposes by intelligent adversaries. A bank can see only a small part of a money launderer’s activities, and many techniques are used to conceal those activities, including false accounts, smurfing, false invoicing, up to infiltration of the bank’s staff. The money launderer can rapidly change strategy, and use the bank’s size and way of doing things against it.
A Rerun of Covid – Covid illustrated the difficulty of getting advice from experts as far apart as epidemiologists and economists, who have no common vocabulary. A machine that can read and “understand” and integrate text from different specialists would have been useful to bridge the gap
Strategic Planning – an obvious example is the changeover from Internal Combustion Engine (ICE) cars to electric cars (EVs). The timeframe is long, there is resistance from the public, a legislated timeframe for the transition, after which ICE cars will not be sold, scarcity of the minerals needed, parallel production of ICE and EVs, an eventual large reduction in the auto manufacturers’ workforce. Alternative technologies – solid state batteries, hydrogen fuel cells – may emerge during the transition, so planning needs to be dynamic and easily reorganized over a long period of time.
Reshoring – geopolitics may demand that some manufacturing be brought back from overseas. Most of the experience will have evaporated, together with a new generation of technology, and the overseas manufacturer is unlikely to be enthusiastic about assisting with reshoring, or training a competitor’s staff. When reshoring begins in earnest, the price for talent will be bid up rapidly, so picking the moment to start the process needs to be done with care.
Economic Modelling – “Economics is less amenable than physics to definitive mathematical analysis because it is about people, whose responses to similar phenomena change over time. We build models in our minds or computers that fit observed reality at one point in time, and reality changes. Then we have to think harder about what’s going on.” – Ross Garnaut.
The recent indecision in the US over whether the surge in inflation would be transitory or long-lasting illustrated the fragility and poor predictive power of economic models that do not merge psychology with more mathematical forms of modelling.
Modelling in English
Semantic Searching – A problem related to those above, and the solution to which they would use. The ability to create a small piece of text describing precisely what you are looking for and have it search through potentially millions of Google hits using a dictionary and the precise meanings of words, rather than just key word searching.
Lane Following – A problem that demonstrates the temporary hack of the human cognitive system in response to a textual command. Modelled problems will require a similar facility when exploring different strategies.