We are exploring the fundamental limits that underlie human endeavor — in learning and understanding, in performance, and in prediction.
What are the fundamental limits that underlie human endeavor, and when might we bump up against those limits?
This theme explores fundamental limits in learning and understanding, in performance, and in prediction, ultimately to probe the limits of our current scientific theories.
Limits to Understanding
The historical role of research science has been to describe, predict, and manipulate elements of the world in which we live. At the beginnings of our scientific understanding, these goals were in harmony. Early theories were designed to accurately describe certain physical processes, and then a detailed study to find mechanism would lead to richer meaning behind the theory, and ultimately to successful manipulation of the system of interest. This underpinning of the scientific process is our traditional definition of what it means to “understand” one’s system of interest.
This clean definition of understanding has been blurred as the scientific endeavor expands to larger systems. From biological evolution, to economic markets, to group decision making, these new fields of study have resisted the simple mathematical compression of earlier theories. The challenge, then, is how to assess scientific knowledge for these adaptive systems: What does it mean to understand without a concise phenomenological explanation? This tension has been at the forefront in the development of complexity science.
An early example of this tension is the proof of the “four color theorem” in graph theory. The proof was begun by human analysis, showing that a non-four-colorable graph must belong to one of 1,936 counterexamples. The proof was completed by a specialized algorithm to show that all of the possible counterexamples are, in fact, four-colorable, and so there can be no counterexample and thus the theorem is true. This proof-by-algorithm is too long to verify by hand, and so we must rely on the trust of a correct implementation of the algorithm. This poses the question: if no one has fully read the proof, can we really say that we understand why the four color algorithm is true?
To truly understand complex phenomena, we may have to generate fundamentally new computational approaches. Three of these approaches — simulation, machine learning, and mathematical theory — have allowed us to predict complex phenomena. As we refine the tools and approaches of complexity science, we must also ask whether we predict in order to test our understanding, or understand in order to predict.
Limits to Prediction
Our ability to predict is the key to our success. Whether we are predicting the future of the market, future weather, future disease outbreaks, or long-range trends in technology and culture, we stand to gain by improving the accuracy of our forecasts.
The inability to predict timing and outcome magnitudes imposes massive costs in terms of economic, cultural, and economic life.
In the United States alone, from 2010-2015, 58 major weather events inflicted an approximate total of 800 billion dollars in damages. Much of this cost derives from severe storms, flooding, and droughts. Some fraction of this cost could have been defrayed by effective preparations. The healthcare system is exposed to similar liabilities. Unanticipated hospital infections account for 10 Billion dollars a year in treatment in the USA. War is in another category altogether, and the annual cost to the USA of engaging in counter-terrorism amounts to around 100 billion. The recent 2008 financial crisis has been estimated to have cost around 22 trillion dollars.
Even small improvements in our ability to predict the behavior of complex systems such as the weather, disease, the economy, and cultural shifts, would lead to huge reductions in cost and significant value to our economy and quality of life.
It is our belief that significant increases in predictive power are achievable by combining what are now largely independent methods of prediction. By sharing data, tools, and theories across domains that span meteorology, epidemiology, machine learning, economics, and evolution, we stand to generate a significant shift in our ability to accurately predict the future.
Limits to Human Performance
It is a well-attested fact that in a surprisingly large range of human activities — including individual record performances in Olympic events, team-scoring performances in collaborative sports, individual problem-solving performance in game and puzzle solving — metrics of outcome show systematic trends towards increased ability.
In a recent study Lippi et al (2008) analyze over a century of world record data (1900-2007) and report positive performance trends in every sporting category, with the strongest trends in Javelin and Shot put (70% and 50% increase in performance) with the weakest performance increase in 100M and 400M races (10% and 8%)
This heterogeneity is confirmed over a 40-year Olympic period (1968 Mexico City - 2008 Beijing Olympics) in which swimming times have fallen by around 10%, whereas short-distance running times have fallen by only 2% and long distance running times by 6%.
Similar trends are found in completion times for the Rubik’s cube puzzle over a 35 year period. The Rubik’s cube became widely available in 1980 and the first official record completion time was 19 seconds in 1982. By 2015 the cube could be completed in under 5 seconds.
In chess, peak average ratings calculated for each player over a three-year interval are available for a roughly 100 year period spanning 1900 to 2000. While there is significant variation in the data, ratings have shown a trend towards increasing by around 3-5%.
More generally intelligence test scores have shown a significant increase in value from 1930 to the present — the Flynn effect. Recent analyses show an increase of around 3 IQ points per decade with little evidence of saturation.
In team events, where multiple individuals contribute towards an outcome, similar trends can be found. In NFL games from 1920 to 2011, the average scores of winners have gone from around 17 points to 28 points a performance increase of around 60%.
These trends suggest multiple deep questions about the limits of human performance in both solitary and social settings: What are the underlying patterns and universalities in positive performance trends and do these patterns encompass all areas of human endeavor — from athleticism to scholarship? And what role does creativity play in increased performance to encompass — new training regimes; new forms of movement, new mental frameworks, and new technologies?
This research theme is sponsored by the Miller Omega Program.