Periodic Reporting for period 2 - PICS (Prediction in Complex Systems)
Okres sprawozdawczy: 2019-09-10 do 2020-09-09
The overall purpose of the research project is to provide a new framework for understanding and evaluating predictions in complex systems. The framework focuses on three related questions: First, why do scientists in some sciences face disproportionate difficulties in making precise and accurate predictions? Second, what are some different roles that predictions can play in theory and practice? Third, given the previous two questions, are there different ways in which certain predictions should be evaluated? These questions have been investigated from the standpoint of Ecology. A revised notion of 'Ecological complexity' explains the difficulties of prediction. This is important not just for philosophical and ecological research, but also because it has implications beyond academia. Failed predictions can place life, property and the environment at risk and thus create and perpetuate a situation of mistrust between scientists and other stakeholders, including the general public. Solving the problem of prediction can help to mitigate these issues.
In philosophy of science complex systems are usually understood as those that have many interacting parts (this is the current state of the art). At first glance, this conception works well for ecological systems. However, closer examination reveals that this complexity does not explain the problems faced by ecologists, namely the difficulty in making generalizations and accurate predictions. These difficulties are only explained fully if we take into account the causal heterogeneity of ecological systems, i.e. the causes of ecological phenomena vary across space and time.
Causal heterogeneity leads to predictive failure. The process of generalizing usually involves omitting factors that are particular to each instance of a phenomenon so as to focus on what is common between the various instances. If the causal factors of the phenomenon are the same across systems, then we only need to identify these and include them in our models in order to have an accurate causal picture of the phenomenon. However, in cases of causal heterogeneity, the differences between systems are relevant causal factors, not mere details. The ‘idiosyncrasies’ of each system, are not irrelevant details, but factors that affect the functioning of the system. Predictions are based on patterns. Scientists make predictions for the behaviour of a system, based on the past behaviour of that system or the current behaviour of a similar system. Causally heterogeneous systems behave differently across space and time; hence predictions are difficult to make and have low chances of success.
Conclusions of the action:
The study of complex systems has important differences when it comes to prediction, especially when these systems are in applied sciences (i.e. scientific investigations that aim to intervene on the world rather than testing investigating theory). I have coined the phrase ‘prediction in the wild’ to denote the context of applied prediction in complex systems. The main conclusion is that we need to reconceptualize the role of prediction in this context, so that it reflects the reality of studying complex systems. The traditional values of wide generalization, predictive novelty and precision are not necessary for successful applied prediction.
In philosophy of science complex systems are usually understood as those that have many interacting parts (this is the current state of the art). At first glance, this conception works well for ecological systems. However, closer examination reveals that this complexity does not explain the problems faced by ecologists, namely the difficulty in making generalizations and accurate predictions. These difficulties are only explained fully if we take into account the causal heterogeneity of ecological systems, i.e. the causes of ecological phenomena vary across space and time.
Causal heterogeneity leads to predictive failure. The process of generalizing usually involves omitting factors that are particular to each instance of a phenomenon so as to focus on what is common between the various instances. If the causal factors of the phenomenon are the same across systems, then we only need to identify these and include them in our models in order to have an accurate causal picture of the phenomenon. However, in cases of causal heterogeneity, the differences between systems are relevant causal factors, not mere details. The ‘idiosyncrasies’ of each system, are not irrelevant details, but factors that affect the functioning of the system. Predictions are based on patterns. Scientists make predictions for the behaviour of a system, based on the past behaviour of that system or the current behaviour of a similar system. Causally heterogeneous systems behave differently across space and time; hence predictions are difficult to make and have low chances of success.
Conclusions of the action:
The study of complex systems has important differences when it comes to prediction, especially when these systems are in applied sciences (i.e. scientific investigations that aim to intervene on the world rather than testing investigating theory). I have coined the phrase ‘prediction in the wild’ to denote the context of applied prediction in complex systems. The main conclusion is that we need to reconceptualize the role of prediction in this context, so that it reflects the reality of studying complex systems. The traditional values of wide generalization, predictive novelty and precision are not necessary for successful applied prediction.
1. Papers:
- Elliott-Graves, A. (2020) The Value of Imprecise Predictions, Philosophy Theory and Practice in Biology 12:4. doi:10.3998/ptpbio.16039257.0012.004
- Elliott-Graves, A. The Future of Predictive Ecology (2019) Philosophical Topics Volume 47, Number 1, pp. 65-82.
- Elliott-Graves, A. Review of Defending Biodiversity: Environmental Science and Ethics by Newman J., Varner G., & Linquist S., (2017), Cambridge University Press.
- Mitchell, S. & Elliott-Graves, A. (in preparation) ‘Biocomplexity’ invited contribution to Stanford Encyclopaedia of Philosophy (submission deadline April 2021).
2. Monograph:
- Ecological Complexity - Under contract with Cambridge University Press
3. Selected Conference & Workshop Presentations:
- Sep 2020 – Models and Policymaking Workshop, Institute for Future Studies - Stockholm University. “Making the most of Uncertainty”
- Jul 2020 – NC3 Colloquium, University of Bielefeld. “Causal Heterogeneity: Effects on Ecological Research”
- Apr 2020 - KLI Colloquia, Konrad Lorenz Institute, Vienna, Austria. “Optimal Model Complexity in Sustainability Science”
- Feb 2020 - Are Climate Impacts Environmental Impacts? Court Re¬view of Complex Environ¬mental Know¬ledge and Im¬pacts, Helsinki Institute of Sustainability Science (HELSUS) Seminar, Helsinki, Finland. Prediction in the Wild (invited speaker)
- Aug 2019 – Philosophy of Biology on Dolphin Beach 13, Moruya, Australia. Prediction in the Wild
- Jul 2019 – International Society for the History, Philosophy and Social Studies of Biology, Oslo, Norway. “Meta-analysis as a Predictive Tool for Invasion Biology”, in the Symposium that I organised: “Tackling Bioinvasions 60 years on: lessons from the trenches”.
- Jun 2019 – Charles University. Idealization Across the Sciences Workshop, Prague, Czech Republic.Agreeing to Disagree: Pluralism about Optimal Model Complexity
- Jun 2019 - Australian National University, Australia. Workshop: Scientific Modelling: Pushing the boundaries. Title of Paper: Agreeing to Disagree: Pluralism about Optimal Model Complexity
- Elliott-Graves, A. (2020) The Value of Imprecise Predictions, Philosophy Theory and Practice in Biology 12:4. doi:10.3998/ptpbio.16039257.0012.004
- Elliott-Graves, A. The Future of Predictive Ecology (2019) Philosophical Topics Volume 47, Number 1, pp. 65-82.
- Elliott-Graves, A. Review of Defending Biodiversity: Environmental Science and Ethics by Newman J., Varner G., & Linquist S., (2017), Cambridge University Press.
- Mitchell, S. & Elliott-Graves, A. (in preparation) ‘Biocomplexity’ invited contribution to Stanford Encyclopaedia of Philosophy (submission deadline April 2021).
2. Monograph:
- Ecological Complexity - Under contract with Cambridge University Press
3. Selected Conference & Workshop Presentations:
- Sep 2020 – Models and Policymaking Workshop, Institute for Future Studies - Stockholm University. “Making the most of Uncertainty”
- Jul 2020 – NC3 Colloquium, University of Bielefeld. “Causal Heterogeneity: Effects on Ecological Research”
- Apr 2020 - KLI Colloquia, Konrad Lorenz Institute, Vienna, Austria. “Optimal Model Complexity in Sustainability Science”
- Feb 2020 - Are Climate Impacts Environmental Impacts? Court Re¬view of Complex Environ¬mental Know¬ledge and Im¬pacts, Helsinki Institute of Sustainability Science (HELSUS) Seminar, Helsinki, Finland. Prediction in the Wild (invited speaker)
- Aug 2019 – Philosophy of Biology on Dolphin Beach 13, Moruya, Australia. Prediction in the Wild
- Jul 2019 – International Society for the History, Philosophy and Social Studies of Biology, Oslo, Norway. “Meta-analysis as a Predictive Tool for Invasion Biology”, in the Symposium that I organised: “Tackling Bioinvasions 60 years on: lessons from the trenches”.
- Jun 2019 – Charles University. Idealization Across the Sciences Workshop, Prague, Czech Republic.Agreeing to Disagree: Pluralism about Optimal Model Complexity
- Jun 2019 - Australian National University, Australia. Workshop: Scientific Modelling: Pushing the boundaries. Title of Paper: Agreeing to Disagree: Pluralism about Optimal Model Complexity
The main achievement of the project is the development of a new framework for understanding 'complexity' and ‘prediction’ in applied science. The cause of the problem of prediction is causal heterogeneity, which creates difficulties for generalisation (results are not transferable across space or time). Failure to generalise leads to predictive failure.
Impacts & Wider Social Implications:
This project has far-reaching implications for philosophy of science and scientific practice. It shows that the traditional approach to improving predictions, popular in philosophical circles in the past but still prominent in many scientific disciplines, is unlikely to work in the case of complex, heterogeneous systems. This approach entails the creation of more general, unifying theoretical frameworks, under which larger numbers of particular phenomena can be subsumed. Yet if the systems are heterogeneous, then the process of generalizing involves omitting the very causal factors that are necessary to make accurate predictions in particular systems. This leads to the second implication, namely that very simple, general models will be of limited use for making accurate predictions in these types of systems, whereas more complex models will have a higher chance of predictive success. The ability to make timely and accurate predictions is expected of applied scientists by their peers, their employers and the general public. Failed predictions can place life, property and the environment at risk. Therefore, the issues of trust, credibility and accountability are of primary importance for the public understanding and appreciation of scientific research and intertwined with the feasibility of the research itself. I believe that philosophers of science, especially those who have empirical knowledge in more than one field, are well suited (and consequently duty-bound) to mediate the dissemination of scientific results not only to other researchers, but also to administrators, policy-makers, non-academic stakeholders and the general public.
Impacts & Wider Social Implications:
This project has far-reaching implications for philosophy of science and scientific practice. It shows that the traditional approach to improving predictions, popular in philosophical circles in the past but still prominent in many scientific disciplines, is unlikely to work in the case of complex, heterogeneous systems. This approach entails the creation of more general, unifying theoretical frameworks, under which larger numbers of particular phenomena can be subsumed. Yet if the systems are heterogeneous, then the process of generalizing involves omitting the very causal factors that are necessary to make accurate predictions in particular systems. This leads to the second implication, namely that very simple, general models will be of limited use for making accurate predictions in these types of systems, whereas more complex models will have a higher chance of predictive success. The ability to make timely and accurate predictions is expected of applied scientists by their peers, their employers and the general public. Failed predictions can place life, property and the environment at risk. Therefore, the issues of trust, credibility and accountability are of primary importance for the public understanding and appreciation of scientific research and intertwined with the feasibility of the research itself. I believe that philosophers of science, especially those who have empirical knowledge in more than one field, are well suited (and consequently duty-bound) to mediate the dissemination of scientific results not only to other researchers, but also to administrators, policy-makers, non-academic stakeholders and the general public.