A Manifesto for Democratically Accountable Modelling
By Bruce Edmonds and Gary Polhill
Policy Actors (politicians, ministers, advisers, etc.) are increasingly facing complex problems – situations whose complexity surpasses the human mind to comprehend in all its detail. One response in such situations is to use complex models to get some leverage on it and enable the exploration of options. The problem with this is that, if an actor trusts a model, they derogate to a model they do not fully understand. In such circumstances it is essential that other researchers can fully inspect, try out and generally check such models.
As a recent government report documents, complex computational and mathematical models are being increasingly used as supporting evidence for decisions that may affect the public. This may be as simple as a spreadsheet or as sophisticated as a machine-learning algorithm. Future public trust may be dependent on this – just saying “well the experts told me this” will not be enough, for any particular expert can be wrong.
In the below we use the following terms.
- A complex model is any computer simulation program or set of mathematical equations that are used as a model of some issue or phenomenon, where the behaviour/outcomes of the model are not obvious to the person interpreting them. In other words, when a decision-maker has to trust that the model works, rather than relying on their own deliberation (at least to some extent).
- Public policy is any decision, strategy or intervention that impinges on the lives of people and/or the environment over which the public has a legitimate claim. This includes all policies made by democratic bodies (except a few made for reasons of security or safety) and many decisions by other institutions where it is established that these are of legitimate public concern (e.g. employment practices, or exploitation of environmental resources).
- Other policy is any decision, strategy or intervention that may impinge on the lives of people and/or the environment about which the public has a legitimate interest.
- The model content is a set of files and documents that allow anybody to fully inspect, alter, experiment with and check the model. This minimally includes access to the source code itself (under a suitable open license) and sufficient documentation to understand how the code is run, its key ideas, its parameters and outputs and how the code works.
- “Fully available to the public” means the legal right to inspect, alter, experiment with and check the model without hindrance or fear of prosecution or other detriment.
- “Maximally available to the public” means as available to the public as commercial, safety or personal interests allow. In other words, there is a good reason for any restriction on inspecting, experimenting with or checking the model. This depends on the circumstances, the authors’ legitimate interests, the potential impact on others and the extent to which the impact is of legitimate interest to the public.
- “Essentially not influenced” means that the decision would have been the same if the model had not existed.
We, who sign this demand the following. Namely that:
- If a complex model is used to help inform public policy, then the model content should be fully available to the public (preferably whilst the policy is still changeable).
- If a complex model is used to help inform other policy, then the model content should be maximally available to the public.
- If public policy is not influenced by a complex model then no support or legitimacy from that model should be claimed to justify that policy.
- Those who use complex models to help decide public policy take due care concerning how it is used and how the models are interpreted for policy purposes.
- Researchers and analysts who offer complex models to help inform public policy should not claim abilities, reliability or legitimacy that are not rigorously established for those models.
We further affirm that we will make our best endeavours to adhere to these ourselves.
Sign at: http://chng.it/zCwYqqtm
This manifesto is published by “Democratically Accountable Modelling” (DAM) – a campaign for transparent, accessible and responsible use of complex models in policy. DAM reflects the view of its members and is not part of any other institution.
For more information see their website (https://damcampaign.org) or their twitter feed (@dam_campaign). The founders of DAM are Bruce Edmonds (email@example.com) and Gary Polhill (firstname.lastname@example.org). All enquiries to be made to them.
 Computational modelling: technological futures (2018), the Government Office for Science and the Council for Science and Technology, UK. http://gov.uk/government/publications/computational-modelling-blackett-review