Computer simulations and experiments Viola Schiaffonati February, 12 th 2015
Overview 2 Simulations and computer simulations Computer simulations and experiments Explorative experiments From verifiability to reliability A fallibilist perspective
Simulations 3 Simulation as the reproduction of the behavior of a system using another system, providing a dynamic representation of a portion of reality (Hartmann 1996) Example: scale model of a bridge built to test the resistance of some materials to atmospheric agents Model not enough for the purpose Need of putting it in a controlled physical environment Model executed in the reality by means of the action performed by the environment
Model + execution 4 Simulation: model + execution Model as a representation of the aspects relevant to a specific purpose Execution of the model as the process performed by an agent (natural environment, human being, computer) Simulation as executable representation
Computer simulations 5 Simulation based on a computational model and executed by a computer Computational model: formal mechanism able to manipulate strings of symbols (to compute functions) Computer simulation as the process resulting from the execution of a computational model representing the behavior of a system whose state changes in time Not every execution of a computational model is a computer simulation!
Classification of simulations: a tentative proposal 6 Possible according to two different dimensions Type of model Agent performing the execution Computational model (formal mechanism able to compute functions): execution performed by a computational machine (computer) Mathematical model (e.g. system of equations): execution performed both by a computer and a human being Physical model (representation of a selected part of the world): execution performed by the nature itself
Computer simulations and experiments 7 Variety of positions Simulations just as techniques for conducting experiments on digital computers (Naylor 1966) Simulations as intermediate tools between theories and empirical methods (Rohrlich 1991) Simulations as substitutes for experiments impossible to make in reality (Hartmann 1996) Simulations as novel experimental tools (Humphreys 2004) Simulations as special kinds of experiments (Simpson 2006)
Simulations used as experiments 8 Possible in case of coincidence between purposesof simulations and experiments Discovering new explanatory hypotheses, confirming or refusing theories, choosing among competing hypotheses But not necessary Simulation with no experimental purposes in mind (e.g. simulation of a protein folding process for didactical purposes)
(Epistemological) justification for such use 9 Similarity between techniques of experimentation and techniques of simulation (Winsberg 2006) Involvement of data analysis Constant concern with uncertainty and error Experiments and simulations (used for experimental purposes) as controlled experience Ability and necessity of controlling the features under investigation Choice and control of the experimental factors (artificial setting)
Practical reasons 10 (Computer) simulations used as experiments To make several accelerated experiments exactly repeated and with a high precision degree non always possible in empirical cases To perform experiments difficult to make in reality being free from many of the practical limitations of real experiments To carried out experiments impossible to make in reality, such as studying parts of reality not physically accessible
From techniques to explorations 11 Different ways of using computer simulations as experiments Just as techniques to derive numerical solutions to systems of differential equations with non analytical solutions (typical of physics) As explorations to develop new hypotheses, models, and hints to be further verified (typical of biology)
Computer simulations as explorations 12 Simulations used to explore knowledge without the grounding in real physical processes (neither in wellfounded theories nor in experimental data) Simulation results suggesting new regularities not extractable from the model assumptions otherwise Simulations seen as explorative experiments getting some hints for new knowledge to be further investigated Explorative in the sense of not giving the assurance of the correctness of a conjecture, even if helping to build it up
Trusting simulations 13 Reasons for trusting simulations Simulation models strongly grounded in well-founded theories Simulation of physical phenomena already modeled by equations Simulation model making the numerical treatment possible Experimental data against which testing simulation results Simulation of artificial phenomena and processes comparable to real phenomena and processes Not always possible
Validation problem 14 Simulations used in substitution of traditional experiments (see biology) To explore reality without the grounding in real physical processes (neither in well-founded theories nor in experimental data) What reasons to believe in simulations and their experimental power even in these cases? How to validate simulation results?
Modeling from above 15 Simulations not anymore directed toward eliciting the implications of well-formulated theoretical models No systems of equations for representing biological development Simulations directed to augmenting the exploration opportunities Possibility for users to test reactivity and adaptability of the model in progress Theoretical model under construction shaped by simulation results Modeling from above (Fox Keller 2003)
The modeling framework in biology 16 Simulation as the construction of a hierarchy of models (Winsberg 1999) Transforming models into algorithms and making the algorithms working General theoretical knowledge (usually not expressed as model) Mechanical model for applying theoretical idealized knowledge to real world systems Computational model for making computationally tractable the mechanical model Simulation model for simplifying assumptions, for removing degrees of freedom, for optimizing code,
Reliability 17 From verification to reliability Reliability without truth (Winsberg 2006) Not only less strong, but implying a shift in the perspective No yes or no answer, but different degrees Different strategies for assessing reliability No strategy Minimal strategy Pool of strategies
No strategy 18 No need of any strategy Very narrow use of simulation Simulations adopted just for pictorial purposes with didactical, explicative, or clarification means Simulations of the 3D protein folding or the DNA helix Simulation strongly rooted in previous knowledge Exploration/experimental purposes reduced to a minimum
Minimal strategy 19 In virtuo simulation, in vitro validation Simulation tools exploited theoretically to explore new scientific knowledge Simulation as a trial theory (Fox Keller 2003) Simulation results validated experimentally in a continuous manner
Pool of strategies 20 Variety of situations in the simulation framework Presence of incomplete empirical data Weakness of theoretical framework Difficulty in making experiments Not a single strategy, but a pool of strategies providing reasonable belief in simulation results
Inference from success to reliability 21 Sources of credibility for simulations Prior success of the model building techniques adopted Production of outcomes fitting well with previously accepted data, observations, and intuitions Capability of making successful predictions Capability of producing practical accomplishments
A fallibilist framework 22 No general rule on how to combine and use these strategies Local solutions to be founded in each different situations Not easy to understand how to locally apply them Good reasons to assess simulations reliability, but fallible (Hacking 1983) No guarantee of the correctness of the results: even with strategies applied, simulation results can be shown later to be incorrect
References 23 Fox Keller, E. (2003) Models, Simulation, and computer experiments, in Radder H. (ed.) The Philosophy of Scientific Experimentation, Pittsburgh University Press,198-21 Hackoing, I. (1983). Representing and Interving. Introductory topics in the philosophy of natural sciences. Cambridge University Press Humphreys, P. (2004). Extending Ourselves. Computational Science, Empiricism, and Scientific Method, Oxford University Press. Hartmann, S. (1996) The world as a process: simulations in the natural and social sciences in Hegselmann, R. et al. (eds.) Simulation and Modeling in the Social Sciences from the Philosophy of Science point of view. Theory and Decision Library, Kluwer, 77-100 Naylor, T. H. (1966) Computer Simulation Techniques, John Wiley Rohrlich, F. (1991) Computer simulation in the physical sciences, in Fine, A., Forbes, M., Wessels, L., eds., Proceedings of the 1990 Biennal Meeting of the Philosophy of Science Association, 145-163 Winsberg, E. (1999) Sanctioning models: the epistemology of simulation, Science in Context, 12, 275-292 Winsberg, E. (2006) Models of success vs. success of models: Reliability without truth, Synthese, 152(1), 1-19