Why Believe a Computer The Role of Quantitative Models in Science

1
Why Believe a Computer? The Role of
Quantitative Models in Science

• Naomi Oreskes

Powerpoint by Fernanda Rossi Jessica Matthews
2
Overview

• Models are used to
• Organize data
• Synthesize information
• Make predictions
• Models never fully represent so therefore make
  uncertain predictions
• Added complexity in model decreases certainty of
  predictions
• Short-time frame model vs. long-range
  deterministic model

3
Models as a Science?

• Testing is the heart of science. Although there
  is no foolproof way to define science,
  testability is the most commonly cited
  demarcation criterion between scientific theories
  and other forms of human explanatory effort.
• A single test is rarely, if ever, sufficient to
  convince anyone of anything

4
The Role of Quantitative Models in Science

• Purpose of model is to gain understanding of
  natural world
• Scientists have sought understanding to
• Advance utilization of earths resources
• Foster industrialization
• Prevent or treat diseases
• Generate origins stories
• Reflect on worlds creator
• Satisfy human curiosity

5

• Until 20th century, the word model in science
  referred to physical model
• Now model refers to a computer model
• A numerical simulation of a highly parameterized
  complex system
• Quantitative models in ecosystem science have 3
  functions
• Synthesis and integration of data
• Guiding observation and experiment
• Predicting or forecasting the future

6

• generated predictions are used as a basis for
  public policy
• government regulators and agencies may be
  required by law to establish their
  trustworthiness (How is this problematic?)
• Demand for verification or validation
• Claims about model verification are now routinely
  found in published scientific literature. Are
  these claims legitimate? Can a computer be proved
  true or false? How can we tell when to believe a
  computer?

7
The Problem of Verification

• There may be several possible configurations of
  nature that could produce a given set of observed
  results
• Therefore, any empirical data we collect in
  support of a theory may also be consistent with
  alternative explanations
• For this reason, many scientists except the view
  that theories can be proved false but not true
  (falsified but not verified)

8

• Purpose of essay is therefore to challenge the
  utility of models for prediction
• Quantitative model output has been used in
  issues such as global climate change and
  radioactive waste disposal
• But it is open to question whether models
  generate reliable information about the future
• Should we create new policies based on the
  prediction of models?

9
Naomis Opinion

• The predictions models offer to us do not aid in
  basic scientific understanding
• Our use of them does not make them important
• More complex models tend to be less accurate

10
Example of Assumptions we Make

• Stellar parallax in the establishment of the
  heliocentric model of planetary motion by
  Nicolaus Copernicus
• Flaws present in instruments we use

11
Another Example

• Earth was thought to be billions of years old
  based on the concept of uniformitarianism the
  assumption that presently observable geological
  processes are representative of Earths history
  in general
• Then, Lord Kelvin calculated the time required
  for a molten body the size of earth to cool to
  its present temperature was at most 98 million
  years, declaring the entire science of geology
  invalid
• This dismissed Charles Darwins theory of natural
  selection and for several decades evolutionists
  were in nearly full retreat
• THEN, radioactivity was discoveredproving Kelvin
  wrong.

12

• In hindsight, it is easy to see where others have
  gone wrong Astronomers thought their instruments
  were better than they were Kelvin thought his
  knowledge more complete than it was. It is harder
  to see the flaws in our own reasoning. (If we
  could see them, presumably we could correct
  them.) When computer models are involved, it can
  be more difficult still, because the systems
  being modeled are very complex and the embedded
  assumptions can be very hard to see. How DO we
  test computer models?

13
The Complexity Paradox

• The more complex the natural system is, the more
  different components the model will need to mimic
  that system
• Complexity decreases systematic bias but
  increases uncertainty
• Should we use complex or simple models to make
  predictions?

14
Models are Open Systems
15

• Hypothetico-deductive model (deductive-nomological
  model)
• Generates hypotheses, theories, or laws and
  compare their logical consequences with
  experience and observations in the natural world
• PROBLEM only works reliably in closed systems

16
Another way to understand

• 2 2 4 therefore 4 2 2
• Is a straight line the shortest distance between
  two points?

17
Open Systems

• All models are open systems
• 3 general categories into which this openness
  falls
• Conceptualization
• Empirical adequacy of the governing equations
• Input parameterization

18
Successful Prediction in Science

• Successful prediction in science is less common
  than most of us think
• Ex. 1 Meteorology Weather Predictions
• Weather prediction is not deterministic
• Spatially averaged
• Restricted to the near term
• Trial and error

19

• Ex. 2 Celestial Mechanics and the Prediction of
  Planetary motion
• Involve a small number of measurable parameters
• Systems involved are highly repetitive
• Enormous database with which to work

20

• Ex. 3 Classical Mechanics
• Scientific laws create an imaginary world that
  requires adjustments and modifications based on
  past experiences and earlier failed attempts

21
Model Testing, Forecasting, and Scenario
Development

• Short-term predictions can be helpful
• Long-term predictions cannot be tested and
  therefore do nothing to improve the understanding
  of scientific knowledge
• Naomi proposes that we focus away from
  quantitative predictions of the future and
  towards policy-relevant statements of scientific
  understanding

22
Complexity is the Strength and Weakness of
Numerical Models

• Computer models have helped us gain a better
  understanding o the Earths complex
  life-supporting processes.
• Strength - the ability to represent such systems
  is the obvious strength of models
• Weakness complex models are nonunique, their
  predictions may be error, and the scale of their
  predictions make them difficult if not impossible
  to test

23
Continued

• No sensible person would wish to court disaster
  by ignoring the threat of global warning, but
  neither would any sensible society wish to spend
  large sums of money solving a problem that does
  not exist.
• Computer models are only as strong as their
  weakest link.

24

• Has your opinion of models now changed?

25
VIDEO TIME!

• http//youtube.com/watch?vhHkbmSjSjbg
• Look for errors that could be found with this
  model process. What do we conclude about models?
  Are they useful but unreliable? Can we ever
  really know what is going to happen?