Todays guest is Dr. Erica Thompson who wrote the excellent book
"Escape from Model Land", which I strongly recommend for reading.
Dr. Thompson is Associate Professor of Modelling for Decision
Making at UCL’s Department of Science, Technology, Engineering
and Public Policy. She is also a Fellow of the London
Mathematical Laboratory, where she leads the research programme
on Inference from Models, and a Visiting Senior Fellow at the LSE
Data Science Institute.
She is working on the appropriate application of mathematical
modelling in supporting real-world decisions, including ethical
and methodological questions. For instance, what is the best use
of models in climate change, public health and economics.
Making and using models in the real world is — as it turns out —
quite a tricky business and in our conversation we go deep into
the question: what constitutes model land and how can we escape
model land to achieve good results for our society from what we
learned in model land.
I covered similar topics in other podcast episodes, because this
question can be tackled from a number of different
perspectives.
The first question I ask Dr. Thompson is the obvious one: What is
model land?
“Nobody actually cares at all about what happens in your model.
[…] unless you make a claim that what happens within that model
land has some relationship to what happens in the real
world. So, how to transfer your judgement about the model to
the judgement about the real world, is the key question?”
What does Steven Wolfram mean with irreducibility of nature? Why
do we have to treat different types of models differently?
What is the difference between interpolation and extrapolation,
and why is this crucially important? Many models of complex
systems incorporate significant amounts of expert judgement,
especially when models are extrapolating. How should we deal with
such models?
“All of these decisions about model construction imply value
judgements about what we think to be important.”
Value judgements per se are not the problem — but are they shared
by the people affected by the model? How did you get to those
judegements? Are the transparent enough? Do the decision makers
know and agree with these judgements?
Under what conditions can we assess the reliability of a model?
In which category do models that are discussed in public fall,
for instance climate models? What are the butterfly and hawkmoth
effect? What is the difference between data driven vs.
“expert driven” models and what role does data quality play in
practice?
Most models also are partial models. What is incorporated in a
model? What is left out? What conclusions are we allowed to draw
from complex models? Do they highly successful data driven models
distort our expections in the more assumption driven ones?
“The model is then very much part of the story. It is not just a
prediction engine.”
There are models that influence the world and the world feeds
back opposed to models that “just” describe the world, and
performative models that actually create the reality they
describe and counter-performative models. Why is it important to
distinguish among these different types?
“Those [counter-performative models] were not made with the aim
to be accurate models and correctly predicting the future. They
were made with the aim of showing what could happen if we didn't
action which would then avoid these worst case scenarios.”
What is the difference between a (conditional or unconditional)
prediction and a scenario?
Models are tools and cannot replace judgement. But did we use
these tools accordingly? Or did models in the recent past (e.g.
Covid) inflict more harm than good on our society?
“This is exactly what models are for—to serve as working
hypotheses for further research.”, Ludwig von Bertalanffy
and
“Build a society that is resistant to model errors”, Nassim Taleb
Is this true?
Models as narrative generating devices and communication tools
and collective thinking — do we want that? Under what conditions
— like flatten the curve? And, how to avoid group think and be
captured by models?
“Plans are worthless but planning is everything”, Dwight D.
Eisenhower “Kein Plan überlebt die erste Feindberührung”, Helmuth
Graf von Moltke
So, there is a significant amount of expert judgement in building
models, but do people know that and which expert do we trust?
“Trust is a social process and expertise is socially determined.
[…] You must follow the science is saying you must agree with my
value judgements.[…]
A decision can never be science based.”
Thus, science is never value free.
Finally we talk about regulation in complex systems and how those
relate to models, the long and short term perspectives and what
skin in the game means. Is Niall Ferguson right when he says:
“Surely, once we have written a regulation for every
possible misdeed, then good behaviour will ensue. This is just an
amazing illustration of our ability as human beeings to
keep doing the wrong thing in the face of all experience. […] the
big players are actually protected by complex regulation.
[…]
Regulation is the disease of which it pretends to be the cure.”
Then, how should we regulate complex systems? Should every
politician be a scientist in the Platonic sense?
»Ultimately the definition of an expert is somebody who's
judegements you are willing to accept as your own.«
References
Other Episodes
Episode 68: Modelle und Realität, ein Gespräch mit Dr.
Andreas Windisch
Episode 67: Wissenschaft, Hype und Realität — ein Gespräch
mit Stephan Schleim
Episode 53: Data Science und Machine Learning, Hype und
Realität — Teil 1
Episode 54: Data Science und Machine Learning, Hype und
Realität — Teil 2
Episode 39: Follow the Science?
Episode 37: Probleme und Lösungen
Episode 2: Was wissen wir?
Dr. Erica Thompson
Personal Website of Dr. Thompson
UCL’s Department of Science, Technology, Engineering and
Public Policy
London Mathematical Laboratory
LSE Data Science Institute
Other References
Erica Thompson, Escape from Model Land, Basic Books (2022)
Lex Fridman #376 in conversation with Steven Wolfram (2023)
Ludwig von Bertalanffy, General Systems Theory (1969)
Cathy O’Neil, Weapons of Math Destruction: How Big Data
Increases Inequality and Threatens Democracy, Penguin (2017)
Niall Ferguson on Regulation in conversation with John
Anderson (2023)
Kommentare (0)
Melde dich an, um einen Kommentar zu schreiben.