“Escaping From Model Land” by Erica Thompson

Written by

in

,
  • (Why do we separate nature from art? They are representations of each other.)
  • We use models to interpret data
  • Risk is quantifiable uncertainty
  • Uncertainty is unquantifiable understanding
  • Few surprising events are Black Swans
    • (We often just ignore the warning signs)
  • Everything has a value judgement
  • “Past performance is no guarantee of future success”
  • Models can only answer certain questions in certain ways
  • When judging models, we are looking for one that is “adequate for the task”
  • All models are “wrong” in that they are not reality
  • Stylized facts are generally true but not true in detail
  • Artists are modelers
  • Community standards are important
  • Stochastic parrots repeat what they have been taught but do not understand any of it
  • We like just so stories in models, but they are not usually actually reflective of reality
  • Statistics are for big data where models are for small data
  • (Be careful with percentages, it is easy to lose reality in them)
  • Probabilities are reflections of beliefs not reality
  • “Counter performative” is when we avoid the prediction of the model
  • We need to differentiate between scientific observation and value judgements
  • (And value judgements embedded in scientific observations. We can only look at so many things and we use value judgements to decide what to look at.)
  • We need to accept that other people have different opinions because they think about things differently and make different value judgements
  • We have a bigger risk of humans descending to the level of AI than AI ascending to human levels
  • You may lose your wealth but it is difficult to lose your privilege
  • What matters is if the model helped you make better plans, not if it was correct
  • The question of how we should act in the world is always a value judgement
  • Principles for modelling:
    • Define the purpose
      • Make sure the people and purposes are included
    • We know nothing for certain but we certainly know something
    • All models require a value judgement
      • If you cannot see them, ask the people who are affected by the model to identify the value judgments
      • All for differences of opinions
    • Write about the real world
    • Use many models

Comments

Leave a Reply