Lecture 4: Explanation vs Prediction Flashcards
Express explanation & prediction formally
We are interested in the explanation of some construct 𝒴through the means of 𝒳.
𝒴changes as a function of 𝒳
𝒴= 𝐹(𝒳)
Name three tasks involved in carrying out explanation in this manner
- We model f as closely as possible to 𝐹
- Data X and Y are tools for estimating f
- This estimation is used to test our explanatory hypothesis 𝒴= 𝐹(𝒳)
How are these tasks different for prediction? (4)
- Data X and Y are the objects of interest
- f is used as a tool for generating good forecasts of Y values
- f is generally constructed from data.
- Even if the underlying structure is 𝒴= 𝐹(𝒳), a function other than our best approximation of 𝐹 and data other than X could be preferable for adequate predictions of Y.
Explain the alien prediction machine though experiment
●We are visited by benevolent aliens in their
spaceships
●After a brief visit they leave, but give us a
marvelous prediction machine
●It will tell us the result of any prediction with
perfect accuracy. For instance:
○Q:“will it rain tomorrow on this spot?”
○A:“yes”
●We want to build spaceships just as the ones the aliens have.
●We try to use the marvelous prediction machine for this endeavour: ○Q:“will we build spaceships just like those of your creators?”
○A:“yes”
What can we ask next? Does this help us recreate the technology?
What is explanation according to philosophy of science and what is it called?
An explanation is a logical derivation from:
A. a non-empty set of particular observations
B. a non-empty set of natural laws
to the non-empty set of particular observations
That is to be explained.
This is the deductive nomological model
You explain the height of a flag pole using the length of a shadow and the height of the sun. What’s the problem with this?
An explanation would consist of more the process of making the flagpole and it would be weird to call it an explanation. Its more of a prediction however it satisfies the deductive nomological model
What is explanation according to statistics?
A explains B, when the Pr(B|A) > Pr(B)
What is the causal explanation?
A explains B when a causal process links B to A
that is characterised by the A’s ability to transmit
a mark in a continuous way to B.
What is the problem with the causal explanation?
You could come up with alternative explanations (?)
What is meant by the productive explanation?
If the nature is as you suppose it is, then it should produce what is to be explained.
More formally: A model that instantiates (the relevant aspects of) a theory, should produce (behaviour that resembles) the phenomenon to be explained.
Describe an important quality in a good explanation
That a phenomenon does not exist regardless of whether the factors used to explain it are there or not (e.g presenting overlapping symptoms as an underlying cause of mental disorders)
Aka necessary and sufficient
What 2 main components are involved in predictivism?
Accommodation: Adding or adjusting hypotheses to fit observations.
Prediction: hypotheses first formulated and then checked against observations.
What types of novel data are there?
Prediction of novel data:
● Temporal novelty: Forming a hypothesis =without data, then gathering data to test it
● Problem novelty: There are certain things in science that need to be solved, don’t use these in to develop the theory then see if the theory can solve them.
● Use novelty: e.g machine learning; train dataset, test dataset
What is the relationship between prediction and novel data?
Prediction only works when you use it on new data
What did Maher claim about predictive success?
Predictive success is symptomatic of the use of a reliable discovery method.
(Maher)
