Lecture 6 - Models, Science and the Scientific Method

Models & Uncertainties

There will always be uncertainties within models because they aren’t perfect representation of the real-world and because the humans who are involved in making them are perfect observers, collectors, thinkers, and so on of data, prcoesses, and so on.

They come in the form of:

• Inexactness, being mostly technical uncertainties that require a lot of time and effort to study (what are the parameters, what are the equations, etc)

• Unreliability, there are many methodologies and approaches when it comes to modeling, which one should be used, and how do we quantify things?

• Ignorance, epistemological, there are things we aren’t aware of, and we are unaware of things that we don’t even know are unknowns

Verification tends to be impossible because there is no valid consistency

The human component of models can introduce biases in many ways.

Uses of Models

• Although they are simplifications of the real world, we can still observe and infer processes and draw conclusions of them through surrogative reasoning

• It provides a formal framework for us to intergrate & synthesise information and allows for comparison across systems

• It allows us to evalue specific scenarios (especially useful when we can’t experiment on the real-world system) and make predictions

• They are used to optimise current monitoring programs and systems to collect better data (what, where, how often to measure?)

Rationale to Modelling

They’re needed to:

• Look into the future

• Understand the impacts of events that have not happened (yet)

• Understand impacts of human behaviour

• Understand impacts on human behaviour

General Definition of Science

There are many definitions of science, but all of them agree that it is an organisation of observed, tested, and verifiable knowledge from the real world that can be built upon.

Science is built upon with our theories, explored and tested further with our hypotheses, and represented by and verfiied with models

Science often involves the simplification of the real world by making abstractions to better understand it

Classical Scientific Method

1. Pose the question in the context of existing knowledge (current tehoreis & observations)

   1. A new question that cant’t be answered by existing theories or expands them

2. Formulate a hypothesis as a tentative answer to the question/s (your take on it)

3. Deduce the consequences of the hypothesis and make predictions

4. Test the hypothesis in a specific new experiment or theory field

   1. Does the hypothesis fit the existing world-view?

   2. If not, adjust to address the contradictions or redefine to adress major discrepancies

5. When consistency is obtained, the hypothesis can become a theory

   1. It must give a coherent set of propositions

   2. Theory is now 'subject to ‘natural selection’ among other competing theories

6. Theory becames a framework within observations and/or theoretical facts are explained and predictions can be made

Essentially, science is always change because there are always knowledge gaps or new questions for us to build upon.

Inductive Reasoning

Moving from specific instances to general statements, maing broad generalisations based on specific observations.

Our generalisations are limited to the data available/what we are able to observe. We can’t always observe everything/have all the data, so there can be discrepancies and exceptions within the generalisations

Data-drive models are all based on observations and are inductive

• We’d need to performance experiments to observe and collect data on a system

• Then, we’d analyse for patterns that allow us to make predictions (generalisation)

• Which could then lead to verifying hypotheses and adding onto existing theories (more generalisations)

Deducitve Reasoning

Moving from general to specific instances, coming to a conclusion on a specific situation based on general principles and known theory. Built using valid rules of inference where the conclusion follows from the premises.

It’d first start with the theory or known knowledge, continuing on with making hypotheses to verify and test to come up with explanations and predictions on that specific scenario

Process-based models are based on theory and are deductive

• We’d first start with the theory and pose a question based on the theory to a problem

• We’d go through the stages of making a model (perceptual, conceptual, procedural, etc)

• We’d then calibrate it using parameterisations, further testing, and so on

• We’d then evaluate it for accuracy & errors where we can either use it for explanations or continue building upon it

Abductive Reasoning

Moving from an observation to a hypothesis that accounts for the observation. Compiling available observations (things that we generally know to be true) and arriving at one (or more) hypothesis that makes the most sense/is the most likely

However, there are still uncertainties present because it is just the most likely conclusion, not the exact conclusion

Hypothetic-Deductive Reasoning

Formulating a hypothesis in a form that could be conceivably falsified by a test on observable data. Sees scientific method as being able to falsify but not prove hypotheses. But argues this is just as valuable as it’s the only way to forward science.

Normal Science

Kuhn argued that typically, we all follow and abide by normal science where we only push for current social, scientific, and world-view standards rather than for new inventive things.

However, there are ocassionally a crisis/revolution that may lead us to a new paradigm, but this is very rare and we tend to return to normal science

Modelling Learning Cycle

Models help us build upon science through a learning cycle that invovles a feedback loop

• We’d first start with a general theory, and we’d pose a question we can use to develop a hypothesis.

• From here, we’d develop models to test and verify the hypothesis

• With these models, we are able to make predictions and compare it with the observable data to see if there are any departures

• From here, we can decide whether the model needs to be revised, if more data needs to be collected, or to start over again with a new or revised hypothesis