teaching-phd-skills

Why this exists

An argument can be perfectly valid and completely false, or true in every part and still invalid. Keeping those apart is the content of a first-semester logic course, and it is the tool you use to say why a paper’s conclusion does not follow rather than just feeling that it does not.

There is now a sharper reason. A language model produces fluent, well-formed, invalid reasoning at scale. Its output has the cadence of a careful argument — the connectives are right, the hedges are in the right places, the structure looks like reasoning — and fluency is exactly the cue humans use to judge whether reasoning is sound. Your students are calibrated on a signal that no longer tracks the thing it used to track.

External material

The Internet Encyclopedia of Philosophy on fallacies — free, thorough, and organised so you can look one up rather than read it front to back.

If you want a book: Hurley, A Concise Introduction to Logic, chapters 1 and 3. Chapter 1 is validity and soundness; chapter 3 is the informal fallacies. Nothing after that is needed here.

Governing part

None.

The mechanic

Four distinctions, which cover most of what goes wrong in engineering papers:

• Valid vs. sound. Valid means the conclusion follows if the premises hold. Sound means valid and the premises are actually true. Most bad papers are valid and unsound.
• Necessary vs. sufficient. Storage is necessary for a high-renewables grid; it is not sufficient. Conflating these produces most overclaiming in energy modelling.
• Correlation, causation, and the third thing. Including the case where your model assumed the causal direction and then recovered it, which is not evidence.
• Base rates. A test, a screen, or a heuristic that is “90% accurate” tells you almost nothing until you know how rare the thing is.

Then find three real fallacies — one in each place:

1. A published paper in your area. They are there. Read for the inference, not the topic.
2. Your own draft. This one is not optional and it is the point of the exercise.
3. AI output. Ask a model something in your field where you know the answer is contested, and read what it produces for the form of the reasoning rather than the content.

For each: quote it, name the failure, and say what would have to be added to repair it.

Deliverable

Text entry: three quoted passages, each with the fallacy named and a repair proposed.

Competency check

• Three examples, one from each of the three sources
• Each passage quoted, with a locator
• The specific failure named, not just “this is weak”
• A concrete repair for each — what evidence or qualifier would fix it
• The one from your own draft is actually fixed, and the draft link shows it

Turning it in

This replaces your reading entry for the week you do it. Submit it to that week’s WR assignment in Canvas — the same one you would have submitted anyway. It is not an extra piece of work on top of the log.

• In the text box: the fallacy you found in your own writing.
• You hand in: A text-box entry: three quoted passages — one from a published paper, one from your own draft, one from AI output — each with the failure named and a concrete repair

You do not submit anything to the milestone yet. When all five skills in M3 are done, submit the M3 — Reading and Reasoning assignment, which is just an index saying which week each one went in.

This site is open source. Improve this page.