Thinking tools
Thought experiments
Run an experiment in your head, following an idea to its consequences, when the real experiment is impossible, expensive or not yet needed.
Ernst Mach is commonly credited with coining "Gedankenexperiment" in the 1890s, but the Danish physicist Hans Christian Ørsted was already writing of a "Tankeexperiment" in 1811. The practice is far older: Galileo used one to attack Aristotle's physics.
A thought experiment sets up an imagined situation, keeps everything you believe about the world fixed, and asks what would happen. You do not need a laboratory, only a precise scenario and the discipline to follow it honestly. Often the result is a contradiction: the assumptions you started with cannot all be true together.
It works because many beliefs hide consequences their holders never traced. Making the scenario concrete forces those consequences into the open, at the price of a few minutes of careful thought. Mach argued that thought experiments draw on the large store of experience we already carry, so they can test an idea before anyone builds equipment for it.
A thought experiment is only as good as its setup. If the imagined scenario quietly smuggles in a false assumption, it will produce a confident wrong answer, and intuition about situations nobody has ever seen is weak. Use it to find contradictions and sharpen questions, then check the answer against the world when you can.
Example
In Two New Sciences (1638), Galileo has his character Salviati take Aristotle's claim that heavier bodies fall faster. Suppose a large stone falls at a speed of eight and a small one at four. Tie them together: the slow stone should hold back the fast one, so the pair falls slower than eight. But the pair is heavier than the large stone, so it should fall faster. Both cannot be true, so the premise is false. The famous drop from the Leaning Tower of Pisa is reported only by his later biographer Vincenzo Viviani; Galileo himself never described it.
Where it applies
Before writing code, walk a single request through the design in your head: two users edit the same record at once, the network drops halfway, the queue doubles. For an AI agent, imagine the most adversarial input a user could send and trace what each tool call would do with it.
Think about it
Which belief about your system have you never followed through to an extreme case, and what happens when you do?
Applied it to something real? Log it.
Test yourselfHow did Galileo's tied-stones argument refute Aristotle without dropping anything?
He showed that the claim that heavier bodies fall faster predicts two opposite results for the same tied pair, so the claim contradicts itself and must be false.
Related models
- First principlesBreak a problem down to what you know is true and reason up from there, instead of copying what others have done.
- InversionWork a problem backwards by asking what would guarantee failure, then make sure you avoid it.
- The map is not the territoryEvery model, metric or plan is a simplification, and the parts it leaves out can matter more than the parts it shows.