Lange argues AI outputs should count as one input alongside our own reasons, not replace them. Run a small study: show volunteers AI answers (some reliable, some subtly wrong) and measure how much they update.
Research with people. Check your university’s ethics requirements first, get informed consent and anonymise any data you share.
As AI becomes everyone’s first source of answers, how much people defer to it shapes collective reasoning.
Output: A comparison of actual deference with Lange’s “total evidence” standard, plus interface design suggestions.
Write down what you want to learn from people.
If you are studying people, check your university’s ethics process first.
Pilot with 3–5 people and fix what confuses them.
Run it properly and take careful notes.
Share the results (anonymised) and what you would change next time.
All steps done
Share what you made, then try another project. The PDF version has room for notes.
In Anthropic and CIP’s study, about 1,000 Americans used the open-source tool Polis to write a constitution for an AI. Run the same process with 30–100 students and compare your result with theirs.
AI could make institutions better or worse at reasoning and coordinating. Prototype one tool or evaluation from BlueDot’s decision-making hackathon or Forethought’s design sketches.
Bender and Koller argue a system trained only on text cannot learn meaning. Use Neuronpedia’s free interface to Gemma Scope to inspect features for concepts that seem to need grounding (colours, spatial relations) and check whether they respond across very different wordings.
