How to Conquer Artificial Intelligence: A Structured Workshop on Causal Hypothesis Generation (Planning-ness 2015)
Bibliographic record
Abstract
About Planning-ness:Planning-ness was a unique "un-conference" for creative thinkers and strategists that ran from 2009 to 2016. Its core philosophy was built on interdisciplinary cross-pollination, frequently inviting experts from domains entirely outside of advertising to provide fresh perspectives on problem-solving. The idea was that participants could borrow mental models from fields ranging from thermodynamics to poetry and apply them to the domain of planners (which in advertising world is the term for strategists who are responsible for responsible for understanding the consumer, the market, and the brand). Organized in partnership with the Account Planning Group of Canada (APG Canada) for the 2015 Toronto edition, the event prioritized "doing" over "observing." Every session followed a "How To" format, requiring a workshop component where participants could immediately apply new frameworks to practical challenges.Workshop Summary:This session explored the intersection of human intuition and machine intelligence, addressing the strengths and weaknesses that both humans and the AI algorithms available at the time brought to the table. The workshop focused on how to elicit information from human stakeholders in ways that are useful for future modeling—specifically causal search. We walked all attendees through a structured hypothesis generation approach using examples heavily influenced by the "sprinkler model" found in Judea Pearl’s Causality: Models, Reasoning, and Inference (2000).Following the introduction, participants were divided into eight groups to put these principles into practice. Each group selected an outcome they cared about and mapped out potential causes (both direct and indirect) and side effects. This provided a practical demonstration of how to push beyond the human tendency to stop at a single potential explanation and generate the kinds of inputs that would be useful for next steps.With the set of potential inputs that might matter, participants were prepared to plan out data collection and experiments that might be necessary for them to answer causal questions their team was interested in solving. The structured approach helped them to make sure not to leave out important variables for alternate explanations that could be later used by causal search and causal inference algorithms.Workshop Framework: The "Rules" for Human Elicitation:To ensure the session moved beyond standard brainstorming, we utilized an iterative, four-step process for hypothesis generation (detailed with a visual build per step on Slides 19–23). Moderators prompted participants to switch cognitive gears to the next step once they had spent a particular amount of time or had exhausted ideas for the current step.Step 1: Start with the outcome you hope to change.Step 2: For every hypothesized variable, add at least two causes. This forces the mind to move past the first "obvious" explanation and consider alternative causal paths.Step 3: For every hypothesized variable, add at least one side effect. These are additional outcomes that could potentially occur but are unintended or secondary to the variables they are related to (we used side effect more broadly to mean any unintended or secondary outcome instead of limited to bad outcomes).Step 4: Find at least two items on the board that share a common cause, and add that cause. This step is helps uncover latent variables and identifying confounders that influence multiple parts of the system.Steps 2 to 4 were repeated, resulting in a robust, multi-layered causal map.This workshop is a practical application of the first steps in the methodology presented in 10.5281/zenodo.19612598
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame distilled prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. Learned from the 10,348 direct Codex labels and 10,348 direct Gemma labels. Candidate is the union of thresholded teacher heads; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels or direct frontier model labels.
Codex and Gemma teacher scores by category
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.002 | 0.004 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.001 | 0.002 |
| Science and technology studies | 0.003 | 0.000 |
| Scholarly communication | 0.008 | 0.001 |
| Open science | 0.002 | 0.002 |
| Research integrity | 0.000 | 0.001 |
| Insufficient payload (model declined to judge) | 0.002 | 0.007 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; both teacher heads agree on what is shown here.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".