LN: MIT AI Lab WP 316 (1988) — How to Do Research at the MIT AI Lab

Full citation: MIT AI Lab. (1988). How to do research at the MIT AI Lab (AI Lab Working Paper 316). Massachusetts Institute of Technology. http://www.ai.mit.edu/lab/howto.html


Pass 1 — Bird’s Eye View (5 Cs)

CAssessment
CategoryGuide / handbook for graduate-level academic research
ContextProduced internally at MIT AI Lab for incoming graduate students. Edited by David Chapman. Not peer-reviewed, but widely cited as a graduate research primer.
CorrectnessBased on collective tacit knowledge of experienced AI researchers. Practical rather than empirical. No falsifiable claims.
Contributions(1) Three active reading questions for papers. (2) Framework for choosing a research direction. (3) Advice on writing, collaboration, and intellectual honesty.
ClarityVery clear. Written for intelligent non-experts (first-year graduate students).

Relevance: ⭐⭐⭐⭐⭐

Foundational document for the MIT Student Method / reading framework used in PUMA.


Pass 2 — Content Grasp

The Three Active Reading Questions

The document proposes that when reading a research paper, the reader should actively maintain three questions:

Q1 — “How can I use this?” Map the paper’s contribution to your own research problems. What technique, dataset, metric, or framing could apply?

Q2 — “Does this really do what the author claims?” Critical scrutiny of methodology, evaluation, and conclusions. Are the experiments rigorous? Is the claim over-stated? Are baselines fair?

Q3 — “What if…?” Generative thinking. What if the assumption were relaxed? What if this method were applied to a different domain? What would break?

These three questions are not a formal named methodology in the document — they are informal suggestions within a broader guide. They are presented as habits of mind, not a structured protocol.

Additional Key Themes in WP 316

On choosing research problems:

  • Work on problems that matter. Don’t chase fashionable topics.
  • Ask: “If I solve this, will anyone care?”
  • The document distinguishes between “thesis problems” (safe, completable) and “research problems” (uncertain, high-value).

On reading the literature:

  • Read deeply, not broadly. Better to understand 10 papers completely than skim 100.
  • Follow citation chains backwards and forwards.
  • For every paper: understand why the authors made each design choice.

On writing:

  • Writing is thinking. Don’t wait until you have results to start writing.
  • Clarity of writing reflects clarity of thought.

On intellectual honesty:

  • Report negative results. Don’t over-claim.
  • Distinguish what you know from what you believe.

Pass 3 — Virtual Reconstruction

What makes WP 316 enduring

The document has survived 35+ years because it captures tacit knowledge that graduate programmes rarely teach explicitly: how to read actively, how to choose problems, how to sustain research momentum.

The three reading questions (Q1/Q2/Q3) are the most-cited element because they give a concrete cognitive framework for transforming passive reading into active intellectual engagement.

How WP 316 Relates to Keshav’s Three-Pass Method

DimensionWP 316 (1988)Keshav Three-Pass (2007)
OriginMIT AI Lab internal guideUniv. of Waterloo, ACM SIGCOMM
StructureInformal questions (Q1/Q2/Q3)Formal 3-pass protocol with 5Cs
FocusHabit of mind while readingSystematic reading procedure
IntegrationOverlaps with Pass 3 mindsetOperational protocol for all passes
Use in PUMAQuestions to ask during any passReading protocol for SLR papers

Synthesis for PUMA: WP 316’s three questions are best understood as the mental attitude to bring to Keshav’s Pass 3. They are not competing with the Three-Pass Method but complementary to it. Together they form the PUMA reading protocol:

  • Keshav Pass 1–2: Assess relevance and grasp content (systematic)
  • WP 316 Q1–Q3: Activate deeper thinking during Pass 3 (generative)

The “MIT Student Method” Name

In various Spanish-language projects guides and prompting communities, the combination of WP 316’s three questions with other cognitive and AI-prompting frameworks has been labelled informally as the “MIT Student Method.” This label does not appear in WP 316 itself, nor in any subsequent MIT publication. It is a pedagogical shorthand that collects:

  • WP 316 Q1/Q2/Q3 active reading questions
  • Keshav Three-Pass systematic protocol
  • AI-prompting frameworks (RCOIF, EGI, AMI, DRCA) for AI-assisted research

For PUMA, the vault uses Keshav Three-Pass as the primary verifiable reading protocol, and WP 316 Q1/Q2/Q3 as the active reading mindset applied during Pass 3.


PUMA Integration

Used in: PN-MIT-Student-Method-Complete (permanent note), MIT-AILab-Reading-Practice

Supports: OE1 (SLR quality), the qualitative depth of literature review, Section 1.4 (methodology)

Three questions applied to key PUMA papers:

  • Angermeir 2025: Q1 → use reproducibility taxonomy as checklist; Q2 → is “only 5 executable” properly operationalised?; Q3 → what if we measured reproducibility after 1 year?
  • Calikli 2025: Q1 → use non-monotonic finding to justify 4-strategy design; Q2 → does their dataset cover our PM domain?; Q3 → what if CoT examples include chain-of-thought reasoning, not just labels?

References to follow up

  • Keshav, S. (2007). How to read a paper. ACM SIGCOMM CCR, 37(3). ✅ Already in vault.
  • Hamming, R. W. (1986). You and your research. Bell Communications Research Colloquium Seminar. — Companion piece to WP 316.