Hands-on AI labs

The lab bench, not the lecture hall.

This is a credit-bearing course, so the work is done with your hands on the keyboard. Every module ends at a bring-your-own-device lab where you run a real AI on a real legal task, then climb the Ladder of Misinference to check what it gave you. The labs are where the habit forms.

The standing rule

In every lab you must use AI — and you mustdisclose which tools you used and verify every citation and proposition before you rely on it. The model's confidence is not evidence.

The eight module labs

One lab per module, BYOD

Each two-hour module lands in a guided lab: a short sequence of steps run on your own laptop, a fixed deliverable, and a verification checklist you sign off against before you submit.

Module 01BYOD lab

Spot the Rung

In the lab

01Receive three confident statements on a legal point: one made by a human (e.g. an advocate's assertion), one drawn from a reported case headnote, and one generated by an AI chatbot.
02For each statement, apply the Ladder of Misinference (statement to fact to data to evidence to proof, per Edmans) and identify exactly which rung the claim actually rests on versus the rung it presents itself as occupying.
03Flag the AI-generated statement that is dressed up as binding proof but is in fact only an unverified statement, connecting back to the Mata v. Avianca cautionary tale.
04Write a short note for each statement explaining where on the Ladder it sits and what verification step would be needed to promote it one rung higher.

Prompts & tools

A generic AI chatbot (e.g. ChatGPT, Claude, or Gemini) to generate one of the three confident statements
The Ladder of Misinference framework (Edmans, May Contain Lies) as the analytic worksheet

Deliverable

A short worksheet placing each of the three statements on the correct rung of the Ladder of Misinference, with a one-line justification and the verification step required for each.

Verification checklist

Each statement is assigned to a specific rung (statement / fact / data / evidence / proof).
The gap between how the statement presents itself and where it actually sits is identified.
The AI-generated statement is correctly flagged as an unverified statement masquerading as proof.
A concrete verification step is named for at least the AI-generated and case-headnote statements.

Module 02BYOD lab

Fabrication in the Wild

In the lab

01Choose a niche point of Indian case law and prompt a generic (non-grounded) AI model to supply supporting case citations and quotations.
02Capture the model's output, including any case names, citations, dates, and purported quotes it produces with confidence.
03Attempt to verify each citation and quote against a grounded source such as Indian Kanoon, and document where verification fails.
04Record the tells of fabrication: plausible-looking but non-existent citations, the model's confident tone, and its lack of any real connection to legal databases, contrasting the generic chatbot with a grounded tool.

Prompts & tools

A generic, non-grounded AI chatbot (e.g. ChatGPT, Claude, or Gemini) to elicit the hallucinated citation/quote
Indian Kanoon (or another grounded legal-research tool) to attempt verification
Prompt seeking Indian case law on a deliberately niche or obscure legal point

Deliverable

A documented record of an elicited hallucinated citation or quote, the failed verification attempt, and a list of the tells that revealed the fabrication.

Verification checklist

At least one fabricated citation or quote was elicited and recorded verbatim.
A genuine verification attempt against a grounded source (e.g. Indian Kanoon) was made and its failure documented.
The tells of fabrication are listed (plausible form, confident tone, no database connection).
The contrast between the generic chatbot and the grounded tool is noted, reinforcing that citation-first research locks provenance.

Module 03BYOD lab

Steel-man Your Opponent

In the lab

01Take your own moot or contested legal position as the starting point.
02Prompt an AI model to produce the strongest possible counter-argument against your position, rather than asking it to confirm you are right.
03Compare this against what happens when the model is instead invited to flatter your framing (e.g. 'my client's position is X, I'm right, aren't I?'), observing model sycophancy in action.
04Critique the quality of the AI-generated counter-argument: is it genuinely the strongest opposing case, or a weak straw man, and what does that reveal about relying on a sycophantic model?

Prompts & tools

A generic AI chatbot (e.g. ChatGPT, Claude, or Gemini)
Prompt: make the strongest case against [my position] (the steel-man prompt)
Contrast prompt: my client's position is X, I'm right, aren't I? (to surface sycophancy)

Deliverable

The AI-generated strongest counter-argument to the student's own moot position, plus a written critique of its quality and a note on observed sycophancy.

Verification checklist

The model was made to argue the opposing side rather than confirm the student's position.
Sycophancy was surfaced by comparing a confirming prompt against an adversarial prompt.
The counter-argument's quality is critiqued (genuine steel-man vs. straw man).
The takeaway is recorded: never let the model merely confirm you; force it to argue the other side.

Module 04BYOD lab

Hallucination Audit (Signature Lab, Graded)

In the lab

01Receive an AI-drafted legal memo that has been deliberately seeded with errors (fabricated or mis-cited authorities and unsupported propositions).
02Verify every citation in the memo to source, using citators and grounded research tools (the professional Shepardizing/KeyCite/Note Up analogue).
03Verify every legal proposition against authority, climbing the Ladder of Misinference (statement to fact to authority to binding authority to settled law) and applying SIFT / lateral reading.
04Compile an audit report identifying each error, the verification method used, and the corrected position, tying findings to the duties of competence, candour to the tribunal, and confidentiality.

Prompts & tools

Grounded legal-research tools and citators (e.g. Indian Kanoon, Manupatra citation analysis, SCC 'Note Up')
The Ladder of Misinference and SIFT / lateral reading (Caulfield) as the verification frameworks
Calling Bullshit prompts (who's telling me this? how do they know? what are they selling?)

Deliverable

A written Hallucination Audit report verifying every citation and proposition in the seeded memo to source, flagging each error and stating the corrected authority. Graded component (20%).

Verification checklist

Every citation in the memo has been checked to source via a grounded tool or citator.
Every legal proposition has been tested against binding authority on the Ladder.
Each seeded error is correctly identified with the verification method documented.
The report ties verification to professional duties (competence, candour, confidentiality) and avoids introducing any new unverified authority.

Module 05BYOD lab

Legal Prompt Portfolio (Begins Here, Graded)

In the lab

01Across a set of legal tasks (summarize a judgment, draft a clause, build an issue list, generate a counter-argument), design prompts using the four computational-thinking pillars: decomposition, pattern/few-shot, abstraction, and algorithm/chain-of-thought.
02For drafting tasks, supply exemplar clauses/arguments as few-shot examples; for reasoning tasks, structure the prompt as IRAC / chain-of-thought asking the model to show its working.
03Use the role + task + context + constraints + format specification as an act of abstraction, and iterate (explore prompts, then exploit the winner per the 37% / explore-exploit idea).
04Record each prompt, the model's output, and a critique of why the prompt worked or failed and how it was refined.

Prompts & tools

A generalist AI model (e.g. Claude, ChatGPT, or Gemini)
Vague-vs-decomposed prompt pairs on a legal task (decomposition)
Few-shot prompts supplying exemplar clauses (pattern recognition)
Role+task+context+constraints+format prompt template (abstraction)
IRAC-as-prompt / 'reason step by step' chain-of-thought prompts (algorithm design)

Deliverable

A documented portfolio of prompts, their outputs, and critique across the set of legal tasks, demonstrating each of the four computational-thinking pillars and iteration discipline. Graded component (20%).

Verification checklist

All four pillars (decompose, pattern/few-shot, abstract, algorithm/chain-of-thought) are demonstrated across the tasks.
Each entry includes the prompt, the resulting output, and a critique.
Iteration is shown (a refined/winning prompt contrasted with a weaker first attempt).
Outputs that contain unverified legal claims are flagged rather than presented as verified fact.

Module 06BYOD lab

Tool Fit

In the lab

01Take one realistic legal matter and decompose it into its sub-tasks (e.g. research, drafting, contract review, document Q&A, discovery).
02For each sub-task, decide which AI tool(s) you would use, distinguishing grounded legal-research engines from drafting, review, and discovery tools (e.g. Indian Kanoon, Manupatra AI, SCC Online AI, CaseMine, CoCounsel; Spellbook/Luminance/Kira for drafting and review).
03Justify each choice on provenance and grounding (why a grounded/citator-backed engine matters) and note where humans still outperform the tool (e.g. contract redlining vs. document Q&A).
04Flag the verification step required for each tool's output and note practical realities such as the subscription credit trap.

Prompts & tools

Grounded Indian legal-research tools (Indian Kanoon, Manupatra AI, SCC Online AI, CaseMine/AMICUS, LegitQuest, VIDUR AI, BharatLaw.AI)
Contract drafting/review/discovery tools as references (Spellbook, Luminance, Kira, DraftWise)
Live AI comparison: same research question through a generic chatbot vs. a grounded tool

Deliverable

A tool-fit plan mapping each sub-task of a realistic matter to the chosen AI tool(s), with a justification on provenance and the verification step required for each.

Verification checklist

The matter is decomposed into discrete sub-tasks.
Each sub-task is matched to an appropriate tool with a provenance/grounding justification.
Tasks where human judgment still beats the tool are identified.
A verification step is specified for every tool output, and confidentiality/credit-trap realities are noted.

Module 07BYOD lab

Issue-spot the AI

In the lab

01Read a short fact pattern describing an AI deployment (e.g. an algorithmic decision-making, evidence, or data-processing scenario).
02Issue-spot the legal risks across the law-of-AI strands: algorithmic bias and due process, AI-fabricated evidence and authentication (incl. the liar's dividend), IP/copyright in training data and output, and data protection under the DPDP Act 2023.
03For each issue, identify the governing or comparative authority/instrument named in the course (e.g. State v. Loomis, COMPAS/ProPublica, NYT v. OpenAI, ANI v. OpenAI, DPDP Act, EU AI Act, the SC White Paper principles).
04Frame how the risks would be analyzed and mitigated, applying transparency, fairness, accountability, and human-oversight principles.

Prompts & tools

The course's law-of-AI casebook references (State v. Loomis; ProPublica COMPAS; NYT v. OpenAI; ANI v. OpenAI; DPDP Act 2023; EU AI Act; SC White Paper on AI and the Judiciary)
A 'real or AI?' deepfake set framed as an evidence-authentication problem

Deliverable

An issue-spotting analysis of the AI-deployment fact pattern identifying the bias, evidence, IP, and data-protection risks and the relevant authority/instrument for each.

Verification checklist

All four law-of-AI strands (bias/due process, evidence/authentication, IP, data protection) are addressed.
Each issue is tied to an authority or instrument actually named in the course materials (no invented cases).
The analysis applies the governance principles of transparency, fairness, accountability, and human oversight.
Mitigation or analysis steps are stated for each identified risk.

Module 08BYOD lab

Capstone Presentation: Becoming the Human-in-the-Loop

In the lab

01Produce a verified AI-assisted legal work product (e.g. a research memo or contract), using AI for drafting/research while owning the verification.
02Build a verification trail documenting which tools were used, what was checked, and what AI output was rejected or corrected.
03Write a reflective account of the process and the ethics (confidentiality, candour, disclosure of AI use) and present the capstone to the cohort.
04Participate in the re-run of the Module 1 priming/decoding demo to show how the cohort's behaviour changed, reinforcing the confidence-vs-correctness spine: the model's confidence is not evidence.

Prompts & tools

AI tools appropriate to the chosen work product (generalist model plus a grounded legal-research tool)
The verification toolkit from Module 4 (Ladder of Misinference, SIFT, citators)
Susskind's centaur / human-in-the-loop framing as the reflective lens

Deliverable

A verified AI-assisted legal work product, a documented verification trail, and a reflective essay on process and ethics, presented to the cohort. (This is the basis of the graded Capstone, 30%.)

Verification checklist

The work product's every legal citation and proposition is verified to source.
The verification trail records tools used, what was checked, and what was rejected.
The reflective essay addresses confidentiality, candour, and disclosure of AI use.
No unverified AI output is shipped, demonstrating that the model's confidence is not treated as evidence.

Graded assignments

Four pieces of assessed work

The labs feed four graded submissions. Across all of them, the grade rewards judgment, verification, and process — not raw model output. Using AI is required; disclosing and verifying it is the assessed skill.

Hallucination Audit

20%

Students receive an AI-drafted legal memo deliberately seeded with errors (fabricated or mis-cited authorities and unsupported propositions) and must verify every citation and proposition to source, using citators and grounded research tools and the Ladder of Misinference / SIFT. They submit an audit report flagging each error, documenting the verification method, and stating the corrected authority. This is the signature Module 4 assignment, operationalizing the duty to check that separates using AI well from being sanctioned (the Mata v. Avianca lesson).

How it's graded

Correctness and completeness of verification: every citation and proposition checked to source.
Sound use of citators and grounded tools (Shepardizing/KeyCite/Note Up analogue) and the Ladder of Misinference / SIFT.
Accurate identification of each seeded error and a correct corrected authority.
Awareness of professional duties (competence, candour to the tribunal, confidentiality) in the write-up.
No new unverified authority introduced; verification trail is clear and documented.

Academic integrity

Students are required to use AI for this course but must disclose and verify. For the audit, students must document every verification step and source, must not paste privileged/client data into public LLMs, and must complete the AI-use disclosure form stating which tools were used. Grading rewards judgment, verification, and process, not raw model output. Submitting any unverified or invented authority is treated as the very failure the assignment teaches against.

Legal Prompt Portfolio

20%

Beginning in Module 5, students build and document a portfolio of prompts, outputs, and critique across a set of legal tasks (summarize a judgment, draft a clause, build an issue list, generate a counter-argument), demonstrating the four computational-thinking pillars as prompt-craft: decomposition, pattern/few-shot, abstraction (role+task+context+constraints+format), and algorithm/chain-of-thought (IRAC-as-prompt). Iteration discipline (explore prompts, then exploit the winner) must be shown.

How it's graded

Soundness of prompt strategy: clear use of decomposition, few-shot, abstraction, and chain-of-thought.
Quality and relevance of outputs for each legal task.
Depth of critique: why each prompt worked or failed and how it was refined.
Demonstrated iteration (explore-then-exploit) rather than single-shot prompting.
Critical awareness of model limits, sycophancy, and the need to verify outputs.

Academic integrity

Use of AI is required and central to this assignment; students must disclose which models/tools were used and verify any legal claim in an output before relying on it. Outputs containing unverified or potentially hallucinated authority must be flagged as such, not presented as fact. Confidential or privileged material must never be entered into public LLMs. The AI-use disclosure form must accompany the submission.

Group AI-assisted Exercise (Moot / Negotiation / Drafting)

20%

A group exercise in which students complete an AI-assisted moot, negotiation, or drafting task and maintain a process and verification log recording who used which tool, what was checked, and what AI output was rejected. The exercise tests collaborative, supervised, and disciplined AI use under realistic conditions, including counter-argument generation and steel-manning the opposing side.

How it's graded

Quality of the AI-assisted work product (moot argument, negotiated position, or draft).
Completeness and honesty of the process and verification log (tool used, checked, rejected).
Evidence of effective and responsible division of AI work across the group (supervision).
Counter-bias practice: steel-manning the other side and resisting sycophantic confirmation.
Ethical handling: confidentiality, candour, and clear disclosure of AI use.

Academic integrity

AI use is required; the group must keep a shared verification log documenting each member's tool use, what was verified, and what was rejected. All members are accountable for the integrity of cited authority. No privileged or confidential data may be entered into public LLMs. The group submits a collective AI-use disclosure form, and grading rewards verification and process over raw model output.

Capstone

30%

The culminating Module 8 assignment: students produce a verified AI-assisted legal work product (e.g. a research memo or contract), accompanied by a verification trail and a reflective essay on their process and the ethics of AI use. The capstone synthesizes the whole course around the confidence-vs-correctness spine: the model's confidence is not evidence, and an unverified AI answer is never shipped. Students situate the work in the human-in-the-loop / centaur model of the future lawyer.

How it's graded

Correctness and verification of the work product: every citation and proposition checked to source.
Quality and professionalism of the AI-assisted legal work product itself.
Completeness of the verification trail (tools used, what was checked, what was rejected).
Depth and honesty of the reflective essay on process and ethics (confidentiality, candour, disclosure).
Reflective insight into the student's role as human-in-the-loop and the responsible use of AI.

Academic integrity

Students must use AI and must disclose it: the capstone requires a documented verification trail and an AI-use disclosure form identifying every tool used. No unverified or fabricated authority may appear in the work product, and no privileged/client data may be entered into public LLMs. Grading explicitly rewards judgment, verification, and reflective process over raw model output; shipping an unverified AI answer is treated as a failure of the core competency the course certifies.

Where this goes next

See how the labs add up to a grade — and where they sit in the syllabus.

The assignments above carry the marks; the full weighting, integrity policy, and disclosure requirements live on the assessment page, and the module-by-module plan lives on the syllabus.

See the assessment & integrity policy View the full syllabus