Subjects › AGI Agents (6-week course)
W4 · Negotiation, voting & debate
Utilities and Pareto optimality, concession strategies, Borda counts and multi-agent debate.
The idea
When agents disagree you need a mechanism, not more talking. Negotiation: each side has private weighted utilities; a deal is Pareto-optimal if no alternative makes one side better without making the other worse; Conceder (β > 1) gives ground early, Boulware (β < 1) holds firm until near the deadline. Arbitration aggregates agents' answers — majority vote, confidence-weighted vote, an LLM judge, and (Exercise 4.2) a Borda count over rankings, with points by position. Debate lets agents see each other's answers and revise simultaneously. In your repo: Week04_MultiAgent_AutoGen_CrewAI/negotiation.py, arbitration.py, exercises/debate.py; Exercises 4.1–4.3.
Lesson
No lesson built yet. The worked solutions below are the material it would be assembled from.
Questions (5)
| Question | Type | Difficulty | Drift | Accuracy |
|---|---|---|---|---|
| In negotiation.py a negotiator's aspiration — the utility it demands at round t of T — is 1 − (1 − reservation… | mcq | Easy -1.10 | — | — (0) |
| Exercise 4.2 — Borda count. With m candidates in a ranking, first place earns m − 1 points, second m − 2, …, l… | numeric | Easy -0.70 | — | — (0) |
| In the negotiation demo, the buyer (side A, receives share x of each issue) weights price 0.6, delivery 0.3, w… | mcq | Moderate 0.20 | — | — (0) |
| Exercise 4.2: Borda count (first of m earns m − 1 points … last earns 0), with answers normalised (lower-cased… | mcq | Hard 1.00 | — | — (0) |
| Exercise 4.3's journal asks you to compare debates among three copies of the same local model with debates amo… | short | Very hard 1.50 | — | — (0) |