Project Collective Intelligence

Doel vak

By the end of this course, students will be able to:

1. Understand the notion of collective intelligence in both natural and artificial systems, and recognize the recurring design pattern that connects flocking, foraging, task allocation, collective decision-making, synchronization, and opinion dynamics. (Knowledge and understanding)

2. Distinguish the microscopic and macroscopic scales of a collective system — the design of the individual agent on one hand, and the emergent group-level behavior on the other — and articulate how local rules give rise to global order. (Knowledge and understanding)

3. Use simulators to conduct empirical research in the field — implementing agent-based models, measuring emergence through appropriate order parameters, and producing reproducible results. (Applying knowledge and understanding) (Making judgements) (Lifelong learning skills)

Inhoud vak

This course introduces the principles and methods of collective intelligence — how groups of simple, locally interacting agents produce robust, adaptive, intelligent group-level behavior. The course follows a single arc: the field map, the individual agent, the emergent collective, and engineering applications.

Themes

Six canonical domains are covered: collective motion (Vicsek, Reynolds boids), foraging and stigmergy (ant trails, ACO), division of labor (response thresholds), distributed decision-making (honeybee quorum, best-of-N), synchronization (Kuramoto), and opinion dynamics (DeGroot, Hegselmann–Krause).

Concepts

Across domains, students develop a unifying toolkit: the agent–environment–interaction framework, local rules with limited sensing, positive and negative feedback, phase transitions and order parameters, mean-field analysis, network topology, and robustness under perturbation.

Research and practice

Collective intelligence is an active interdisciplinary field at the intersection of computer science, statistical physics, biology, and the social sciences. The course connects these foundations to the engineered systems where they now operate — swarm robotics, federated and multi-agent machine learning, distributed consensus and blockchain protocols, multi-agent LLM systems, and crowd computing — and teaches students to recognize the collective-intelligence structure underlying them.

Outcome

Students leave with a working vocabulary for any collective system, hands-on experience implementing and analyzing an agent-based model, and the ability to formulate, measure, and communicate a collective-intelligence research question — transferable to research and to the design of modern decentralized systems.

Aanvullende informatie onderwijsvormen

The course combines literature study and moderated discussions with experimental work in computer simulations. Students will work in small groups.

Regarding genAI, students are allowed to use Level 3 — Use of generative AI is permitted, with explanation and reflection
In this course, generative AI (GenAI), such as ChatGPT or similar tools, may be used as part of the working process. The student must explicitly explain how GenAI was used and critically reflect on the quality, usefulness, and limitations of the output. Assessment also focuses on how GenAI has been used consciously, critically, and responsibly. Incorrect use of GenAI, including failure to comply with the requirements for explanation and reflection, is considered fraud. If established, the submitted work or assessment result will be declared invalid, and the Examination Board may impose further measures.

Toetsvorm

Students will work in teams of three, four, or five (depending on total student numbers).

At the start of the course, each team chooses one collective-intelligence topic from the canonical domains introduced in Lecture 1 (collective motion, foraging and stigmergy, task allocation, best-of-N decision making, synchronization, opinion dynamics) or another topic agreed with the instructor. The team works on this single topic throughout the four weeks, deepening it from a basic implementation to a full empirical study, and presents the final result at the end of the course.

They will be subject to a compulsory evaluation every week that involves a presentation and demonstration with simulation videos. Answering questions during the group presentation and discussion is part of the assessment.

Grading is based on:

1) Quality of the presentation and degree of achievement of the task

a. Week 1 — Foundations (formative). The team sets up the simulation environment, presents the chosen topic, and demonstrates a basic working agent following the skeleton developed in Lecture 2.

0% of the final grade — only a formative grade will be given, assessed at the end of Week 1.

b. Week 2 — First results. The team presents an initial working simulation of the chosen topic, with a first qualitative characterization of the emergent collective behavior.

20% of the final grade, assessed at the end of Week 2.

15% will evaluate the group presentation and delivery.

5% will consist of individual evaluations where students will be asked individual questions, and the quality of the answer will be assessed.

c. Weeks 3 and 4 — Empirical study and final presentation. The team deepens the work into a full empirical study, applying the analysis framework introduced in Lecture 3. The deliverable must specify the agent, the environment, and the interaction, and must include at least one order-parameter curve as a function of a swept control parameter. Week 3 is a mid-point checkpoint; Week 4 is the final presentation.

40% of the final grade (20% assessed at the end of Week 3, 20% at the end of Week 4).

30% (15% each week) will evaluate the group presentation and delivery.

10% (5% each week) will consist of individual evaluations where students will be asked individual questions, and the quality of the answer will be assessed.

2) Short report of the approach and experimental results

40% of the final grade. The report is written in a group, and a group grade will be given. The report must specify the agent, the environment, the interaction, the order parameter, and the control parameter, and must include the corresponding figure(s) with error bars over multiple runs.

Final grade composition

45% group presentation, performed in a controlled environment

15% individual presentation (answers to questions), performed in a controlled environment

40% group report

Resit and late submission

In case one of the evaluations listed under point (1) is missed (the Week 2 milestone, or one of the Week 3 / Week 4 components), students have the opportunity to submit a video containing a presentation and a live demo. In this case, the grade of the specific component will be capped at 55%.

In case of late submission of the report, a penalty of 10% on the grade of the report component will be applied for each 24 hours of delay. A maximum of 5 days delay is allowed (corresponding to a maximum penalty of 50% on the grade of the report component).

Literatuur

Introductory papers on swarm robotics and collective dynamics, e.g. the review here https://link.springer.com/article/10.1007/s11721-012-0075-2

Aanbevolen voorkennis

• Coding skills in Python are necessary.
• Students are required to download, install, and use simulation software.
• Students should also have the will to switch to a more research-oriented mindset (the course is not about following instructions, but proposing research problems and approaches to solve them, with a strong motivation).