GFlowNets: Sampling as sequential decision making

Policy about the use of AI agents and chatbots

This document describes the policy about the use of AI agents and chatbots in the development of the course work, in particular the project code, the final report and the presentation.

Lire la politique sur l’utilisation des agents d’IA et des chatbots en français

Summary

In a nutshell, the most important aspects of the policy are the following:

  • The use of LLM chatbots, AI agents and coding assistants is generally discouraged in this course, as it may be detrimental to the learning objectives.
  • For all the writing assignments, especially the final report, it is explicitly not allowed to generate text with LLM chatbots or AI agents. All the text must be written by you. Spell-checking tools are allowed.
  • For coding assignments, especially the final project, it is not allowed to use coding agents and the generation of large amounts of code from scratch. The use of chatbots is allowed as support and for the generation of small pieces of code. You are expected to understand all the code used for the project.
  • The use of chatbots for the generation of new ideas and brainstorming is explicitly discouraged. You are expected to learn about the topics of this course by reading trustworthy bibliography and attending the course sessions, to then come up with your own ideas.
  • All uses of LLM chatbots, AI agents, and coding assistants must be disclosed and described in the final report, in a separate section titled “Disclosure of the use of AI-based tools”.
  • The violation of this policy may imply failing the course.

This policy corresponds to level 2 - allowed, but restricted (utilisation permise, mais restreinte) - of authorisation, according to the Université de Montréal framework about the use of generative AI.

Motivation

The objective of this course is that you learn both the theoretical aspects of GFlowNets as well as its practical applications. As graduate students, you are also expected to learn and practice skills such as coding, organising a coding project, plan the work, training machine learning models, critically analysing the results, drawing conclusions, writing a technical report, etc.

The widespread deployment of AI-assisted tools such as chatbots and assistants has resulted in their adoption by many people for many of the tasks you are expected to carry out in this course, such coding, writing and even brainstorming and ideation. However, in the best case scenario, it is highly uncertain whether using these tools would help you develop these skills. In the worst case scenario, it is certain that the abuse of these AI-based tools would be detrimental to the accomplishment of the learning objectives of this course. For example, if you ask a chatbot or agent to generate the code and final report of your machine learning project, you may obtain an OK-looking result, but you will likely not learn anything beyond interacting with a chatbot, which is not an objective of this course.

For these reasons, the use of AI agents, assistants and chatbots is generally discouraged for the development of the course work, and it is explicitly not allowed for certain tasks such as generating the text of the written assignments and non-negligible amounts of code.

The ability to clearly write technical and scientific content to describe complex ideas and processes is a very important skill to develop. The tangible output of writing - a text - may be replicated by a language model, but the cognitive processes that take place while we write and the skills that are acquired while we learn to write cannot be replicated if we outsource the process to an automatic tool. In other words, learning how to write technical and scientific content teach us to think better. For these reasons, text generation is specifically not allowed and your expected to write all the content of the final report.

Similar to the pedagogical benefits of writing, coding plays a fundamental role in the process of learning complex algorithms and methods in computer science and machine learning in particular. The process of writing code, reviewing it and testing is not only an important skill that is part of the objectives of the course, but it also helps deepening and fixing our understanding of the methods at hand.

Besides the reasons mentioned above, it is also important to mention that the extensive use of AI-based tools presents environmental and ethical concerns. The proliferation of AI-based tools and their widespread deployment has triggered the construction and extension of large data centres around the world. These data centres require extensive use of raw materials and consume large amounts of water and electricity, which result in environmental impacts that pose serious concerns in the current context of climate crisis. Furthermore, both the development and use of large AI systems have been associated to multiple ethical concerns, including the discrimination of certain groups, increase of inequality due to the appropriation of resources and wealth accumulation, copyright infringement, and the spread of mis- and dis-information, among many other issues.

References