The AI•ON values



  • Anyone can join AI•ON as a research contributor. We do not screen members based on education credentials, affiliation, or any other attribute. Contributions should speak on their own merit.
  • All code deliverables are open-sourced. In particular, all experiments released by AI•ON should be accompanied by properly documented open-source code to reproduce them.
  • The entirety of the research process should take place in the open, and in particular all communications between researchers should be public and archived.




  • We are committed to advancing science in a meaningful way. We will only publish on our website problems that we believe are both important and scientifically sound, and that can maximize the impact we can have using our limited resources. We achieve this through rigorous curation of the proposed research problems.
  • The fact that anyone can jump into the research process regardless of affiliation or credentials should lead to more robust findings and faster progress, not to shoddiness. We achieve this through an uncompromising adherence to the scientific method throughout each and every of our projects. Because the entire research process takes place in the open and involves many independent and disinterest individuals, we can set a standard of scientific rigor that would be unrealistic in academia or industry. No trust required: every claim can be quickly verified, every experiment can be easily reproduced.




  • There is a tremendous amount of student interest in artificial intelligence and machine learning, and the supply of learning opportunities is no longer keeping up with the demand. Machine learning conferences and machine learning classes at universities everywhere are overflowing. This will only get more dire in the next few years. As a result, there is a pressing need for alternative learning avenues for newcomers to the field. We are committed to growing AI•ON into a viable, large-scale process to educate the machine learning experts that the world will need in coming years.
  • We believe that the best way to learn is by working as a team on applied projects. Learn by doing, and collaborate with others both in quality of mentor and mentee.
  • We believe that a core tenet of learning and teaching, which is often overlooked, is sustainable motivation. Student projects are sometimes devoid of meaning beyond the walls of the university, which undermines student motivation and thus adversely affects the learning experience. Applied research propects proposed by AI•ON all deal with meaningful, real problems, and as result the solutions developed by AI•ON contributors are likely to find impactful applications in the real world. We believe this is the best way to solve the problem of sustainable motivation.




  • Research can be done in isolation. Learning can be done in isolation. However, we believe that contact with others can accelerate research and learning by several orders of magnitude. Communication is a catalyzer of progress, and in an age where communication is free and reaches globally, we intend on maximally leveraging this catalyzer.
  • This goes beyond teaching, passing on best practices, or mentorship. As a community, we can tackle problems that would have been out of reach for us as a collection of individuals: we can crowd-source data on a large scale, we can brainstorm ideas far and wide across scientific fields, and we may even be able to pool computing resources in the future.




The Inspiration behind AI•ON

  • Arxiv, the pulse of the artificial intelligence research community.
  • Github, and the open-source community that has grown around it.
  • Kaggle and its community.
  • The Polymath project.



The organizing team

  • Francois Chollet, Google Brain (Mountain View)
  • Gabriel Pereyra, DeepMind
  • Hugo Larochelle, Google Brain (Montreal)



  • October 18, 2016

    First launch

    AI•ON officially launched. The initial website lays out our process, our values, and lists ten starter projects.

  • December 1st, 2017


    New website, including three projects proposed by Yoshua Bengio, Hugo Larochelle, and Max Welling.