Call for Papers

Reinforcement learning or unsupervised learning? Neural networks or Bayesian frameworks? Statistics or symbolics? Since the beginning, artificial intelligence has been driven forwards by a set of key debates over technical approaches and research philosophies.


Machine Learning: The Debates (ML-D) will be a day-long series of unscripted discussions that will match up researchers to present their best arguments and points of view on major issues within the field and around the future of the technology.

We’ll be tackling a range of questions, from technical discussions about the promise of various techniques to big questions around the social impact of the technology. Our hope is that these sessions will work to sharpen community understanding of the nuances of these issues, generate useful falsifiable predictions, open new routes of research, and change some minds.

ML-D will feature four moderated match-ups, with a pair of speakers or teams offering opening statements, rebuttals, and concluding arguments as they contest a proposition. ML-D will also feature a poster session tackling a range of simmering debates within the field. We are seeking 2-5 page research papers or short position pieces (example) which make an affirmative or negative argument around the following issues:

  • “Current deep learning methods possess certain inherent limitations that will place a ceiling on their ability to advance machine intelligence and limit the societal impact of these technologies.”
  • “The field of machine learning will not be able to contend effectively with issues of fairness and interpretability without significantly changing the core methods of the discipline.”
  • “Developments in simulation learning, one-shot learning, and meta-learning will, over time, dramatically reduce the dependence of the field on data for learning good representations.”
  • “Different attitudes to consumer privacy will drastically influence the rollout of AI on a per-country basis.”
  • “The risks and opportunities of artificial general intelligence should be taken seriously."
  • “The ever-increasing compute requirements for large-scale AI research via techniques like neural architecture search, distributed training, and domain randomization, means AI research will bifurcate into low-compute and high-compute research areas.”

This list is designed to be open-ended, and we welcome position pieces on topics beyond those mentioned. Please reach out to the workshop organizers if you have any questions on whether or not your idea would be within scope.  

Submitters will be able to specify whether their paper is exclusively for the poster session, or if they would be willing to participate in a live debate if invited. Submitters will also be able to opt out of the formal workshop archive and only present their paper during the workshop if desired. Papers may be drawn from previously published or otherwise existing work.

In addition to industry and academic researchers, we also strongly encourage doctoral students to utilize this novel debate-style workshop to share their points of view with the larger research community.