[Paper Review] Choose Your Weapon: Survival Strategies for Depressed AI Academics

Julian Togelius, Georgios N. Yannakakis|arXiv (Cornell University)|Mar 31, 2023

Scientific Computing and Data Management8 citations

TL;DR

A Point-of-View article outlining multiple strategies for AI academics facing resource disparities with big tech, including scaling down, niche focus, collaboration, startups, and policy suggestions for universities and industry.

ABSTRACT

Are you an AI researcher at an academic institution? Are you anxious you are not coping with the current pace of AI advancements? Do you feel you have no (or very limited) access to the computational and human resources required for an AI research breakthrough? You are not alone; we feel the same way. A growing number of AI academics can no longer find the means and resources to compete at a global scale. This is a somewhat recent phenomenon, but an accelerating one, with private actors investing enormous compute resources into cutting edge AI research. Here, we discuss what you can do to stay competitive while remaining an academic. We also briefly discuss what universities and the private sector could do improve the situation, if they are so inclined. This is not an exhaustive list of strategies, and you may not agree with all of them, but it serves to start a discussion.

Motivation & Objective

• Motivate AI researchers to reflect on career options in the context of unequal compute and data access between academia and industry.
• Provide a catalog of pragmatic strategies for sustaining impactful AI research within academic settings or through alternative paths.
• Encourage discussion on how universities and industry can better support academic researchers in a rapidly scalable AI landscape.

Proposed method

• Present a diverse set of strategies (e.g., give up on grand goals, scale down, reuse or analyze existing models, pursue small models or edge AI, focus on specialized domains, seek niche problems, or spin out/startups).
• Discuss practical considerations and trade-offs for each strategy (compute costs, publication venues, collaboration, and IP implications).
• Advocate for university and industry reforms to better support high-risk, high-gain research and open collaboration.

Experimental results

Research questions

• RQ1 What strategies can AI academics adopt to remain relevant and productive given growing compute/SOC constraints of academia vs. industry?
• RQ2 How can universities and large industry players collaborate to support open, high-risk AI research that benefits the broader community?
• RQ3 What role do startups, collaborations, and niche domains play in sustaining academic AI innovation?

Key findings

• There is a perceived widening gap between academic and industry compute resources, influencing researchers’ career choices.
• Strategies include scaling down, focusing on toy problems, reusing or analyzing pretrained models, pursuing small models, or targeting niche domains.
• Startups and collaborations with industry can provide advantages but introduce IP and publication trade-offs; universities can reform incentives to encourage high-risk research.
• Open-sourcing models and stronger academia-industry collaboration could help alleviate competitive pressures.
• Universities are encouraged to reward risk-taking and support foundation-model style collaboration to keep academic innovation open.