Managing Autonomous Research: Key Principles and Considerations

In recent talks at a Department of Energy-sponsored workshop and a Midwestern multinational corporation, I was asked: How does one effectively manage autonomous research? It’s a pertinent question, given that autonomous research necessitates a multidisciplinary collaboration and a digital shift in lab culture. My own track record in this domain is a blend of notable triumphs and instructive missteps. Drawing from these experiences, I’ve distilled a few principles.

?? Principle 1: Find a Vision that Resonates Among Domain Experts. During my tenure as the founding director of the AMDM Programme in Singapore in 2018-2019, together with my deputy director Kedar Hippalgaonkar and the guidance of our leadership team including Dr. Karthik S/O Kumar , we crafted a vision: to enhance productivity 10x by upskilling domain experts in machine learning (ML) & automation. This vision resonated with a team of driven individuals eager to transcend routine lab work, while still passionately pursuing science. I invite you to compare our foundational manifesto of 2018 and our evolved perspective in 2023.

?? Principle 2: "Matricize" ML & Mechatronics as a Means to Mass Mastery: The AMDM program empowered each PI-driven domain-specific project with core research efforts in ML and Mechatronics. Our strategy was two-pronged: we introduced advanced ML and automation skills into the project teams, while encouraging all team members to become hybrid domain-ML experts through fun educational initiatives. Success stories from these efforts, including outcomes from community workshops and hackathons, underscore the value of our methodology. My experience with the DARPA SD2 program (with Joshua Schrier & Alex Norquist ) further reinforced this principle, emphasizing the importance of nurturing a cohesive team culture that fosters the development of hybrid experts. Further reading here and here.

?? Principle 3: Alternate Innovation Styles for Rapid Prototyping. Autonomous lab development thrives on rapid prototyping. An analysis spearheaded by Erin Looney & Ian Marius Peters of 55 hardware startups revealed that the most agile were those where both management and teams adeptly alternate between "natural" and "structured" innovation styles as needed. To best avoid the pitfalls of a rigid preference for one style over the other, it’s recommended to work as teams and not as individuals, adopting a flexible approach akin to an active learning algorithm’s adjustability. Further in this pre-print.

?? Principle 4: Embrace Iterative Innovation. The iterative innovation model, as described here and here by former Bell Labs innovators, advocates that technologies be developed not in a vacuum, but through dynamic exchanges with the marketplace. By eschewing the binary of applied versus basic research, this model encourages an exploratory and iterative journey toward impactful outcomes, rather than treating publications as the end goal.

Confronting Reality in Academia. In the academic realm post-2020, I’ve noticed a shift towards individualistic research efforts and a gravitation away from team-based science, partially due to economic pressures and job market trends. While individually rational, such tendencies collectively hinder the progression towards fully autonomous research labs. Nevertheless, leading academic centers continue to attract excellent talent.

Optimism for What Lies Ahead. I anticipate some of the most groundbreaking innovations will originate from environments that are well-funded, mission-driven, and equipped with core ML and automation expertise. These settings facilitate an environment where teams, guided by visionary management, can swiftly transition between natural and structured innovation, propelling rapid prototyping and discovery.

Watch entities like SMART, led by Eugene Fitzgerald and his "innovation mapping" exercises, ARIA led by Ilan Gur and his "opportunity space mapping" exercises, and the philanthropically funded Future House directed by Andrew White and Samuel G. Rodriques with their "challenge book." Many of these operate with greater insulation from economic fluctuations (e.g., pressure on government R&D funding from increased debt servicing).

In Summary, the management of autonomous research is an intricate venture, requiring sustained funding, a collective vision, a conducive culture, strategic flexibility, and dynamic engagement with market and society. As we look ahead, I believe it is these attributes that will define the future of research and innovation.

This article was written by me, summarized by GPT-4, and re-edited by me.