Two MIT Hertz Fellows reveal AI's research direction split
The Hertz Foundation's 2026 MIT-affiliated fellowship cohort includes two AI researchers whose work points in opposite directions: one toward machines that reason like people, the other toward autonomous drone swarms for defense. The five-year funding structure that makes the selection possible is the part worth examining. A standard AI PhD typically runs on advisor-controlled grants that constrain the research direction; a Hertz Fellowship decouples the student from that constraint, and the foundation has used that runway to back both cognitive-inspired research and applied military autonomy in the same cohort.
The Hertz Foundation announced its 2026 fellowship cohort, naming four MIT-affiliated recipients: Annika Marschner, an incoming mechanical engineering PhD student whose undergraduate work spanned bio-inspired robotics and surgical hardware; Alvin Q. Meng, a doctoral student in inorganic chemistry studying iron-sulfur clusters in Daniel L.M. Suess's lab; Zachary S. Siegel, an EECS PhD student in the Computer Science and Artificial Intelligence Laboratory working at the intersection of robotics, cognitive science, and AI; and Matthew Wanta, an incoming operations research PhD student whose prior work centered on machine learning for autonomous aerial systems. The foundation is awarding each fellow five years of financial support, a stipend and full tuition equivalent, on top of lifelong access to the foundation's mentoring and event programming. The Hertz Foundation says the cohort is drawn from 19 total 2026 fellows selected from across the United States, and that its alumni network now exceeds 1,300 people since the program was established in 1963.
The split inside the AI component is the part the source's list format obscures. Siegel's doctoral work, advised by Leslie P. Kaelbling, Tomás Lozano-Pérez, and Joshua B. Tenenbaum, is described in the foundation's announcement as an effort to build machines that learn and reason more like people, grounded in Bayesian inference and combinatorial generalization. The framing borrows from his Princeton senior thesis, where he modeled how humans infer the goals of others in open-ended environments. Wanta's work sits at a different pole. According to the source, his research integrates probabilistic modeling and computer vision into cooperative drone search and swarm control frameworks, including simulation architectures for probabilistic target localization built in collaboration with U.S. Special Operations Command and Army C5ISR organizations. The foundation does not present these as competing bets, but they are bets all the same: one points toward AI as a model of human cognition, the other toward AI as a tool for autonomous military operations.
The five-year funding structure is what makes the divergence worth flagging. A standard PhD in AI typically runs on four to five years of advisor-controlled grant funding, with the research direction constrained by the lab's existing grant portfolio. A Hertz Fellowship decouples the student from that constraint for the full duration. The foundation is, in effect, subsidizing cross-disciplinary risk for two students whose research questions would likely be harder to fund through a single grant mechanism. Siegel's combination of robot planning, Bayesian inference, and combinatorial generalization spans three subfields with distinct publication cultures and review standards. Wanta's integration of computer vision, probabilistic modeling, and multi-agent coordination across defense and academic settings faces a different kind of friction, since defense-adjacent AI work has its own funding pipelines, publication norms, and review barriers. The fellowship smooths over both.
The Hertz Foundation's announcement is structured as a celebration of individual excellence, quoting selection-committee co-chair Philip Welkhoff praising the cohort's fearlessness and creativity. That framing is standard for fellowship announcements and not wrong on its own terms. What it leaves implicit is the selection logic that produced two AI researchers working at opposite ends of the field's research spectrum. The foundation says the cohort was chosen for fearlessness in taking on new challenges, but the source does not describe a selection framework that explicitly weights research direction, cross-disciplinary ambition, or downstream application domain. A signal about where the foundation thinks AI research should head is not in the announcement itself; it can only be read by comparing the individual profiles against each other.
The non-AI components of the cohort are not filler. Marschner's work on bio-inspired robotic limbs and surgical hardware, and Meng's inorganic chemistry on iron-sulfur clusters, fit a broader Hertz pattern of funding physical-science and engineering research with applied medical and energy implications. The Hertz Foundation's stated track record includes contributions to advanced medical therapies and the James Webb Space Telescope, which signals an institutional orientation toward science that produces tangible systems rather than purely theoretical outputs. The AI selections extend that pattern in two directions at once: cognitive-AI research that may or may not produce deployable systems, alongside applied defense-AI research that is already operating in production-adjacent contexts.
Whether the cognitive-versus-military split is a new pattern or a continuation of one remains unclear, because the source does not address what the foundation's prior AI fellowship cohorts looked like. The source also does not specify how the foundation's selection committee handles conflicts between academic AI research and defense-adjacent AI work, or whether applicants are screened for the application domain of their proposed work. The cohort composition itself is the most concrete signal in the announcement: the same five-year funding mechanism is currently supporting AI research at both ends of the field's research spectrum.