{"id":2312,"date":"2026-02-17T08:04:07","date_gmt":"2026-02-17T08:04:07","guid":{"rendered":"https:\/\/petermbach.com\/?page_id=2312"},"modified":"2026-04-05T06:47:54","modified_gmt":"2026-04-05T06:47:54","slug":"058-juan-pablo-carbajal","status":"publish","type":"page","link":"https:\/\/petermbach.com\/index.php\/podcast\/058-juan-pablo-carbajal\/","title":{"rendered":"TGC Podcast Episode 058 – Juan Pablo Carbajal"},"content":{"rendered":"\n
\n
\n
\"\"<\/a><\/figure>\n<\/div>\n\n\n\n
\n

<\/p>\n\n\n\n

The Grand Challengers Podcast Episode #58<\/h3>\n\n\n\n

Tadpoles, tolerance and why AI alignment is really a human problem<\/h1>\n\n\n\n

Guest: Juan Pablo Carbajal<\/h2>\n\n\n\n

March 3rd, 2026<\/p>\n<\/div>\n<\/div>\n\n\n\n

\n\n\n\n

Overview<\/h2>\n\n\n\n

We train AI systems to optimize for metrics \u2014 but what if the real alignment problem isn’t in the machine? What if it starts with us?<\/p>\n\n\n\n

Juan Pablo Carbajal is a physicist, interdisciplinary researcher, and passionate educator at the Eastern Switzerland University of Applied Sciences (OST). Originally from Argentina, his career has spanned physics, robotics, biomechanics, agronomy, machine learning, and water research \u2014 driven not by any single discipline but by a fundamental need to understand the dynamical systems underlying everything he encounters. He holds a PhD in Informatics from the University of Zurich, has worked at Eawag (Swiss Federal Institute of Aquatic Science and Technology) and the University of Ghent, and currently chairs Dwengo Helvetica, a nonprofit advancing STEM education.<\/p>\n\n\n\n

In this wide-ranging conversation, Juan Pablo joins host Peter Marcus Bach to reframe the AI alignment debate as fundamentally a human one. He argues that the same narrow, single-metric optimization we apply to AI systems \u2014 rewarding output over understanding \u2014 is already distorting science, education, and society at large. From the “publish or perish” crisis in academia to how we evaluate students through one-dimensional scores, the pattern is the same: pressure a system to hit a number, and unintended behaviours will emerge. Like Makoto Kern’s discussion of the AI “competence mirage” in enterprise software (Ep #52)<\/a>, Juan Pablo warns that optimizing for the appearance of intelligence is not the same as building genuine understanding.<\/p>\n\n\n\n

The conversation traces Juan Pablo’s remarkable journey from collecting tadpoles in an Argentine river \u2014 complete with a memorable food-chain disaster involving crabs and a hungry bird \u2014 through studying embodied AI at the University of Zurich’s AI Lab, where intelligence is treated as inseparable from its physical body. Along the way, he shares a profound lesson from plant biology: how nitrogen-fixing bacteria and their “cheater” free-riders coexist through tolerance rather than punishment \u2014 a metaphor with deep implications for how we design both human institutions and AI systems.<\/p>\n\n\n\n

Juan Pablo explains why physics-informed AI may be the necessary antidote to purely data-driven approaches. Using the analogy of a flat earth \u2014 perfectly valid at local scales but catastrophically wrong at global ones \u2014 he illustrates how data alone cannot reveal fundamental principles. True understanding requires integrating prior knowledge, physical constraints, and the kind of interdisciplinary thinking that our increasingly siloed institutions struggle to support.<\/p>\n\n\n\n

Topics discussed include the limits of data-driven machine learning, Gestalt psychology and machine vision, the etymology of “robot” (from the Czech for “slave”), complex systems and emergent behaviour, the role of culture as humanity’s alignment mechanism, and why understanding our history remains our best defence against repeating its mistakes. Also featured: an expert recipe for homemade yerba mate and a childhood ecological massacre.<\/p>\n\n\n\n

Episode Teaser<\/h2>\n\n\n\n
\n