Why language is lossy and not enough

This is a special episode, highlighting a session from ELC Annual 2025! The true promise of AI isn’t in replicating human intelligence. It’s in developing entirely new forms of non-human intelligence that perceive and understand the world in fundamentally different ways. Jamie Lien (Co-Founder and Chief Scientist @ Archetype AI) and Rashi Agarwal (Head of AI Engineering @ GoodLeap) explore the emergence of "Physical AI" - machines that sense the world through modalities beyond human biology to form internal representations free from our biases. This means building machines that can directly sense the physical world through modalities beyond human biology, form their own internal representations and interpretations free from our biases, and then translate that understanding back to us in human terms.

ABOUT JAIME LIEN

Jaime Lien, Ph.D. is Co-Founder and Chief Scientist at Archetype AI, a pioneering startup advancing Physical AI, artificial intelligence that understands the real world through real-time sensor data fusion.

With over a dacade of experience in radar-based sensing, signal processing, and hardware engineering, Jaime’s career bridges cutting-edge research and consumer-ready innovation. Before Archetype, she led radar sensing development for Google ATAP’s Project Soli and contributed wireless communication and localization expertise at NASA’s Jet Propulsion Laboratory.

ABOUT RASHI AGRAWAL

Rashi Agrawal is Head of AI Engineering at GoodLeap, where she leads enterprise-wide AI initiatives that deliver real business impact. An accomplished speaker, she covers the latest in AI, including context engineering, evaluations, and multi-agent collaboration, while driving Applied AI innovation in the enterprise. Previously, she scaled engineering teams at Yahoo, advancing its multibillion-dollar advertising business. A passionate world traveler to 40+ countries, Rashi brings global perspective and energy to her leadership and storytelling.

SHOW NOTES:

• Archetype AI’s mission: Building a foundation model for physical reality (2:24)
• The potential for discovery: Using AI to observe phenomena humans cannot perceive (3:36)
• Augmentation vs. Replacement: Giving humans "superpowers" rather than automating them away (5:48)
• The "Perfect Storm" for Physical AI: Transformers, self-supervised learning, and commodity sensors (6:04)
• Defining “Non-Human Intelligence” and removing the constraints of human labels (8:34)
• Why language is inherently lossy and insufficient for true physical understanding (10:28)
• Real-world application: How Physical AI aids safety decision-making in the solar industry (12:35)
• Use case: Improving pedestrian safety and traffic signaling in Bellevue (14:51)
• The biggest engineering leadership challenge: Embracing the “messiness” of real-world data (16:21)
• Q&A: Why we shouldn't teach AI physical laws, but let it discover them (18:50)
• Q&A: Validating models when there is a defined ground truth vs. subjective language (20:49)
• Q&A: Compute requirements and the future of active learning at the edge (22:05)

LINKS AND RESOURCES

• Video version of Jaime and Rashi’s session at ELC Annual 2025

This episode wouldn’t have been possible without the help of our incredible production team:

Patrick Gallagher - Producer & Co-Host

Jerry Li - Co-Host

Noah Olberding - Associate Producer, Audio & Video Editor https://www.linkedin.com/in/noah-olberding/

Dan Overheim - Audio Engineer, Dan’s also an avid 3D printer - https://www.bnd3d.com/

Ellie Coggins Angus - Copywriter, Check out her other work at https://elliecoggins.com/about/

Hosted by Simplecast, an AdsWizz company. See pcm.adswizz.com for information about our collection and use of personal data for advertising.