In this episode of Agents of Tech, we dive into agentic AI, a new kind of artificial intelligence designed to enhance human agency rather than automate people out of the loop.
We’re joined by Dr. Niloufar Salehi, Assistant Professor at UC Berkeley and Chief Product Officer at Across AI. Her work spans healthcare, education, criminal justice, and the creative industries. She explains why most AI systems misunderstand how people actually work, and what it will take to build systems that empower rather than override human judgment.
Topics include:
- The rise of agentic AI and what it means for work
- Case studies in medicine, law, and content creation
- Why most automation fails in the real world
- The hidden risks of synthetic data and algorithmic bias
- What we can learn from creatives trying to outsmart YouTube’s algorithm
This is a must-watch if you’re thinking about AI and ethics, human-computer interaction, or the future of decision-making in high-stakes settings.
00:00 Intro to Agentic AI
00:16 Meet Dr. Niloufar Salehi
00:24 What is Agentic AI?
00:48 Hosts Discuss Human-AI Interaction
04:00 Guest Interview Begins
04:40 HCI and Interdisciplinary Design
06:34 Algorithmic Misconceptions
07:11 Xerox PARC & HCI Origins
10:48 AI in Healthcare: What Works
13:26 Translation Risk in Medicine
16:08 AI in the Courtroom
19:02 Synthetic Data: Power & Pitfalls
21:00 YouTube Creators & Algorithm Personas
26:00 Designing Interfaces Around Human Strengths
29:00 AI in Hiring, Policing & Due Process
31:00 AI That Offers Options, Not Orders
32:30 Future of AI & Human Collaboration
📘 Plans and Situated Actions – Lucy A. Suchman
• Chapter “Situated Actions” in Human Machine Reconfigurations
DOI: 10.1017/CBO9780511808418.008 (asmepublications.onlinelibrary.wiley.com, Cambridge University Press & Assessment)
📘 AI Snake Oil: What Artificial Intelligence Can Do, What It Can’t, and How to Tell the Difference – Arvind Narayanan & Sayash Kapoor
• DOI for an excerpt in Stanford Social Innovation Review: 10.48558/0Z9Z DR86 (Stanford Social Innovation Review)
• DOI for a full-length review article: 10.1215/2834703X 11700273 (Cambridge University Press & Assessment, Duke University Press)
📕 Study: Effect of Prior Diagnoses on Dermatopathologists' Interpretations…
• DOI: 10.1001/jamadermatol.2022.0000
📕 Study: Disparities in Dermatology AI: Assessments Using Diverse Clinical Images
• DOI: 10.1126/sciadv.abq6147 (Science)