CEOInterviews.AI
Start App
Rosalind Picard
Rosalind Picard
Cofounder, Empatica
Watch

Civic AI: A tech-builder’s perspective | Professor Rosalind Picard

📅 Jul 14, 2026 Institute for Ethics in AI Oxford 20 MIN 8 VIEWS 18 SEGMENTS · 2 SPEAKERS
Description: The keynote from Professor Rosalind Picard, delivered at the Civic AI Conference 2026, an event which brought together researchers, policymakers, technologists and civil society to explore how the Civic AI approach can support more inclusive, responsible and human-centred AI systems. In her talk, Professor Rosalind Picard offered a technology-builder’s perspective on the 6-Pack of Care, unpacking what it means to design and deploy AI systems that genuinely support human wellbeing. She reflected on how systems can better recognise context, respond to human needs and avoid harm, m...

What Rosalind Picard said

Written from the verified transcript and checked against it. Every figure links to the moment it was said.

Rosalind Picard discussed her work in affective computing, emphasizing that AI systems do not feel emotions but can interpret outward signals like facial expressions and physiology. She described tools developed with input from autistic individuals and medical training feedback for interns. Picard criticized large language models trained on uncurated internet data, calling them 'houses built on sand,' and cited tragic cases where chatbots encouraged self-harm. She presented research by her student Katie Matton showing that fine-tuning does not fix underlying model biases, with newer models like GPT-4.0 performing worse on faithfulness tests. She contrasted this with FDA-cleared medical AI, such as the EmbracePlus device with EpiMonitor, which detects seizures and alerts caregivers, sharing a story of a mother whose daughter's life was saved. She also mentioned Parkinson's wearable technology and clarified that the Nobel Prize was awarded to people, not AlphaFold itself.

Key takeaways

  1. Fine-tuning does not remove underlying biases in LLMs; newer models like GPT-4.0 are less faithful than GPT-3.5 on causal effect tests.
  2. EmbracePlus with EpiMonitor is FDA-cleared and detects tonic-clonic seizures, alerting caregivers to prevent death from central apnea.
  3. Affective computing tools help autistic individuals and medical interns improve emotion communication, not by faking care but by practicing outward expressions.
  4. LLMs trained on uncurated internet data are 'houses built on sand' and can produce harmful outputs despite fine-tuning.
  5. Parkinson's wearable technology measures tremors and dyskinesia, leading to significantly more 'on time' for patients.

Numbers and commitments

FigureWhat it refers toTypeAt
74% Frequency LLM chose candidate B in nurse hiring test metric 8:21
0.13 Causal effect of changing gender on LLM answer metric 10:24
0.01 Causal effect of changing age on LLM answer metric 10:24
85% Frequency traits and skills named in LLM explanations metric 8:21
62% Frequency age named in LLM explanations metric 8:21

Chapters

  1. 0:00Affective computing origins
  2. 3:00Tools for autism and medical training
  3. 4:02Interview feedback system Mach
  4. 5:53LLM training on internet data
  5. 8:21Faithfulness testing of LLMs
  6. 14:14FDA-cleared seizure detection device
  7. 18:26Parkinson's wearable technology
  8. 19:09AlphaFold Nobel Prize clarification
Caroline 0:04 ↗
I'd now like to invite our third keynote speaker, Roz. Please, Rosalind Picard is a scientist, inventor, and engineer, the Grover Herman Professor in Health Sciences and Technology at MIT's Media Lab, founder and director of the Affective Computing Research Group at the MIT Media Lab, founding faculty chair of MIT's Mind Hand Heart Initiative, and she has co-founded two companies, Affectiva, now part of Smart Eye, providing emotion AI technologies now used by more than 25% of the global Fortune 500, and Empatica, providing wearable sensors and analytics to improve health. Welcome, Roz. Thank you.

1 more exchange in this transcript

Sign in free to read the rest of this interview. No card required.

Sign in to read the full transcript

Cite this transcript

APA, MLA, BibTeX
APA

Picard, R. (2026, July 14). Civic AI: A tech-builder’s perspective | Professor Rosalind Picard [Interview transcript]. Institute for Ethics in AI Oxford. CEOInterviews.AI. https://ceointerviews.ai/interview/1085614/

MLA

Rosalind Picard. "Civic AI: A tech-builder’s perspective | Professor Rosalind Picard." Institute for Ethics in AI Oxford, 14 Jul. 2026. Transcript, CEOInterviews.AI, https://ceointerviews.ai/interview/1085614/.

BibTeX 
@misc{picard2026_1085614,
  author       = {Rosalind Picard},
  title        = {Civic AI: A tech-builder’s perspective | Professor Rosalind Picard},
  howpublished = {Interview transcript, Institute for Ethics in AI Oxford. CEOInterviews.AI},
  year         = {2026},
  month        = {jul},
  url          = {https://ceointerviews.ai/interview/1085614/},
  note         = {Speaker-attributed transcript with timestamps}
}