“AI’s Moral Dilemma: Are We Building Our Own Nightmare? w/ Dr. Rana el Kaliouby” — Moonshots EP #49

Episode summary

Recorded at Abundance360 (March 2023), Diamandis interviews Dr. Rana el Kaliouby, co-founder and former CEO of Affectiva (MIT spin-out for emotion recognition AI). The conversation covers building empathic AI — training systems to detect 50+ emotional/cognitive states from facial expressions — and the ethical questions around deployment. El Kaliouby discusses data bias in emotion AI, the decision to turn away intelligence agency funding in favor of values-aligned investors, and the potential for AI health companions that detect mental health biomarkers continuously. A recurring theme is that AI is a mirror of its creators and consumers; the data we feed it determines its behavior.

Key arguments / segments

[00:02:00] El Kaliouby’s journey: 25 years building emotionally intelligent machines; human intelligence includes EQ and social intelligence, not just IQ
[00:05:00] Emotion AI progress: from detecting 3 expressions to 50+ emotional/cognitive/behavioral states using deep learning on global facial data
[00:06:00] Automotive application: detecting drowsiness and distraction in drivers; computer vision enables intervention before accidents
[00:10:00] AI as mirror: large language models learn from all human content; we are inadvertently teaching them our values through our behavior
[00:13:00] AI ethics in two buckets: development (data/algorithmic bias) and deployment (privacy, exploitation); tying executive bonuses to ethical implementation
[00:15:00] Turned away intelligence agency funding in 2011 to avoid surveillance applications; nearly ran out of money but held to consent-based values
[00:18:00] USC study: PTSD patients more forthcoming with digital avatar therapists than human ones — raises questions about human-AI relationship displacement
[00:22:00] Advertising disruption: personal AIs that know you better than you know yourself could bypass advertising entirely
[00:27:00] Mental health AI companion: continuous biomarker monitoring to detect depression deviations from baseline; flag to loved ones or doctors

Notable claims

93% of human communication is non-verbal
Affectiva’s emotion AI can now detect 50+ emotional, cognitive, and behavioral states from facial expressions
USC study found PTSD patients were more forthcoming with digital avatar therapists than human therapists
Affectiva turned down intelligence agency venture funding in 2011 over surveillance concerns

El Kaliouby is promoting her work (Affectiva) and her new AI venture fund; the conversation is framed around opportunities she is positioned to capture
Episode sponsored by Eight Sleep (mid-roll ad ~[00:07:00])
Recorded at Abundance360, a paid optimist community — no adversarial questioning on emotion AI privacy risks
The “93% of communication is non-verbal” statistic is a widely cited oversimplification of Mehrabian’s research

Guests

Dr. Rana el Kaliouby — Co-founder and former CEO, Affectiva (MIT Media Lab spin-out). PhD from Cambridge University. Pioneer in affective computing and emotion AI. Now running an AI-focused pre-seed/seed venture fund.

Mapping against Ray Data Co

Low-medium relevance. The emotion AI space is tangential to RDCO’s core operations, but the ethics framework (two buckets: development bias vs. deployment privacy) is a clean mental model worth noting. The decision to tie executive compensation to ethical implementation is a concrete governance pattern. The USC PTSD avatar study is a good Sanity Check data point on human-AI relationship dynamics.

2023-06-22-moonshots-ep51-mo-gawdat-ai-threat