Interview
Rosalind Picard: Affective Computing, Emotion, Privacy, and Health | Lex Fridman Podcast #24
- Affective computing is expected to remain confined to narrow, pre-specified contexts due to a persistent lack of genuine consciousness, awareness, or the ability to interpret nuance, with timelines for general emotional intelligence remaining unpredictable and contingent on societal focus.
- Technology is anticipated to evolve from purely recognizing emotions to incorporating "embodied" AI agents that offer greater engagement and task assistance, while non-embodied software will continue to serve as virtual companions, though machines will likely only mimic consciousness to appear aware rather than achieving it.
- Significant risks are identified regarding the non-consensual reading of affective states, including potential government surveillance to detect skepticism and the misuse of data by hostile entities, prompting a shift toward privacy-preserving techniques like "jamming" emotion data with fake metrics.
- The speaker predicts that AI development will increasingly be driven by societal concerns rather than pure technological ambition, with expectations for regulations to separate emotional analysis from sales agents and to extend lie detection protections to hiring, alongside a preference for regulatory "carrots" over "sticks."
- Advances in wearable sensor technology and machine learning applied to non-medical data are expected to predict future health, mood, and stress levels over a week or more, with specific applications in detecting pre-seizure biomarkers for SUDEP prevention and monitoring deep brain region activity.
- Future scenarios include the possibility of legal frameworks where AI agents are granted rights, potentially making it a crime to disconnect devices in the distant political landscape, while the industry faces the challenge of preventing AI from exploiting human emotions for profit, such as manipulating user moods to increase spending.
- The outlook emphasizes extending human intelligence to underserved populations through low-cost, green solutions for diseases like epilepsy and diabetes, rather than focusing on expensive consumer electronics, while noting that the FDA approval process remains a significant barrier compared to peer-reviewed research.
- Societal impacts are anticipated where AI interactions alleviate loneliness and extend human connections, yet the technology is not expected to replicate deep human-to-human emotional bonds like "falling in love," with a caution that relying solely on science for all knowledge may be myopic.