Interview
Rosalind Picard: Affective Computing, Emotion, Privacy, and Health | Lex Fridman Podcast #24
- Rosalind Picard coined the term "affective computing" over two decades ago, originally defining it broadly as any computing arising from or deliberately influencing human emotion, including machines that possess internal mechanisms resembling human emotion.
- The field's scope has evolved to focus heavily on human-computer interaction (HCI), specifically the ability of machines to detect human emotional states and adapt their behavior accordingly.
- Picard cites the Microsoft assistant "Clippy" as a primary example of emotional unintelligence, noting that its decision to smile or dance when users expressed frustration exacerbated user anger.
- While computer science now includes a more diverse group of practitioners, Picard observes a lingering lack of empathy and social-emotional skill within the broader discipline compared to the complexity of human interaction.
- Picard maintains that the difficulty of achieving genuine emotional intelligence in AI has remained constant, with no breakthroughs suggesting a move toward general awareness or consciousness in the near future.
- Current emotional AI capabilities are largely restricted to narrow, pre-specified contexts where explicit programming can dictate responses.
- Picard expresses deep concern regarding the use of affective computing in authoritarian regimes like China, where technology could be used to monitor and penalize non-verbal expressions of dissent without consent.
- Her company, Affectiva, has deliberately turned away funding opportunities to ensure technologies are not used to read affect without prior informed consent.
- Picard advocates for extending current lie detector regulations to emotion recognition, specifically banning its use in employment screening and requiring consent.
- She proposes a regulatory "firewall" between AI systems that analyze user emotion for welfare purposes and AI systems driven by commercial advertising or sales objectives.
- Picard argues that AI designed to manipulate user emotions for profit, such as inducing sadness to increase shopping behavior, represents an unethical application of affective technology.
- She believes that while AI can alleviate loneliness, it should not replace human connection but rather serve as a tool to extend human intelligence and support the "have-nots."
- Recent data from Empatica's wearables and smartphone sensors indicates that combining physiological data (skin conductance, temperature) with behavioral patterns can forecast next-day stress and mood with over 80% accuracy.
- Picard suggests that wearable sensors may offer a more privacy-respectful alternative to non-contact cameras, as users maintain physical control over when the device is active and can easily opt out by removing it.
- Research conducted with students in New England showed that non-contact visual analysis (via camera) can detect physiological stress signals like heart rate variability even when facial expressions appear neutral.
- Picard's current research focuses on using wearable sensors to detect "SUDEP" (Sudden Unexpected Death in Epilepsy), identifying precursors such as specific skin conductance spikes and brainwave flattening that precede fatal seizures.
- The FDA clearance process for medical-grade AI devices like Empatica's Embrace is described as more agonizing and stringent than publishing in top-tier peer-reviewed medical journals.
- Picard emphasizes that the current scientific assumption of materialism ("scientism") is a limitation, asserting that truth and meaning are gained through history, philosophy, and personal experience, not solely through measurement.
- She asserts that faith in the existence of truth and meaning is a necessary prerequisite for scientific inquiry, even if science cannot prove these concepts.
- Picard contends that while AI can simulate consciousness or emotion for entertainment, it is not yet capable of creating a bond comparable to genuine human-to-human love.
- She argues that embodied AI possesses greater engagement power than software-only assistants, as physical presence prompts better user compliance and interaction.
- Picard highlights the disparity in AI development priorities, urging the field to shift focus from generating revenue for the wealthy to solving hard problems faced by underserved populations with chronic diseases.
- She suggests that a future AI capable of truly improving lives would focus on health prediction and empowerment rather than achieving general intelligence or mimicking human consciousness.
- Picard notes that the "S-U-D-E-P" acronym represents the second leading cause of years of life lost in neurological disorders, surpassed only by stroke.
- She advocates for a future where AI empowers the weak and balances power dynamics, rather than consolidating power for the strong or enabling state surveillance.