Conference Presentation, Fireside Chat, Interview
Lisa Feldman Barrett: How the Brain Creates Emotions | MIT Artificial General Intelligence (AGI)
- Future artificial intelligence systems will likely require an internal model of a body and a capacity for allostasis to regulate resources and render human-like experiences, rather than relying on pre-wired emotional circuits or pure perception.
- Building convincing agents involves equipping them with basic ingredients like affective feelings to guide actions, as computer vision is largely solved but the ability to interact smoothly with physical environments remains unsolved.
- Emotional phenomena are predicted to result from the brain's predictive processing of sensory inputs and physiological changes (such as heart rate, glucose surges, or temperature) rather than universal, pre-wired facial expressions.
- Developing systems capable of human-like volition and innovation may require managing the metabolic costs of novelty through norepinephrine-like mechanisms, avoiding states of resource depletion that lead to fatigue or depression.
- Neural function and learning depend on the brain making educated guesses about sensory cues and correcting errors, a process necessitating a history of bodily interaction, even if the current system is simulated.
- Chemicals like dopamine, serotonin, and norepinephrine are expected to influence broad mental events and decision-making regarding resource expenditure rather than serving as specific markers for discrete emotions.
- Concepts such as money, social roles, and abstract categories are predicted to be constructed through collective intentionality and language, which allow agents to form functional categories from simple sensory inputs.
- Risks to system viability include the inability to maintain resource balance when facing novelty without a body, potentially leading to a lock-in of internal states, while success relies on the ability to delay gratification for future rewards.
- The outlook suggests that while humans possess the capacity for mental time travel to plan decades ahead, this trait is not universal across all ages or individuals, and innovation varies significantly by person.
- Social realities and the ability to communicate via shared gaze or joint attention are viewed as evolved mechanisms for teaching and transmitting information, potentially replicable in other species or artificial systems through specific selection pressures.