Interview, Fireside Chat
How AI Will Change Medicine -Stable Diffusion Creator Emad Mostaque
- Plans to form an AI team to conduct literature analysis and drug repurposing aimed at identifying commonalities in autism research.
- Predicts that balancing GABA and glutamate mechanisms in the brain will enable speech reconstruction and facilitate attendance at mainstream schools.
- Forecasts that deploying approximately a thousand GPT-4 equivalent models will reveal exact mechanisms for Multiple Sclerosis (MS) and identify effective treatments.
- Envisions a shift from one specialist per thousand people to personalized medicine facilitated by access to thousands of AI equivalents.
- Asserts that current technology has reached the threshold for organizing world knowledge on Alzheimer's, longevity, autism, and MS into integrated systems.
- Predicts an evolution in language models from "one-to-one goldfish memory" to a persistent "one-to-one" state capable of remembering user queries via cookies or embeddings.
- Anticipates agent-based systems comprising "you plus a thousand" language models to autonomously execute user tasks.
- Claims AI can resolve economic misalignments in healthcare, such as making viable a six-dollar annual treatment for a seven percent subset of Autism Spectrum Disorder (ASD) cases.
- Foresees a transformation in the role of doctors, where individuals have AI agents with specific objective functions acting as their personal health monitors.
- Predicts that doctors will gain access to rich individual data while maintaining privacy, leading to improved healthcare outcomes.
- Estimates that applying this information density to monitoring wound care will make elderly individuals eight times more likely to survive improper treatment and increase efficiency.
- Claims language models function effectively as few-shot learners without requiring extensive information sets.
- States that federated learning standards like FL7 HLR will enable the construction of systems supporting full federated learning.
- Predicts fully auditable open-source language models with 400 million parameters will run on-device, citing a recent announcement regarding Google Pixel phones.
- Envisions a future architecture where big global models coexist with on-device models to balance global knowledge sharing with individual privacy preservation.