x.com/gmiller/status/2033933582830563500
1 correction found
There are no known methods of Reinforcement Learning from Physiological Feedback.
This is too absolute. Published reinforcement-learning systems already use physiological signals and patient state as feedback in medicine, including closed-loop glucose control and ventilation management.
Full reasoning
The sentence says no known methods of reinforcement learning from physiological feedback exist. That is incorrect.
There are published RL methods in medicine that explicitly learn from physiological measurements or biosignals:
- Closed-loop blood glucose control: Fox et al. (PMLR, 2020) describe an RL system for diabetes management that uses continuous glucose-monitor data and insulin-delivery actions in an artificial-pancreas setting. The paper states: "Here, we develop reinforcement learning (RL) techniques for automated blood glucose control."
- Systematic review of RL for diabetes control: A 2020 PubMed-indexed review says RL algorithms have been successfully implemented for "closed-loop, insulin infusion, decision support and personalized feedback" in diabetes, where the patient's body is the environment and blood glucose is the controlled physiological variable.
- Ventilation control from biosignals: A 2023 npj Digital Medicine study developed and validated an RL model for ventilation control during emergence from anesthesia using ventilatory and hemodynamic parameters from 14,306 surgical cases and notes that the model was developed from "real-world data" and "high-resolution intraoperative biosignals."
These examples do not show that RL has solved longevity or full-body biology. But they do directly contradict the blanket claim that there are "no known methods" of reinforcement learning from physiological feedback.
3 sources
- Deep Reinforcement Learning for Closed-Loop Blood Glucose Control
Longitudinal data streams captured from wearables, like continuous glucose monitors ... Here, we develop reinforcement learning (RL) techniques for automated blood glucose control.
- Reinforcement learning application in diabetes blood glucose control: A systematic review
RL algorithms could be used to design a fully closed-loop controller ... reporting successfully implemented RL algorithms in closed-loop, insulin infusion, decision support and personalized feedback in the context of DM.
- Development and validation of a reinforcement learning model for ventilation control during emergence from general anesthesia
We develop and validate a reinforcement learning-based Artificial Intelligence model for Ventilation control during Emergence ... The strength of this study is that the reinforcement learning model was developed based on real-world data ... consisting of high-resolution intraoperative biosignals.