Thoughts on combining machine learning and bioenergetics?

[EnergeticLeo]

I've been thinking about this a lot recently, since I'm a programmer by trade, with an interest in stats and machine learning (ML).
I have a sense that there is huge untapped potential in the combination of the metabolic view of health and ML.

More health-tech companies I don't think have a coherent view of health, and narrowly focus on things like genes, drug discovery, calorie tracking etc... I haven't yet seen a company providing a service that is inspired by bioenergetic view.

haidut mentioned there was a thread here a while ago about someone making an app that uses ML to predict thyroid status based on symptoms.

What are people's thoughts?

 

[challenger]

Machine learning requires vast amounts of input data to produce reliable or useful outputs. I think in theory the potential is definitely there for creating a good system, but there’s not many metrics that you could test frequently enough to train the AI in a reasonable amount of time. If you were to focus on thyroid health you would likely have to test your levels multiple times per day as well as log your diet and activity rigorously. It’s not really attainable with current commercially available tech.

I’ve seen some diet apps(MacroFactor, RP diet) get around this by using their entire user base to predict common interactions between energy intake and weight fluctuations. That system seems to be quite successful, but calorie logging is a much simpler process to begin with.

Perhaps systems such as consistent glucose monitoring using CGM devices could better lend itself to machine learning applications with the larger inflow of data.
As cortisol increases blood glucose, you could probably build a system that used blood glucose as a proxy for stress levels while taking into account the impact of your meals and using machine learning to remove the noise that eating introduces.

Would definitely be a cool project.

 

[piro444]

Machine learning requires vast amounts of input data to produce reliable or useful outputs. I think in theory the potential is definitely there for creating a good system, but there’s not many metrics that you could test frequently enough to train the AI in a reasonable amount of time. If you were to focus on thyroid health you would likely have to test your levels multiple times per day as well as log your diet and activity rigorously. It’s not really attainable with current commercially available tech.

I’ve seen some diet apps(MacroFactor, RP diet) get around this by using their entire user base to predict common interactions between energy intake and weight fluctuations. That system seems to be quite successful, but calorie logging is a much simpler process to begin with.

Perhaps systems such as consistent glucose monitoring using CGM devices could better lend itself to machine learning applications with the larger inflow of data.
As cortisol increases blood glucose, you could probably build a system that used blood glucose as a proxy for stress levels while taking into account the impact of your meals and using machine learning to remove the noise that eating introduces.

Would definitely be a cool project.

This is really good
Maybe add a “common sides” tab to know what to look for

I thought about this a couple of days ago, on how computers could (and I guess should) predict health and restore health parameters
IIRC even Tim Cook said that apple will be an health company in the future, and that’s where iWatches are going I guess
If you add a blood monitoring option (even drawn by labs, you just input your markers) you can track your health and see where you’re going
Even better if it makes suggestions based on that and how well you digest foods

The “digestion” might be the hardest part, how do you track someone’s digestion? Measure how much they bloat after a meal and input the circumference of the belly (?) maybe

 

[EnergeticLeo]

Machine learning requires vast amounts of input data to produce reliable or useful outputs. I think in theory the potential is definitely there for creating a good system, but there’s not many metrics that you could test frequently enough to train the AI in a reasonable amount of time. If you were to focus on thyroid health you would likely have to test your levels multiple times per day as well as log your diet and activity rigorously. It’s not really attainable with current commercially available tech.

I’ve seen some diet apps(MacroFactor, RP diet) get around this by using their entire user base to predict common interactions between energy intake and weight fluctuations. That system seems to be quite successful, but calorie logging is a much simpler process to begin with.

Perhaps systems such as consistent glucose monitoring using CGM devices could better lend itself to machine learning applications with the larger inflow of data.
As cortisol increases blood glucose, you could probably build a system that used blood glucose as a proxy for stress levels while taking into account the impact of your meals and using machine learning to remove the noise that eating introduces.

Would definitely be a cool project.

Yes that makes sense.
I was thinking of predicting some 'gold standard' marker of metabolism - like the basal metabolic rate test you would do in hospitals - using easy to measure metrics, like body temp, pulse, blood glucose, breathing rate etc.. kind of like broda barnes did when he correlated the body temp test with the BMR test and found the temp test to be a good predictor.
A CGM type device would be cool. maybe an always on body temp monitor; or maybe some other signals like an emotibit.

And yes, that reminds me of Ray once saying that stress may be perceived by the organism as a lack of sugar.

 

[EnergeticLeo]

I thought about this a couple of days ago, on how computers could (and I guess should) predict health and restore health parameters
IIRC even Tim Cook said that apple will be an health company in the future, and that’s where iWatches are going I guess

This is cool - I didn't know he said that.
I know Ray is generally against AI, because he thinks it could be used as a way to increase the obedience of people by telling them 'oh the algorithm said this is what you should do and the algorithm is really smart', but if we could use AI to improve people's own decision making ability that would be awesome.

I just have a strong sense that there must be some amazing un-tapped ideas that combine bio-energetics and ML/software engineering - there aren't many people in the world who are interested in both! It would be tool to create a simple but powerful tool that people could use to increase their metabolism.