.. an interesting effect while thinking about my aim is how to describe: having fun .. it's about the rhythms I can choose from and sort for an satisfying output .. all the intervals of cadence, power, speed, recovery, force, tension, spans, focus summ up to an instant tactic in the given surrounding to rise the bar, to look for advantages and soften the current absences .. because loosing that fun stops my drive [to surprise me and my senses] .. the surprise is not being that ore that fast, it's the awareness of structures I live with & start to play with ..

---

*
___/\___/\___/\___ the s u r f b o a r d of the K u r p f a l z is the r o a d b i k e .. oSo >>

Francois wrote:

But there are also new methods being investigated to address such issues (e.g. transfer learning)

I looked at transfer learning but it seems to be stuck right where general AI is - handwaving..
have you seen actual useful results ?

https://scottlocklin.wordpress.com/...from-a-great-height/

https://www.nature.com/...es/s41746-019-0122-0
for the why TL. 31 is a good read and has led to actually useful clinical stuff, with interpretability.

There's an ongoing discussion about interpretability of models btw, and whether it's always needed.
And folks in general (myself included) think that it is not. At least, not right away. There are many areas of medicine where things were done in a specific manner without really understand why it worked, e.g. going beyond causality, and finding the mechanistic interpretations. We should thrive to get there. But in the meantime, if we can show that something works, without understanding why is still fine.

I guess, as an analogy if everyone started using Alan's platform and within a year we were all under 9hrs at Ironman, no one would really care to understand how a RNN worked ;-)

thanks, interesting - so a few domain-specific limited results. That's good but ;-)

By analogy with the self-driving car bit - In 1995 two Carnegie Mellon robotics professors wrote software and installed a desktop in a Pontiac minivan, which then drove itself across the US achieving 98.2% autonomous function.
No-one has done better in uncontrolled environments, since then.
The last mile of AI (and transfer learning) is the hard part..

Domain specific is the key part. Multi purpose AI I think is not possible. Our team does a lot of clinical predictive stuff. And even within medicine we focus on very specific areas.

I think these will be great efficiencies for people when AI is developed and hopefully will be optimized through an app on the phone that also integrates nanotechnology into knowing our nutritional levels and what intake or foods we need. If an alarm went to your smart watch and said you needed vitamin C or to eat an orange - that might be helpful. Real time nano nutrition and fitness connected to the watch would be optimal. I don't think we are that far away. I think eventually we will take pills that have nanotechnology in it and these bots will have certain functions like surveillance and analysis, and also DNA editing.

But, for instance if you just did a workout and had a meal, if an alarm came up and said you needed 10g more protein for full muscle repair and optimization, that would be perfect.

Efficiency in training and doing the exact right amount of effort and intake to achieve your desired result will be the key benefit to nanotech in bio and the AI movement. Also, it should point some big arrows to the quality of food in the general population. America's arteries are clogged and the second nano hits bio nutrition, healthcare companies will probably factor this in and it will change the system hopefully and eradicate certain things from our diet, as we currently use our bodies like a giant black box.

---

https://www.strava.com/...tes/zachary_mckinney

While nanotech might be a few years into HumanGo's future , we are committed to building a holistic platform and a large part of that will be nutrition. Your point on seeing the body as a "black box" is a good analogy and ties in nicely with Francois's point on the extent to which we need to see what's going on inside the black box to be able to foresee/achieve a result.

In other words, while internal surveillance 'nanobots' would provide an incredible level of explanation of what's going on inside the system, they are not entirely necessary to see the patterns between an input and an output of the system, i.e. a certain food intake and its relationship with health and performance (providing both of those things are being tracked, which is a much nearer term reality for current technology), e.g. per your example, your A.I. coach 'sees' that when your personal daily Vitamin C content falls below 100mg when in heavy training, your personal risk of illness goes up by 5x, so one of the simple 'optimized actions' it presents to you is to add an extra Orange to your diet when training volume exceeds 14 hours a week etc. - beamed to your app as "Looks like you've got a big training day ahead. Have you had your Orange today?" This level of personalized nutrition is entirely feasible in the very near future and we are currently exploring partner options with nutritional apps to facilitate it.

Even with current technology, ML to date has only scratched the surface of potential application to health and wellness. We have some very exciting times ahead of us!

---

Alan Couzens, M.Sc. (Sports Science)
Exercise Physiologist/Coach
Twitter: https://twitter.com/Alan_Couzens
Web: https://alancouzens.com