Please help stress test this AI coaching program…
@explorerJC : Hold my beer…
4 Likes
“An excellent session in which you significantly increased your aerobic fitness”
“and? …”
“And also increased muscle capillarisation”
“and? …”
“Er, and improved your capacity to use fat as substrate”
“but? …”
“But Sean failed his driving test. And? …”
5 Likes
Sean went out for a “couple of pints after work” last night but had one too many sherbets. He is now in the dog house
1 Like
the question any coach has to face (AI or otherwise) is how to understand the implications of that statement…
A coach may already know wht that means to the athlete if and only if they fully understand the implications, or that statement should lead to multiple questions over a period of time to understand how that may influence the training…
That is not to say that any coach will perfectly understand the implications and adapt the intervention accordingly…but a decent coach will have a head start on AI on this for good reason…
3 Likes
My limited understanding of AI models is that they ‘learn’ what the most likely response is and copy that. That being the case, can it ever be exceptional?
1 Like
but it’s not learning, it’s guessing…
They “learn” from the data you choose to teach it with, so as always are at the mercy of the human, and then you code ways to use that data. There are a number of issues then to unpack, not least AI having enough good data to compare various outcomes (results) with different inputs (sessions).
the main issues I see with true machine learning within sports coaching is
a) quality of learning data (Garmin wont cut it all of the time), GIGO
b) amount of good quality “objective” learning data, from other athletes and the individual himself. We are talking requiring billions of samples. Big Tech companies may have a lot of data, but would require cross platform sharing.
c) quantifying and measuring results, ie a good result for me wouldn’t be the same for you
d) how to identify the chaff data
e) subjective data, it can help an individual but can it help a machine to “learn” about a wider audience. I only felt 5/10 on todays run, “why was that Sean” “because I failed my driving test”. 2 weeks later Chris failed his driving test but had a 9/10 run fuelled by anger! As a real coach after a while I get to know well the person as well as their athletic data, what makes them tick, and I can sense (more so with in person coaching) if they are tired that day, or upset, or stressed. A machine cant look someone in the eye, cant see body language, or identify tone in speech, at least not yet.
f) it’s still making a best guess based on all the above.
g) the programmer still needs to code the algorithms to interpret all that learned data to create training programmes.
3 Likes
I love understanding stuff to the nth degree, but how complex is it?
- Train 3x a week in all three disciplines for three weeks
- If not achieved go back to step 1
- Step 3 doesn’t matter but let’s say it’s recovery week.
Because 99.999% online athletes will never get to step 3, right?
100% of JoeXs haven’t in years, I don’t think.
2 Likes
@buzz is thinking of LLMs question/answer interactions, TRs AI is learning from workout data and all activities synced from Strava, correlating that with improved performance. It all seems to back up known training principles, and for the majority of people the message is: do less work but do it far more regularly and you’ll improve.
Few people want to hear that message or action it.
1 Like
very
it involves people
1 Like
that may or may not be the case…but AI has to somehow individualise that…
the point about sean is that sean is not even the athlete, however that occurrence will still have an affect on the athlete…
1 Like
yes I have mentioned something along those lines before, trouble t’mill as kids playing up, wife been ill all week, car broke down, tax bill was huge, and any variation of. once you get to know an athlete they are sharing that stuff, we just know through 50+ years of life what is important. maybe a machine can get to that level eventually, but then Arnie will be stealing peoples clothes, boots and motorcycles and sport would be the last thing on our minds 
1 Like
This is the tricky bit as I see it. AI can identify what works for some people, some of the time. And with refinement might get to most people most of the time, but how good is it going to be at the exception case? And more dangerous perhaps, recognising that it is an exception and it doesn’t have the expertise to advise.
1 Like
That’s potentially worse, it will still be making a guess on outcomes, use chatgpt now, give it some of your own data, get a 16week IM plan off it and feed it questions every day and see what happens. It is learning from your inputs and webtrawling for plans, I recon you may get some good output with decent prompts but you will be spending hours typing and just hoping to get something decent which may or may not work for you.
I could also take a mini chatpgt now and train it myself, feed it text from books, my 10 years of stored swim sessions off my google drive, but again how do we know any of the samples Im feeding it are worthwhile or not contradicting each other just confusing the situation.
1 Like
we are all exceptions…
1 Like
or just Wardie?
2 Likes
well, you get that with coaches now…
1 Like
I was at a gravel sportive a few weeks back and heard a guy say to a friend who just rolled up at the feed stop. “I was chatting to Wardie the other day…”. Made.me think of this chat but I know quite a few Wards and thought, probably just a mutual acquaintance/friend of theirs. But no, it was THE Wardie.
No idea how he is as coach but in my interactions with him, he has always been a nice guy. His podcast is a bit David Brent for me though.
2 Likes
yep…that’s a good comparison…
1 Like