I am fascinated by machine learning, perhaps because I don’t know much about it. I assume that my obvious ignorance concerning the subject allows my fascination to persist and thus perpetuates my metaphorical association of the concept to my life in the gym. In other words, I don’t plan on letting the truth get in the way of a good story for this week’s post.
As I understand it, machines learn distinctly faster than humans when given the proper guidance, information, and run time. The ability of an algorithm to focus on a single subject, say chess, allows that program to sift through infinitely more data at an astronomically faster rate than the human mind. Consequently, the machine learns every possible and probable combination of moves and can appear to play at a grandmaster level at a fraction of the time than its human opponent who has been playing his or her entire life.
In his book Life 3.0, Max Tegmark defines this type of artificial intelligence as Artificial Narrow Intelligence, ANI – an intelligence that can significantly outperform humans and mimic what appears to be human ingenuity, but only within the narrow scope as defined by its parameters. This would be machine learning.
The other two types of artificial intelligence accepted in this type of philosophical exploration include Artificial General Intelligence, AGI, aka Machine Intelligence; and the more sci-fi, sit-you-on-the-edge-of-your-chair version known as Artificial General Super Intelligence, or End of Days, Cyberdyne-style, humans now live in the the machines’ zoos, aka Machine Consciousness. Gulp.
I will be addressing the former because first, it has not been shown that #2 and #3 are even possible, and second, I have an innate problem with assuming that if we ever do devise a machine or some kind of software that we define as generally intelligent, why would we assume it will be evil. Maybe instead of Nietzsche, the computer decides to read a lot of Brenè Brown and teaches us world-wide vulnerability. It’s possible.
For my purposes, I want to muse a bit on the idea of narrow machine learning because I think that’s the metaphor that serves us better in the gym.
Too often, an athlete finds herself stuck in a rut. Perhaps it’s been a few weeks since she made it to class. Maybe she can’t seem to find herself a routine that works. Maybe she completes some kind of competition like the CrossFit Games Open or an in-gym challenge, and feels an overwhelming sense of frustration at all the THINGS she still needs to work on.
Either way, she starts to contemplate what she can change, add to her schedule, or improve upon and the list quickly grows out of control – double unders, barbell lifts, gymnastics skills, running, nutrition, better shoes, new weight belt…must.fix.all.the.THINGS!
If this was an algorithm or some kind of program that I was writing for my computer science class, my professor would tell me, “Whoa, junior nerd of the developing computer nerd herd, you are trying to do too much. Write your program to solve just one problem. Then move on to the next. Too much complexity will overwhelm the code, son.”
Quick little aside – I am not a computer science student, I’ve never written code, and I have never had a conversation with a computer science professor. I am sure you all are supremely cool, physically intimidating, and emotionally superior beings. If any of this post offends you or your computer overlords, I apologize and hope you will take mercy on me when the singularity occurs.
For our demoralized athlete, her internal code has a 505 error and instead of finding her groove her glitches continue.
We call it programming at the gym for a reason. The metaphor suggests that in each day’s workout, there exists a single stimulus upon which we can focus. However, when we get stuck in a loop where we think we must fix all the things immediately, we can quickly become overwhelmed.
So, in the words of the wise and slightly misguided Morpheus,
It sounds so simple, and perhaps it is, but the hardest part is learning to let go of all the extra crap that does not serve us. For many of us, the single, most important thing we can do is just show up. If we do that, the rest will work itself out.
- Don’t have double unders, that’s ok…make one step forward each time we do them.
- Can’t do a pull-up yet, that’s ok…make one move in the direction of getting a little bit higher on that bar the next time you see them in the workout.
- Not sure how to get under the bar in your snatch, that’s ok…land it just one inch deeper in the squat the next time we lift some barbells.
Patiently stay at this game long enough and like that machine learning chess, at some point in the future when you compare yourself to the “you” who started on this journey, the “you” now may be surprised at just how much learning you’ve accomplished.
Decide to flood the code, complicate the process, and hold on to the idea that you can fix all the things at the same time, well…
Simple ain’t easy folks, but if we are willing to keep that horizon low and the trajectory long, the possibilities are near infinite. It might just take a bit longer than Facebook took to learn what you like to buy on Amazon.
See you on the Creek.