Ten Threes in a Row: AI’s Strength Must Become Human Strength
Illustration · not a service recordAfter 30 years of playing basketball, I made 10 three-pointers in a row for the first time. At the end of the video there is also a talking ball-retrieving robot. (My previous record was 8 in a row.)
The last 13 seconds of the video are imagined. The ball-retrieving robot that catches the ball, passes it back, and even speaks up to remind me is a 3D animation that AI made for me. But it is there because of a very real number: the original video of these 10 straight makes is 106 seconds long, and for nearly half of that time I am picking up the ball and walking back.
This also happens to be AI’s next step. The AI we have come to know over the past two years lives inside a screen: it writes, draws, and analyzes. Now the hottest term in the tech industry is “physical AI,” which lets AI grow eyes, hands and feet. Taiwan is keeping up too. The Ministry of Economic Affairs will hold the ROBOT READY forum on November 12, where 24 companies will jointly release guidelines for general-purpose robots. Some research firms also estimate that the global humanoid robot market will grow from about US$5 billion today to about US$50 billion by 2035.
When it comes to robots, people usually fall into two camps: one worries that “robots will replace people,” and the other hopes that “robots will get the work done for people.”
My court gave a third answer: AI serves as the eyes and the brain, the robot serves as the hands and feet, but what really has to change is my own body.
In one sentence: making good use of AI, and even of robots, has only one purpose, which is to make people stronger.
Why make a point of saying this? Because the most effortless way to use a tool is often also the way that sets you back the most: letting it do the work for you. The way that really helps is actually more tiring: let it tell you where you went wrong and save you time, and then make the changes yourself.
This time AI helped me “see”: of the 4 shots I missed, 3 were long. An extra 0.1 m/s of ball speed, roughly 1% more force, and the landing point moves 15 centimeters farther back.
If the robot really existed, it would help me “save time”: the time spent retrieving the ball would drop to 0, and in the same amount of time I could take nearly twice as many shots. Shooting machines that catch the ball, pass it back, and count the makes actually already exist. What my imagination adds is only the other half: the part that understands my movements and speaks. Every line the robot says in the animation comes from real data on the AI analysis page.
But AI also said that I still have not managed any of the three pieces of homework I set last month: I drift 1 meter forward when I land, I squat too deep, and I raise the ball too early. AI can see these three things and a robot can remind me of them, but only my legs can fix them.
You might say: with a robot to retrieve the ball and AI to do the calculations, won’t people just get lazier? That may well happen. But the next sentence is the real point: with the same machine, some people use it to take 500 more shots, and others use it to walk 50 fewer steps. The difference is not in the machine but in what you do with the time you save.
I drew up a job list for the ball-retrieving robot of the future.
What it will do: retrieve the ball, pass the ball, calculate angles, remind me about my landing, and remember every shot.
What it will not do: squat a little shallower for me, pull my landing back for me, or leave the house and go to the court for me.
Children’s learning is the same. The best use of an AI tutor, or even a robot teaching assistant in the future, is not to finish the work for the child, but to let the child practice a few more times and see where they are stuck. AI’s strength must become human strength.
A robot can pass the ball back to you. The one who takes the shot is always you.
If there really were a robot that could help you practice one thing more, what would you most want it to help you practice?
- Original: Ray Lu’s personal Facebook post (published 2026-11-01)
Course titles follow the source. Check each course page for its teaching language.
Introduction to Claude Cowork
Delegate multi-step work with workspaces, Skills, Plugins, and safety reviews. Free course; hands-on use needs paid Claude. The course retains the Cowork name as interfaces merge into Claude.
Applied AI Foundations
Free learning with a ChatGPT account. Turn recurring work into reusable steps with human review; AI Foundations is recommended first. Complete the course and pass its assessment at 80% for a badge.