I recently discovered the communication gap between what Ai sees and what we see. Originally I thought that Ai was getting confused between left and right when analyzing images. It turns out that it was a much bigger but simple issue to fix. I'm currently testing this in CoPilot but I'm sure this can be tested in with othe AI Platforms.
Pictured used in test 1 of AI Visual Analysis. Originally AI was analyzing this image from the human perspective of the viewer as if the body is facing you. For example, if the body is facing you, then the arm on our right side of our view is that person's left arm. The arm that is on the left of our view is that person's right arm. AI did not distinguish between a person's back is facing us and the person's front facing us. LAZY PROGRAMMING.
Here's the raw analysis that I get with the default AI image Analysis.
Why this information is important?
This fixes the problem of Ai not knowing the difference between left and right which screws up anything it has to say after. If this holds true then you can start building your training combinations into a formal library of techniques.
Here's my original AI instruction for CoPilot based on the Default image Analysis.
Using your default Image Analysis. What can the person in the black clothing do next?
From that default perspective, the possible next actions looked like:
Here's my original AI instruction for CoPilot based on the Subject perspective image Analysis.
You can clearly see how the analysis changes. When AI knows the difference between Right and Left. The analysis above is "straight out of the box" replies for CoPilot. I didn't need to make any special instructions for Ai to be able to understand what can be done. Naturally you would need to be able to teach AI your Martial Arts and your style of fighting before you can get to things.
Let me know if you would like me to help you get this fix with the Ai that you use.
Pictured used in test 1 of AI Visual Analysis. Originally AI was analyzing this image from the human perspective of the viewer as if the body is facing you. For example, if the body is facing you, then the arm on our right side of our view is that person's left arm. The arm that is on the left of our view is that person's right arm. AI did not distinguish between a person's back is facing us and the person's front facing us. LAZY PROGRAMMING.
Here's the raw analysis that I get with the default AI image Analysis.
Step 1: Viewer Orientation (default way that CoPilot and probably other AI sees things)
- I looked at the image as if I were standing behind the camera.
- From that angle, the arm extended toward the center of the photo appeared on the right side of the image.
- So I labeled it as the subject’s left arm forward (because I was mapping “image right” to “subject left”).
Step 1: Subject’s Orientation Analysis (I specifically asked AI to analyze the image from the subject's perspective)
Right arm forward, right foot forward His stance shows his right side leading. That’s a “orthodox” guard in martial arts terms.Why this information is important?
This fixes the problem of Ai not knowing the difference between left and right which screws up anything it has to say after. If this holds true then you can start building your training combinations into a formal library of techniques.
Here's my original AI instruction for CoPilot based on the Default image Analysis.
Using your default Image Analysis. What can the person in the black clothing do next?
From that default perspective, the possible next actions looked like:
- Advance with a strike
- Extend the arm further into a jab or punch.
- Step forward with the same side foot to add power.
- Retract after a strike
- Pull the arm back to guard after having thrown a punch.
- Reset stance to prepare for a counter.
- Hold a guard stance
- Keep the arm extended as a defensive barrier.
- Use the forward arm to block or parry incoming strikes.
Here's my original AI instruction for CoPilot based on the Subject perspective image Analysis.
Possible Next Actions
Offensive
- Right jab/lead punch: extend the already‑forward right arm to strike.
- Right front kick: use the forward right leg to attack low or mid‑level.
- Left cross: rotate hips and drive the rear left arm forward for power.
- Combination: jab → cross → right kick.
Defensive
- Guard/cover: keep the right arm extended as a barrier.
- Parry/block: use the right arm to deflect an incoming strike.
- Step back: retreat with the right foot to reset distance.
- Angle change: pivot on the lead foot to shift position.
Transitional
- Switch stance: pull the right foot back, bring the left forward.
- Feint: extend the right arm slightly to bait a reaction, then counter.
- Close distance: step in with the rear foot to pressure the opponent.
You can clearly see how the analysis changes. When AI knows the difference between Right and Left. The analysis above is "straight out of the box" replies for CoPilot. I didn't need to make any special instructions for Ai to be able to understand what can be done. Naturally you would need to be able to teach AI your Martial Arts and your style of fighting before you can get to things.
Let me know if you would like me to help you get this fix with the Ai that you use.