What Good are Forms?

[Kung Fu Wang]

I don't actually think it's possible to record all possible applications.

I will let the computer to do that task for me. For example, if I have 8 kicks and 6 punches, how many kick, punch combos that I can come up with? By using computer, I can also eliminate those combos that don't flow well.

It's 21th century. MA and computer technology should be integrated.

 

[skribs]

I will let the computer to do that task for me. For example, if I have 8 kicks and 6 punches, how many kick, punch combos that I can come up with? By using computer, I can also eliminate those combos that don't flow well.

It's 21th century. MA and computer technology should be integrated.

This is where concepts come in. Don't do every possible punch and kick combo. Do the principles of how they apply. If you take these basic punches:

• Straight punch
• Hook
• Uppercut
• Underpunch
• Body Hook

Teach how to generate power from lead hand, strong hand, and how to oscillate between them, you can create infinite possibilities with those 5 punches and 3 concepts. If I show you how to generate power with a lead straight punch (jab) and a strong underpunch, you should also be able to generate power with a lead underpunch. Also, when you pass 2 or 3 punches in the combination, it's really just several combinations chained together. For example, in a lead straight - strong straight - lead hook - strong uppercut, it's really just a lead straight - strong straight, strong straight - lead hook, lead hook - strong uppercut. Just overlap them enough and it becomes the longer combination.

That's how you oscillate a combination. There's also:

• Repeating (i.e. multiple lead straights)
• Flowing (i.e. hook - backfist)
• Spinning (i.e. hook - spinning elbow)
• Simultaneous (more often used with a block or parry)

There's not much difference between a hook-spinning elbow as there is for roundhouse kick - back kick. Except one uses your feet.

 

[Gerry Seymour]

I will let the computer to do that task for me. For example, if I have 8 kicks and 6 punches, how many kick, punch combos that I can come up with? By using computer, I can also eliminate those combos that don't flow well.

It's 21th century. MA and computer technology should be integrated.

How can a computer determine what applications exist?

 

[skribs]

How can a computer determine what applications exist?

It works on Indeed.

Ba-dum-tsh.

 

[Gerry Seymour]

It works on Indeed.

Ba-dum-tsh.

You are an awful person. Simply awful.

 

[skribs]

You are an awful person. Simply awful.

I don't think many people on this forum would argue with you on that.

 

[pdg]

Both.

1. A form that record 30 different principles - only one example of each.
2. 30 forms that records all different usage.

For example, a single leg form that records all different ways to obtain the single leg, and also all different ways to finish it.

1. uniform stance single leg.
2. mirror stance single leg.
3. uniform stance crisscross single leg.
4. diagonal shaking single leg.
5. diagonal shaking arm dragging single leg.
6. knee striking single leg.
7. shoulder striking strike single.
8. foot sweep single leg.
9. foot scoop single leg.
10. shin bite single leg.
11. inner hooking single leg.
12. outer hook single leg.
13. arm guiding single leg.
14. ….

Of course if I have 20 ways to get and 10 ways to finish, I should have 20 x 10 = 200 ways to get it and finish it. By using video and store in computer, it can be done. we live in the 21th century. we no longer live in the ancient time. This database will be expanded through the future generation.

One day when I finish these forms, I will just record it on video. I won't train those forms. I won't ask my students to train those forms. Those forms are used for "recorded information" only.​

I'm not sure I see an awful lot of use in this approach.

Showing a range of possible application, yes.

Trying to state that it's every possible application? Not so much.

It's dismissing the value or possibility of improvisation in a way.

What happens if you do all this, then you (or someone X number of generations forward) see someone do something different? Do you think "oh I missed one" or do you tell them that it's not part of the art or an incorrect application?

A generation or two down the line I foresee the latter becoming common.

Unless strict codification of the art is what you're aiming to achieve?

 

[TSDTexan]

How can a computer determine what applications exist?

A question i have been researching for close to 18 years. Which hasn't yielded results yet.

I suspect AI will help us find connections between patterns of attack and unknown or lost responses (applications).

for an example, when human1 is in this position doing this thing, and traditions (kata interpretation) say a lowline kick by human2, while guarding upper body is what goes here.

but the ai looks at this same set of positions and recommends a specific sweep by underhooking with the (former guarding arms) and twisting the hips while keeping the (former kicking) leg straight.

And if no one has seen or heard of that response before, then thats a new discovery or a recovery of something that was known at the time the form was created but, was lost, in a failure of transmission.

The challenge is developing a comprehensive coding language that works using motion capture data (populated initially, from 2man and solo forms but could eventually use HAPV data) that can utilize a neural network.

It has to account for 3dspace, time, gravity and human kinematics.

I suspect that if we can have the neural network create billions of fights and apply form based responses to different attacks, and sort them between effective and ineffective... then we will have a lot of bunkai that is either new or lost and rediscovered.

Self driving cars are a thing, and the AI behind that is a step in the right direction.

There was a thing i was watching on a new Chess AI Google's AlphaZero that taught itself how to play chess in 4 hours. It didnt have a database of historical moves. after teaching itself by playing against itself billions of times., it was faced off with the best chess AIs (like stockfish) in the field and beat them handily, yeilding 0 loses, but a high number of draws.

 

[dvcochran]

FWIW, I think adding the computer element will create a level of dependency that will have an adverse effect on learning a technique or MA. Yes, a computer can calculate and store a near infinite quantity of scenarios for a technique or form and it great for study or to brush up.
The average human brain is running near infinite calculations on, who knows how many, Different things/topics at the same time. I get that we have a multiplex parallel processor but it is more about recall when you really need a technique to work. Getting the Physical repetition to establish rock solid muscle memory is paramount. Remember I said Physical repetition. I don't care is my sonal momtom makki is not grammatically perfect as a block as long as it works for the given attack. I do care that my recall is correct and it was the right choice of block.
This is one of the greatest values of forms or partner drills or shadow boxing or however you do it. Techniques have to get engrained. Some people do it faster than others. Why? I wish I knew. I could make billions.

Historically speaking, I Love hearing how well versed some are in there MA history. It is truly a beautiful thing. I personally feel so much has changed in fight/war philosophy/method that some original method is irrelevant. A large component is that the opponents/attackers of today are smarter and more versed in battle/fight strategy today. So it is behovant for each of us to seek out what our MA learnings mean to ourselves and reconcile this with the educations of history and the understanding and opportunity of today.

 

[Gerry Seymour]

A question i have been researching for close to 18 years. Which hasn't yielded results yet.

I suspect AI will help us find connections between patterns of attack and unknown or lost responses (applications).

for an example, when human1 is in this position doing this thing, and traditions (kata interpretation) say a lowline kick by human2, while guarding upper body is what goes here.

but the ai looks at this same set of positions and recommends a specific sweep by underhooking with the (former guarding arms) and twisting the hips while keeping the (former kicking) leg straight.

And if no one has seen or heard of that response before, then thats a new discovery or a recovery of something that was known at the time the form was created but, was lost, in a failure of transmission.

The challenge is developing a comprehensive coding language that works using motion capture data (populated initially, from 2man and solo forms but could eventually use HAPV data) that can utilize a neural network.

It has to account for 3dspace, time, gravity and human kinematics.

I suspect that if we can have the neural network create billions of fights and apply form based responses to different attacks, and sort them between effective and ineffective... then we will have a lot of bunkai that is either new or lost and rediscovered.

Self driving cars are a thing, and the AI behind that is a step in the right direction.

There was a thing i was watching on a new Chess AI Google's AlphaZero that taught itself how to play chess in 4 hours. It didnt have a database of historical moves. after teaching itself by playing against itself billions of times., it was faced off with the best chess AIs (like stockfish) in the field and beat them handily, yeilding 0 loses, but a high number of draws.

Okay, that's driving from forms, and seems to require that the AI already knows what possible applications are for a technique (as opposed to a form movement).

 

[TSDTexan]

Okay, that's driving from forms, and seems to require that the AI already knows what possible applications are for a technique (as opposed to a form movement).

well it can infer from a database of actions. and figure out what can plug in, in a specific moment.
making an intuitive leap. perhaps even discover a new kick or strike or see power generation methods we have not used before.

The big issue is money and time resources to get the technology.

There is a very old saying that "there are no blocks in karate". Superficially, this is a lie. but that isn't the purpose of the statement.

The purpose is to teach you to reconsider everything that acts like a block in the forms. "is this a setup, an entry, a strike, a trapping action, a feint, a lever or fulcrum etc. ?"

And then plug in alternate actions in place of a "mere block" and test the results.

This become even more meta when strikes themselves are reexamined.

Kata is a many layered thing like an onion.
but AI doesn't have to be restricted to that paradigm alone.


The fact that the ai can run through the permutations far faster, and 24 hours/365 days without stopping to eat, sleep, poop, or shower... shows us a lot of potential.

 

[skribs]

FWIW, I think adding the computer element will create a level of dependency that will have an adverse effect on learning a technique or MA. Yes, a computer can calculate and store a near infinite quantity of scenarios for a technique or form and it great for study or to brush up.
The average human brain is running near infinite calculations on, who knows how many, Different things/topics at the same time. I get that we have a multiplex parallel processor but it is more about recall when you really need a technique to work. Getting the Physical repetition to establish rock solid muscle memory is paramount. Remember I said Physical repetition. I don't care is my sonal momtom makki is not grammatically perfect as a block as long as it works for the given attack. I do care that my recall is correct and it was the right choice of block.
This is one of the greatest values of forms or partner drills or shadow boxing or however you do it. Techniques have to get engrained. Some people do it faster than others. Why? I wish I knew. I could make billions.

Historically speaking, I Love hearing how well versed some are in there MA history. It is truly a beautiful thing. I personally feel so much has changed in fight/war philosophy/method that some original method is irrelevant. A large component is that the opponents/attackers of today are smarter and more versed in battle/fight strategy today. So it is behovant for each of us to seek out what our MA learnings mean to ourselves and reconcile this with the educations of history and the understanding and opportunity of today.

There are some sci-fi movies that have taken this AI approach to martial arts. Have you seen the movie Equilibrium? The idea was they had taken footage from thousands of gunfights and analyzed the typical trajectories, and created a gun-kata based on this information.

Of course, that was a movie about how everyone had their emotions forcibly taken away by the government, so maybe your point about it removing the soul from martial arts is a salient one!

 

[Gerry Seymour]

well it can infer from a database of actions. and figure out what can plug in, in a specific moment.
making an intuitive leap. perhaps even discover a new kick or strike or see power generation methods we have not used before.

The big issue is money and time resources to get the technology.

There is a very old saying that "there are no blocks in karate". Superficially, this is a lie. but that isn't the purpose of the statement.

The purpose is to teach you to reconsider everything that acts like a block in the forms. "is this a setup, an entry, a strike, a trapping action, a feint, a lever or fulcrum etc. ?"

And then plug in alternate actions in place of a "mere block" and test the results.

This become even more meta when strikes themselves are reexamined.

Kata is a many layered thing like an onion.
but AI doesn't have to be restricted to that paradigm alone.


The fact that the ai can run through the permutations far faster, and 24 hours/365 days without stopping to eat, sleep, poop, or shower... shows us a lot of potential.

You've gone back to the dissection of the forms (which I don't think was KFW's suggestion - he was talking about cataloging applications to a technique). I think we're far from the day when a computer (or network) can start with a basic technique and determine the applications. It needs to learn (like a human) the basic principles, then the variables that influence them, etc. Then it could start to identify the basic applications on its own. It will take much AI learning time to get beyond the basic applications, I think - and more processing power.

 

[TSDTexan]

You've gone back to the dissection of the forms (which I don't think was KFW's suggestion - he was talking about cataloging applications to a technique). I think we're far from the day when a computer (or network) can start with a basic technique and determine the applications. It needs to learn (like a human) the basic principles, then the variables that influence them, etc. Then it could start to identify the basic applications on its own. It will take much AI learning time to get beyond the basic applications, I think - and more processing power.

yes, yet i did say:

"Kata is a many layered thing like an onion.
but AI doesn't have to be restricted to that paradigm alone."

The deep learning neurologic of alphazero was based on rules (principles) not forms (programmable databases).

A bishop can only move diagonally on its assigned color. etc.

If you treat the body's joints like pieces, each with ranges of motion, and power generation chains. then it can learn how to move it.

if you assign attack values to different body stuctures, and damage values to others.

and let it fight itself, it will figure out how to attack and defend by fighting itself through millions of matches.

it doesnt need the input of kata to do that.
but my interest has been in leveraging computing power to decode kata, to discover forgotten or unknown 2man forms, and the related unseen applications.

 

[JowGaWolf]

If people can't figure a martial arts application then there is no way Ai would figure it out. Ai can't even drive down the street without killing people. You know who else can't do it? Humans.

There is more to applying techniques finding how many attacks are possible. Ask people wow try to get robots to walk and run. They'll tell you the she thing. There's a lot more to walking than what we realiaze.

 

[TSDTexan]

If people can't figure a martial arts application then there is no way Ai would figure it out. Ai can't even drive down the street without killing people. You know who else can't do it? Humans.

There is more to applying techniques finding how many attacks are possible. Ask people wow try to get robots to walk and run. They'll tell you the she thing. There's a lot more to walking than what we realiaze.

1st point. robotics is a seperate thing from what we are discussing. virtually moving bodies is done in 3d modeling all the time, for decades.
this isnt about piloting an artificial robot body in meatspace.


There are quite a few tesla cars driving on autopilot, not killing people, on the streets right now.
And its got amazing collision avoidance systems.

Amazon is dumping millions into self driving 18 wheelers, and the trucking industry is about to 180 from a driver shortage, to massive layoffs within the next 8-10 years.

 

[skribs]

yes, yet i did say:

"Kata is a many layered thing like an onion.
but AI doesn't have to be restricted to that paradigm alone."

The deep learning neurologic of alphazero was based on rules (principles) not forms (programmable databases).

A bishop can only move diagonally on its assigned color. etc.

If you treat the body's joints like pieces, each with ranges of motion, and power generation chains. then it can learn how to move it.

if you assign attack values to different body stuctures, and damage values to others.

and let it fight itself, it will figure out how to attack and defend by fighting itself through millions of matches.

it doesnt need the input of kata to do that.
but my interest has been in leveraging computing power to decode kata, to discover forgotten or unknown 2man forms, and the related unseen applications.

I prefer to think of a Kata like a parfait. Everyone loves a parfait. Ain't nobody say "no, I don't want a parfait, I don't like parfaits."

 

[TSDTexan]

I prefer to think of a Kata like a parfait. Everyone loves a parfait. Ain't nobody say "no, I don't want a parfait, I don't like parfaits."

unless their lactose intolerant.

 

[Kung Fu Wang]

Q: What's wrong with the following "group" MA training pictures?
A: There are training form.
Q: Should they only train form when partners are not available?
A: They need to learn the form so they can train when they don't have partner.
Q: Should they spent their training time in partner drill training while partners are available?
A: ...

What's your answer for the last question?

Here are examples of partner drill training for both striking art and throwing art.

 

[skribs]

Q: What's wrong with the following picture?
A: There are training form.
Q: Should they only train form when partners are not available?
A: They need to learn the form so they can train when they don't have partner.
Q: Should they spent their training time in partner drill training while partner are available?
A: ...

What's your answer for the last question?

Here are examples of partner drill training for both striking art and throwing art.

The image you're showing is foundational technique training. It's a drill done together to build unity in the class, and to help teach you what to practice when you get home. Training solo forms in class is so you can learn what to practice at home, and get feedback for when you are at home.

The image does not prove that they do NOT do partner drills. The image only proves that they do solo forms.

This type of post is the exact type of response that is starting to become a pet peeve of mine in martial arts discussions. To judge a curriculum on a single photo or video of a technical demonstration.