Aikido Intelligence

On my way home, I turned on the radio.  The topic of the day was AI -- Artificial Intelligence.  

I listened with enthusiasm with the goal of catching up with new technology.  I did not expect it to be a source of insight for me with Aikido teaching. . .

According to the scientist featured on the show, the development of AI started with language-based models.  It did not work because the technology operates solely on statistical prediction of texts, rather than having genuine understanding, reasoning or interaction with the real world.

Then, scientists tried to mimic the human thinking process by way of rational logic.  They put the process into codes, hoping that the computer can reproduce the thinking process.  It failed terribly.  Since it is not possible to prescribe all possible scenarios and write them all into a program, the computer system was very rigid and lacked intuitions.  There was no sense of creativity.

The newer generation of AI is mostly data-driven.  Their training is based on pattern matching and statistical probability rather than true, causal understanding of the world.  From there, the system generates a response based on the prediction of what might happen next.  As the AI Chatbot on my phone points out, "These systems lack contextual awareness, common sense, and the ability to generalize to novel situations, often failing when encountering ambiguity or data outside their training set. "

The radio host and the scientist tried an experiment on air, asking a chatbot to write an original joke.  The chatbot did come up with something that bears the typical structure of a joke -- a setup and a punch line.  As soon as she read it, the host laughed out loud -- not because the "joke" was funny, but because it was peculiar.  Nobody got the joke.  The scientist suggested asking the chatbot why the joke is funny.  The AI explanation was rather twisted.  It did not make much sense.  Obviously, the chatbot did not get human humor.

The radio show actually helps to explain the behavior of a certain kind of Aikido students . . . 

This kind of students want absolute certainty in an uncertain world.  They seek formulas that can lead them to do a perfect technique every time.  Therefore, as nage, they are extremely studious and methodical in their training, following strict steps to attain a perfect, replicable set of movements.  

When their perfect technique does not happen, they get confused and frustrated.  "I did everything the way I am supposed to, why is it not working?" they wonder.

As uke, they dutifully deliver a well-rehearsed attack.  Because they have memorized what they considered as the perfect responses for various techniques, they pull out corresponding ukemi for techniques like a programmed response.  You may use them as uke to perform a technique in multiple different ways, but they will reliably give you the exact same response over and over.  

At a lower level practice, this learning approach may work because the pace is slow and the practice is simpler.  These students may appear as if they were very proficient at their level.  As they move into higher ranks, the pace picks up, and the situations grow more complicated.  Their memorized ukemi is only as good as a broken clock.  Their inability to sense a situation and respond in an organic manner begin to take a toll on them.  

Whatever verbage is used during instruction or explanation is often taken by these literalists as absolute.  They get upset if you answer their questions by telling them," It depends."  "But you said . . . " is a common comeback from them.  They want their instructors to offer a magic recipe that works universally.  Unfortunately, the truth is that, in Aikido, we all make adjustments depending on situations.   There are no exceptions.

Because of the nature of how these students operate, they tend to have a hard time connecting with their partners and/ or staying connected.  They tend to be behind in their responses.  Since they need to pause and think before every movement, they cannot keep up with reality.  They get hit, they get squished.  They get startled and confused a lot, as a result.

With repetition, nonetheless, they eventually will accumulate enough experience for a data-driven approach.  Aikido is a complex activity that involves multiple variables.  Particpants have to, in real time, make adjustments to multiple variables simultaneously, while ranking the relative importance of the variables, so as to reach an optimal point instantaneously. 

The problem arise as these students tend to approach Aikido interactions as a single variable problem.  They lose sight that they are actually working in a multi-dimensional space.  As uke, one has to try to stay alive and continue to attack nage.  This kind of students often put themselves in a spot where they may live for the moment, but they cannot continue with their attack.  They also have the propensity to move to a spot where the partner can hit next.  

Building AI solely with data-based problem-solving is limited because AI lacks contextual understanding, struggles with novel situations, and depends on high-quality, non-biased data. Data-driven models (machine learning) correlate inputs to outputs without true comprehension, leading to failure in unpredictable scenarios, high susceptibility to poor data quality, and an inability to generalize like human intelligence. 

What do you do when humans behave like AI systems?