Behind the “thoughts” of AI

Behind the “thoughts” of AI



AI


They begin to study it as a living organism.


Today AI systems operate in hospitals, companies, schools, even in religious environments, they suggest diagnoses, interpret documents, support important decisions and as that power grows, a question begins to become increasingly impossible to ignore, we really understand how it “thinks.” And this is where the story changes tone, because in the face of this uncertainty, researchers began to abandon the idea that it is just applied mathematics and began to treat these models as if they were something more similar to a living system, a phenomenon that needs to be observed, tested and understood from within.


This movement has a name that sounds technical, but the idea is simple, Mechanistic Interpretability, instead of just looking at the final answer that AI delivers, scientists try to trace the internal path that led to that answer, almost as if they were doing an MRI and an artificial venous.


Reducing the problem.


Enhropic researchers, for example, developed tools to follow what happens inside the model while they reason, the logic is not what responded but how they got there and the most curious thing is that some of these scientists say that this type of study is more like biology than engineering, because instead of a predictable machine, you start to see a system with internal patterns, decision circuits, emergent behaviors and sometimes even strange routes that were not in the script.


Another strategy along these lines is to create simplified versions of these as if they were controlled miniatures, a comparison that makes sense here is with organoids, those small versions of human organs that scientists use to study the body on a smaller scale. The idea is if the giant models are too big to understand, then let's create smaller, more transparent models and observe how the internal structures form.


But understanding intelligence is only the first step, because when that same AI leaves the data center and enters the factory, the discussion stops being theoretical and becomes physical.


References 1


Follow my publications with the latest in artificial intelligence, robotics and technology.

If you like to read about science, health and how to improve your life with science, I invite you to go to the previous publications.


Posted Using INLEO



0
0
0.000
0 comments