AI Systems Seen as Forests of Knowledge Trees
-
A.I. systems like large language models are complex, but can be intuitively understood as "trees" that classify information. Multiple trees form a "forest" that can recognize more concepts.
-
Generative A.I. works by activating combinations of trees to conjure "phantom trees" representing new combinations of concepts prompted by users. It creatively fills gaps, but likely does not transcend what it has seen before.
-
A.I. creates value by leveraging similarities in large amounts of past human effort and data to enhance productivity. It maximizes the value of data.
-
Concerns about safety, quality, and bias must be addressed through guardrails on inputs/outputs and scrutiny of training data. More transparency is needed about how inputs lead to outputs.
-
A.I. should be demystified and explained intelligibly, not oversold as magic. Doing so promotes accountability and realistic understanding.