"The Genesis of AI" (5/18) Investment in AI development struggles

We AI developers evaluate our creations much like God assessed his creations in Genesis, by giving life to AI and judging whether it is good or bad.

However, entrusting the same AI with measuring what is “good” or “bad” can be difficult.

For example, there are mathematical metrics like F1 score and ROC curves, but they alone cannot fully reflect the developers’ purposes and values — how you want to use it or under what conditions you want it to operate.

In a textbook-like learning method, you may achieve high accuracy by training on scenes where the price moves significantly after a single candlestick, but for day-trading AI, the entry points are too few to be practical. If AI evaluates each other, this model would be adopted simply because it is highly accurate.

In the end, adoption or rejection requires an evaluation axis defined by the developers themselves about what to prioritize, and that is where individuality and purpose show strongly.

In short, the process of AI learning and evaluation cannot be performed by AI with the same cognitive framework.

Only when there is a higher-order being, humans, can a better AI be created.

When developing AI, don’t insist on evaluating AI based on mechanical indices or the like,

but instead look at the actual output values with your own eyes and verify what they indicate to assess the AI.