Why can't one AI keep winning? The reason 19,136 varieties are needed

In a previous article, I wrote the breakdown of 19,136 kinds (4 price range types × 4 time frames × 299 auxiliary indicators × 4-hour bars). Today is about “why such a number is needed.” The answer is that letting a single AI battle all situations is impossible.

A single AI can have only one personality. A model that aims for high win-rate with small price movements will miss large trends. A model that extends trends will be whittled away in ranges. A model strong during Tokyo market movements will falter when London market movements predominate. Any personality is a weapon in favorable markets and a weakness in unfavorable ones. In other words, trying to build one AI that wins in every market is equivalent to erasing its personality, resulting in a bland, mediocre one for any market.

Therefore, we invert the approach. Prepare 19,136 AIs with different personalities, and select the personality that fits the current market via testing. The number 19,136 is not meant to be used all at once (as I noted in a past article that cycling through everything is meaningless); it’s to ensure that whatever market comes, there is a model in stock that fits that market.

As the market changes, the chosen personality also changes. Even if today’s correct choice is a high-wins model, next month’s correct choice may be a different type. Rather than falling in love with one AI, we keep reselecting from inventory. In a world where there is no single perpetually winning AI, this is, to my knowledge, the most sensible strategy.