Enmusubi AI Development Story Episode 6 — The Reason We Didn’t Use Machine Learning (No external communication, completed with EA alone)
Last time I talked about making it an "explainable AI." This time, the foundation —Why I did not use so-called “machine learning AI”— In short, Enmusubi AI does not rely on external servers or DLLs,it is built as a chain of rules that completes entirely inside the chart with a single file (ex5)rather than trying to look smart. Instead of seeming clever,it is Unstoppable, Consistent, and Explanable. Those are the three aspects I chose.
When you hear AI, many people thinkMachine Learning— feeding vast amounts of past data to learn patterns —Being unable to explain why you judged something to be so
Enmusubi AI belongs to a different lineage. It is aReasoning engine (expert system) built on a stack of rules. It might be closer to writing out the decision steps as humans would when looking at the market. Because it is rule-based, the preceding explainability holds. The base is different.
Many so-called “smart AI” move bycommunicating with external servers. Convenient, yet if communication breaks, it stops; if delays occur, judgments shift; information may leak somewhere. It is also influenced by its operating environment. Enmusubi AI cut that dependency from the beginning.It completes inside the chart with just ex5, without WebRequest or DLL.
- If communications fail or servers go downit won’t stop
- Not bound by external services’ costs, terms, or discontinuationfree from constraints
- Runs in the customer’s account with the same behavior as my testsreproducibility
- No external communication or DLL, which meets listing-review criteriafrankly
Rather than being “clever but dependent and prone to stopping,” I chose to be “simple but standing on my own two feet.” It’s a long-term tool, so I prefer the latter.
The biggest risk of using machine learning in trading isoverfitting (curve fitting)— fitting too tightly to the past leads to sudden breakdown in the future.Backtests can look divine but fall apart in actual trading— I’ve seen this many times.
In my own tests, Gold’s “next direction” could hardly be predicted by any method (this will be covered in another episode).Trying to force predictions with more complex models is dangerous. Therefore Enmusubi AI is composed of a small number of meaningful rules. The fewer knobs (parameters), the less it pander to the past and the more honest it remains to unfamiliar markets.
*The diagram is an illustration of Enmusubi AI’s approach. The actual judgement items and behavior may differ.
If you don’t use machine learning, why call it AI? For me, AI is not about bragging about intelligence. It is about perceiving the state of the market and the account,“perception” and with rules“inference” and consistently“stop new buys / wait”and execute — thatautomation of decision-making (a system without human emotions)is what I call AI.
And that decision’s reasons are projected in Japanese at the “Enmusubi place.” Because the content is rule-based, I can show the reasons as-is. Rather than a smart black box, I prefer a straightforward and explainable system— that is what Enmusubi AI’s “AI” means.Summary
Not using machine learning isn’t due to lack of ability.Unstoppable, consistent, and explainableare prioritized. The design with no external communication and complete inside-EA completion reflects that philosophy. It’s not flashy, but it’s a reliable, quietly powerful tool that grows on you over time.
But to be clear: being straightforward does not guarantee profits. Enmusubi AI is designed not to use stop-loss, so if a full-scale yen appreciation continues, unrealized losses can expand,and in the worst case, bankruptcy is possiblefor this EA. Therefore, please monitor withrisk capitalonly.
※This article is intended for information provision and not for investment solicitation. The performance results shown are past results and do not guarantee future profits. FX/CFD trading involves risks. Please make investment decisions at your own responsibility.