I started by looking at gold charts, and the truth is, many people think AI can read charts well. But when we actually tried it with gold trading, we found that there are still some things AI doesn't get quite right. We used to believe that AI could read gold charts and RSI-MACD signals better than humans, but when we put it to the real test, we discovered certain limitations that we need to understand before implementing it in practice.
Testing AI on Gold Chart Reading
We conducted actual tests by taking real chart images from terminals and having the model analyze them to read RSI-MACD indicator values displayed on the charts. We used the FLUX model on a Mac Mini M4 Pro that our team uses. For this live test, we used 3 real chart images and counted how many times AI correctly read the indicator values, along with the response time for each instance.
The test results showed that AI can read indicator values reasonably well, but not well enough to be satisfactory for actual gold trading, especially in highly volatile market conditions. Inaccurate RSI-MACD readings can significantly impact trading decisions.
Limitations in Image-Based Chart Reading
Although modern AI has improved in visual recognition, reading gold charts with multiple overlapping layers still presents challenges. In this test, we found that AI could clearly see the main price lines, but when it came to reading indicators like RSI in the lower section of the chart, the error rate was higher.
Additionally, reading numerical values on charts remains problematic because display systems differ across terminals, requiring AI to adapt and read values in new environments each time.
Comparing Results with Actual Indicators
After AI read the indicator values from the charts, we compared them with actual values from MetaTrader 5, which we use in our real testing. Across the 3 test images, AI correctly read values about 70% of the time. Notably, AI performed better at reading MACD values than RSI values.
In terms of response time, it took an average of about 1.2 seconds per query, which is fast enough for real-time applications. However, the less-than-perfect accuracy remains a significant obstacle.
Issues in Environmental Awareness
Another problem we encountered in testing is that AI still cannot fully recognize the environment around the chart. For example, reading indicators when other lines overlap or when there are additional texts on the chart can easily confuse AI.
Furthermore, reading very small indicator values on charts or when different scales are used also affects AI's reading accuracy. These issues limit the reliability of using AI for gold chart reading.
Future Development Directions
Although the test results weren't as many had hoped, we see positive trends for the future. Developing models specifically for trading charts, such as training with numerous gold chart images, could help AI read indicators more accurately.
Additionally, incorporating data from other sources like economic news or actual trading data might help AI better understand context, leading to improved gold trading decisions in the future.
Lessons and Next Steps
From this test, we've learned that while AI has some capability in reading gold charts, it's not yet ready for direct practical use. Currently, AI should be used as a decision-support tool in gold trading, not something to be relied upon 100% for actual trading.
For those interested in trying real gold trading, open an XM account at: https://clicks.pipaffiliates.com/c?c=72816&l=th&p=6
