AI have tendency to overfit and go broke. AI is only good for board games like chess where you can brute force search. Stock markets are way too complex and too large a market to be fearful of AI and length of time in the market can be years compared to few mins a move in a board game so there are many variables involved. I think humans will still beat AI traders at least for my lifetime.
I get it. If you are interested in applying AI technology to trading and it gives your life more meaning, then do it. I am still expanding my skill sets, but I prefer to focus on combining traditional TA indicators with regression and forecasting techniques to compensate for the lag effect of the traditional indicators to predict the reversals. Some traders don't use any indicators and just rely on price action/candlestick patterns and experience.
Since you have a mathematical result leaving minimal space for opinions would you care to share those numbers? Humbly, if you know that your approach using AI is (more) profitable, why would you even need to discuss it publicly on these boards with people who may not even know the first thing about AI? No, because discussions on EliteTrader is mostly a waste of time. There may be other venues where you could have a meaningful discussion, but discussions on ET generally turns into a flame war and name calling. My humble opinion on the subject is that AI could be a worthwhile tool, but that does not preclude a fundamental understanding of trading and trading mechanics. If it did - every AI geek and programmer would be rich through trading and that's clearly not the case. I worked closely together with a guy who claimed to be proficient in ML, but in terms of results it didn't amount to much and he clearly had little to no knowledge on trading. I know Jim Simons said that the Medallion Fund used machine learning for their trading models. I think he spent about 10 years to get fully operational and that was with a full team of top notch scientists.
Have you done any analysis of the average performance of funds at Renaissance which weren't the Medallion fund? How can just one fund outperform so much? They are closed to outside investors so it doesn't really matter anymore but I have reason to believe that ML was a marketing tool to attract investors and that the outsized returns weren't due to any special machine learning or mathematical algoritms. That is all I am going to say about it. Do your own research.
"..and/also" keeps opportunities possible whereas "either/or" prunes I, for one, enjoy the development of Ai and it's past/present/future roles in all areas of human endeavor and specifically trading. Just screen your responses, those that don't see any benefit are most likely expressing their opinion via limited experience and some are simply uninformed. Just look up any job description for any DMMs, SLPs and forward looking firms, they all are on the lookout for those with training in Ai/ML/LLMs.
Don't know much about AI, but I'm an amateur mathematician. I do a lot of algo design and data transformation/normalization/indexation/differentiation and integration, etc. If you know what data(s) to feed into well designed algorithms, you can get ahead of the game. Examples are like creating a synthetic instrument and using it to get edge on it's related tradable instrument(s). Stuff like trading the vol or rates markets using indexed returns data and analysis algorithms or trading based on the third-party execution market (price relative to VWAP), using market-maker spreads, etc. In machine learning, they use loss functions (AKA error functions) -- which are important in statistics. For example, when you design auto-pilot or self-correcting systems, the inputs are managed using this type of function. An error function like the mean squared error measures how well a linearization of a data series is tracking the data. So, a rolling MSE loss function is like a trading robot in the sense that it is measuring how well the price is being predicted by parsimonious estimates of its trend. The stuff is useful, not because it's related to ML, but because it's of fundamental importance in statistics. Error functions and the analysis of errors (referred to as residuals and residual analysis in the literature) are critical for model validity. This is of fundamental importance, since the math theories that govern the accuracy and validity of models (like BLUE/Gauss-Markov for linear regression) and so forth are intimately involved in what ML/AI folks are calling loss functions or learning functions, etc.