Why AI Still Can’t Beat the Market

The tech and investment industries recently witnessed an experiment that sounded like something from a science fiction movie up close. $10,000  was given to each of some of the most sophisticated AI’s in the world (Claude 4.5 Sonnet, DeepSeek V3.1 Chat, Gemini 2.5 Pro, GPT 5, Grok 4, Qwen 3 Max) to trade in the actual cryptocurrency market through the Alpha Arena which was hosted by nof1.ai¹

The showdown was thrilling: Could emotionally driven, impulsive humans ultimately be surpassed by cold, logical machines? 

The response was prompt, and it was not good for AI. The majority of these online traders were losing money within less than a week. The unexpected conclusion of the great AI dual was that machines couldn't consistently outperform a market driven by human unpredictability, no matter how sophisticated they grow. This result circles back to one of economics’ most frequently asked questions: Can anyone truly beat the market?

The answer is “almost never”, in keeping with Eugune Fama’s Efficient Market Hypothesis (EMH)², a pillar of current finance. According to Fama, market pricing already takes into account all information that is accessible to the public. This means that utilizing information that everyone already knows, it is almost impossible to achieve “alpha” – returns that regularly outperform the market — regardless of how quickly or thoroughly AI computes. The Alpha Arena experiment showed that even AI is no exception in the face of this huge theory.  But theory is not the whole story, because the markets aren't just numbers, they’re human. Behind every trading button is a mix of fear, greed and hope. AI can analyze the past 10 years' worth of data in a second, but it does not understand the 'horror' or 'joy' investors feel in a sudden crash. It is often these human emotions that determine the direction of the market. Ironically, those emotional surges often move markets more than logic ever could. 

This is where Behavioral Economics can help. Humans frequently make irrational decisions because we are over confident, loss-averse and easily influenced by noise, as pioneers like Daniel Kahneman have demonstrated. The twist is as follows: AI picks up knowledge from us. It is training on imperfect human conduct since it examines the past, which was influenced by our prejudges and errors. 

Even more concerning, in erratic markets like cryptocurrency, the “meaning” of information is always changing. A single celebrity tweet or meme has the power to instantly render months’ worth of analysis outdated. Therefore, even the best and most sophisticated AI models, equipped with flawless logic, falter in the face of illogical reality. Of course, this does not mean that "AI is useless." Far from it. AI is a remarkable tool; it can see patterns, analyze data and warn of impending danger more quickly than people could. However, it cannot take the place of human judgement. Ultimately, Alpha Arena brought attention to the market’s wild, unquantifiable spirit rather than revealing AI’s vulnerabilities. The experiment served as a reminder that the financial world is a living mirror of human unpredictability rather than a mathematical puzzle. 

So can AI beat the market?

Not yet. But the best part is, that's just what might be keeping the game interesting. 

[1] Alpha Arena | AI Trading Benchmark, nof1.ai/. 

[2]  Fama, E. F. (1970). Efficient Capital Markets: A Review of Theory and Empirical Work. The Journal of Finance, 25(2), 383–417. https://doi.org/10.2307/2325486