Beyond the AI Hype: How Trading Firms Actually Use Artificial Intelligence

According to Bloomberg Business, Hudson River Trading’s head of AI research Iain Dunning recently revealed how the major US market maker is deploying artificial intelligence in practical trading applications. The firm is using AI not just for efficiency improvements but specifically for making short-term price predictions that give traders an edge. Dunning discussed the key constraints facing AI implementation in trading, including labor shortages, power requirements, and chip availability. He also explained how his work at the trading firm both differs from and resembles the approaches taken by major AI research labs like those behind ChatGPT. This insight into real-world AI deployment in finance reveals both the potential and limitations of current technology.

The Evolution from Traditional Quant Methods

What makes current AI approaches fundamentally different from the quantitative analysis that trading firms have used for decades? Traditional quantitative methods typically rely on statistical models with clearly defined parameters and relationships. These models work well when market conditions remain within historical norms, but they struggle with regime changes or truly novel scenarios. Modern artificial intelligence, particularly deep learning approaches, can identify complex, non-linear patterns that traditional models might miss. The key distinction lies in AI’s ability to learn representations directly from data rather than relying on human-engineered features.

The Real Constraints on Financial AI

Dunning’s mention of labor, power, and chip constraints reveals the practical realities that separate AI hype from implementation. The talent shortage for AI researchers who also understand financial markets creates intense competition and compensation pressures. Power requirements for training large models have become a significant operational cost that many firms underestimate. Most critically, the competition for advanced chips creates a bottleneck that could determine which firms succeed with AI trading strategies. These constraints mean that successful AI implementation requires not just technical expertise but substantial capital investment and infrastructure planning.

Why Short-Term Prediction Matters

The focus on short-term price prediction represents a strategic shift from traditional quant approaches. While long-term forecasting remains notoriously difficult due to countless unpredictable variables, short-term price movements often contain patterns that AI can detect. These might include micro-structural market inefficiencies, liquidity patterns, or reaction dynamics that human traders or traditional algorithms miss. For a market maker like Hudson River Trading, even marginal improvements in predicting price movements over seconds or minutes can translate into significant competitive advantages in high-volume trading.

How Trading AI Differs from ChatGPT

The comparison to ChatGPT highlights a crucial misunderstanding about AI in finance. General language models excel at pattern recognition in text but lack the specialized training and real-time processing capabilities needed for trading. Financial AI systems typically process structured numerical data rather than unstructured text, operate under extreme latency constraints, and must manage risk in ways that don’t apply to conversational AI. More importantly, trading AI requires explainable outputs—traders need to understand why a model made a particular prediction, whereas ChatGPT’s “reasoning” is often opaque.

The Critical Role of Risk Management

What the discussion doesn’t fully address is the risk management framework necessary for deploying AI in live trading. Unlike traditional algorithms with bounded behavior, AI models can exhibit unexpected behaviors when encountering novel market conditions. The machine learning systems used in trading must include robust monitoring, circuit breakers, and fallback mechanisms. The greatest risk isn’t that AI will make bad predictions—that happens regularly with any trading system—but that it might fail in correlated ways across multiple positions or during periods of market stress when human oversight is most needed.

How AI is Reshaping Trading Competition

The adoption of AI is creating a new divide in the trading industry between firms that can afford the necessary investment and those that cannot. We’re likely to see consolidation as smaller firms struggle to compete with the AI capabilities of larger players like Hudson River Trading. However, this doesn’t mean the end of human traders—rather, it shifts their role toward designing, monitoring, and interpreting AI systems. The most successful firms will be those that combine AI capabilities with deep market understanding and robust risk management frameworks.