Martin Lueck, co-founder of systematic hedge fund Aspect Capital, has warned that handing investment decisions entirely to artificial intelligence could undermine transparency and risk management in quantitative investing, according to a report by the Financial Times.
Lueck, one of the early architects of systematic trading through his work at AHL, said he is reluctant to rely on models where the rationale for positioning is not clearly understood, arguing that investors should be able to explain the logic behind portfolio decisions.
Speaking in an interview, he said that maintaining a clear hypothesis remains central to his investment philosophy, adding that he would not be comfortable allocating capital to strategies where the drivers of returns are opaque.
His comments come amid a broader shift across the hedge fund industry, where firms are increasingly incorporating artificial intelligence and machine learning into trading and portfolio construction. Some leading quant investors have gone further in embracing model-driven decision-making even when the underlying signals are not fully interpretable.
That approach has been championed in part by Cliff Asness, founder of AQR Capital Management, who has previously described the industry as increasingly willing to “surrender” to machine-based strategies as firms seek to capture complex market patterns at scale.
Lueck acknowledged the growing appeal of AI tools, particularly in areas such as data processing, research support and testing investment ideas. However, he drew a distinction between using machine learning as an analytical aid and delegating core portfolio construction decisions entirely to algorithms.
He also pointed back to earlier stages of his career, noting that one of the motivations for leaving the then Man Group-owned AHL in the 1990s was a lack of transparency around model-driven trading decisions, which he described as effectively “black box” systems at the time.
While modern quantitative strategies have evolved significantly since then, with more sophisticated computing power and broader datasets, Lueck said the principle of understanding what drives returns remains essential, particularly during periods of market stress when model behaviour can become harder to interpret.
He added that large language models could play a useful role in supporting researchers, from organising data to running analysis and preparing outputs, but cautioned against using them to replace the intellectual process of forming and testing investment hypotheses.
