.png){kind=logo}
Artificial intelligence has become deeply embedded in global finance, yet for most people it remains opaque—an invisible force shaping markets, investments, and economic outcomes without clear explanation. In December 2025, two figures working at the intersection of finance and technology attempted to address that gap from different directions at once.
Joseph Plazo and Mark Sullivan, co-founders of Plazo Sullivan Roche Capital, released a public whitepaper detailing the design of their artificial intelligence system for market analysis, Athena AI. In the same month, they published a book aimed not at specialists, but at general readers, explaining how generative AI systems—including GPT-style models—are created.
The timing was notable. By the end of 2025, artificial intelligence had moved from experimental novelty to everyday infrastructure, powering everything from online search to trading desks. Yet public understanding of how these systems function—and where their limits lie—has struggled to keep pace.
The Athena AI whitepaper outlines a system intended to analyse financial markets by combining multiple layers of data, rather than relying on single predictive signals. According to its authors, Athena is designed to function as a decision-support system, offering probabilistic assessments of market conditions instead of automated trading instructions.
The document adopts a technical but accessible tone, more reminiscent of academic research than commercial marketing. It describes how Athena integrates macroeconomic indicators, market structure data, liquidity patterns and behavioural inputs to form what the authors describe as “contextual market narratives”.
This approach reflects a broader shift in quantitative finance. Following periods of extreme volatility—most notably during the Covid-19 pandemic—many institutions have grown more cautious about over-reliance on narrowly trained models. Systems that can adapt to regime changes and provide interpretable outputs are increasingly favoured over opaque algorithms that perform well only under stable conditions.
Athena is currently in Beta 3, indicating it remains under active development. The whitepaper acknowledges unresolved challenges, including data bias, structural breaks in markets and the difficulty of modelling geopolitical risk—limitations that are often underplayed in commercial AI products.
Before the December 2025 release, Athena AI had already been referenced in media coverage examining new approaches to market analysis. Earlier reporting focused on its attempt to bridge traditionally separate domains—such as macroeconomic research and market microstructure—within a single analytical framework.
Plazo Sullivan Roche Capital operates more as a research entity than a consumer-facing technology firm. This positioning has allowed it to publish documentation and theoretical justifications that would be unusual in highly competitive trading environments, where secrecy is often seen as essential.
The company’s decision to release a detailed whitepaper reflects a belief that transparency and scrutiny are becoming increasingly important, particularly as regulators and institutional investors demand clearer explanations of how AI-driven decisions are made.
Alongside the whitepaper, Plazo and Sullivan published a book explaining how generative AI systems are built, explicitly targeting readers without technical backgrounds. Rather than focusing on how to use AI tools, the book concentrates on how they work—covering concepts such as neural networks, training data, reinforcement learning and alignment.
The authors rely heavily on analogy, drawing comparisons with legal reasoning, economic modelling and human decision-making. The goal, they argue, is not to turn readers into engineers, but to equip them with enough understanding to ask informed questions about systems that increasingly affect their lives.
The book’s release coincided with growing public debate about the influence of AI on employment, media, education and financial stability. While governments and institutions have issued guidelines and principles, critics have argued that meaningful accountability is impossible without broader public literacy.
The dual release of a technical whitepaper and a popular book highlights a tension at the heart of modern artificial intelligence. On one hand, AI systems are becoming more complex and specialised; on the other, their societal impact is expanding rapidly.
In finance, this tension is particularly acute. AI-driven models influence asset prices, credit decisions and risk management, with potential knock-on effects for pensions, savings and economic stability. Yet the mechanisms behind these systems are often understood by only a small group of specialists.
Plazo, who has previously spoken about the parallels between legal reasoning and probabilistic modelling, has argued that treating AI as infallible is a category error. Sullivan, with a background in quantitative systems, has emphasised the operational limits of machine learning, particularly in environments shaped by human behaviour.
Their shared position places them within a broader movement advocating for “human-in-the-loop” AI—systems designed to support, rather than replace, human judgment in high-stakes domains.
Whether Athena AI itself becomes widely adopted remains to be seen. Its future influence will depend on performance across different market conditions, institutional uptake and regulatory acceptance. As with many AI systems, its most significant contribution may lie as much in its design philosophy as in its technical outputs.
What is already clear is that the December 2025 publications form part of a wider effort to shift how artificial intelligence is discussed—away from mystique and towards explanation. In a field often characterised by exaggerated claims or guarded silence, that approach stands out.
As artificial intelligence continues to shape economic decision-making, the question of who understands these systems—and on what terms—will only grow more pressing. The work of Plazo and Sullivan suggests that, at least for some developers, the answer lies in opening the black box rather than sealing it tighter.
