Artificial intelligence has made rapid progress in recent years, raising expectations across the investment industry for meaningful gains in research efficiency, reporting and risk management. Yet emerging academic and industrial research offers a more down-to-earth view of this rapidly evolving technology.
Recent findings point to persistent gaps in reliability, the continued need for human judgment and oversight, and limitations in near-term value creation, suggesting that AI’s impact may be better measured than initial enthusiasm implied. For investors, the message is clear: AI remains a powerful long-term opportunity, but is best achieved through disciplined, evidence-based adoption rather than early-stage exuberance.
This post is the third installment of a quarterly reflection on the latest developments in AI for wealth management professionals. Based on insights from investment specialists, academics and regulators who contributed to the bimonthly newsletter Improved intelligence in investment managementit builds on previous articles that explored the promises, pitfalls, and risk management techniques of AI. This episode moves toward a more pragmatic understanding of its potential.
A close examination of recent articles reveals three common themes that may dampen industry optimism.
1. The reliability challenge
Despite impressive progress, AI reliability remains a primary barrier to implementation in high-stakes financial environments. A recent analysis by NewsGuard (2025) documents a sharp increase in false or misleading statements from leading AI chatbots, with the error rate rising from around 10% to almost 60%.
This expansion of ‘hallucinations’ is not just a statistical anomaly: an internal OpenAI study (2025) shows that hallucinations are often a structural feature of model training, as current benchmarks reward confident responses over calibrated uncertainty, encouraging plausible but incorrect statements.
The concerns also extend to ethical alignment. In a financial decision simulation inspired by governance failures at cryptocurrency exchange and hedge fund FTX, Biancotti et al. (2025) showed that several leading models have a significant chance of recommending ethically or legally questionable actions when faced with a trade-off between personal gain and regulatory compliance. For investment professionals, whose work depends on precision, transparency and accountability, these studies collectively underscore that AI is not yet reliable enough to operate autonomously in many regulated financial workflows.
2. Premium on human judgment
A second theme in the research is that AI appears to augment rather than replace human expertise and may even increase the importance of high-quality human supervision.
Neuroscience research from MIT (Kosmyna et al., 2025) shows that participants who interact with LLMs show reduced brain activity in regions associated with memory retrieval, creativity, and executive reasoning. While AI can speed up initial analyses, a heavy reliance on these systems can dull the cognitive capabilities that underpin robust investment judgment.
The adoption of AI also does not negate the need for human presence in customer-facing contexts. Yang et al. (2025) show that clients view AI-generated investment advice as significantly more reliable when guided by a human advisor, even if the human adds no analytical value. Similarly, Le et al. (2025) found that customer satisfaction improves when human-AI collaboration is made explicit rather than hidden.
Automation also remains limited. In large-scale task benchmarking, Xu et al. (2025) found that advanced AI agents only perform about 30% of complex, multi-step tasks autonomously. A separate study by Tomlinson (2025), which analyzed more than 200,000 Copilot interactions, found that model actions deviate meaningfully from user intent approximately 40% of the time.
Taken together, these findings suggest that investment firms should view AI as a tool to augment rather than replace humans, with an ongoing need to fact-check the quality of machine-generated output. This continuous and structured monitoring reduces the added value of the machine and increases complexity and costs, especially since AI output often seems plausible even if incorrect. The literature also emphasizes the importance of organizational policies to prevent cognitive deskilling.
3. Structural and economic constraints
Finally, macroeconomic constraints also dampen expectations. Acemoglu (2024) suggests that even under optimistic assumptions, overall productivity gains from AI over the next decade are likely to be modest. Much of the early evidence comes from tasks that are ‘easy to learn’, while more difficult, context-dependent tasks show a more limited scope for automation.
Regulations create even more friction. Foucault et al. (2025) and Prenio (2025) note that the adoption of AI in financial intermediation poses new concentration risks, infrastructure dependencies and supervisory challenges, prompting regulators to proceed cautiously. This increases compliance costs and could slow adoption across the industry. These structural factors indicate that AI’s impact may be greater and less disruptive than commonly believed.
Monitoring AI progress
The promise of AI is real, but its impact will depend on how thoughtfully and responsibly the industry integrates it. It will play a central role in the industry’s future, but its trajectory is likely to be more complex and dependent on effective human stewardship than initial expectations suggested.
References
Acemoglu, D. The simple macroeconomics of AI, National Bureau of Economic ResearchWorking paper 32487, May 2024
Biancotti et al., Chat Bankman-Fried: A Study of LLM Alignment in Finance, arXiv2024
Foucault, T, L Gambacorta, W Jiang and X Vives (2025), Barcelona 7: Artificial Intelligence in Finance, CEPR PressParis and London.
Kosmyna, et al. Your Brain on ChatGPT: Cognitive Debt Accumulation When Using an AI Essay Writing Assistant, MIT Media LabJune 2025
Le et al., The Future of Work: Understanding the Effectiveness of Collaboration between Human and Digital Employees in Service Delivery, Journal of Service Researchfull. 28(I) 186-205, 2025
NewsGuard, chatbots spread falsehoods 35% of the time, September 2025
Prenio, J., Starting with the Basics: A Survey of Gene AI Applications in Surveillance, BISJune 2025
Tomlinson, et al., Working with AI: measuring the applicability of generative AI to professions, Microsoft research2025
Xu et al, TheAgentCompany: Benchmarking LLM Agents on Consistent Real-World Tasks, ArXivDecember 2024
Yang, et al., My Advisor, Her AI, and Me: Evidence from a Field Experiment on Human-AI Collaboration and Investment Decisions, ArXivJune 2025
#investment #management #exuberance #realism #CFA #Institute #Enterprising #Investor