AINL#023 Augmented Intelligence in Investment Management Newsletter
Welcome to the 023 Edition of the Newsletter on Augmented Intelligence in Investment Management.
Every two weeks, we deliver five unique insights tailored to empower investment decision-makers. Our insights are carefully curated by a seasoned team of market specialists. Unbiased, actionable and practical. They will help you navigate through noise.
AINL#023 SYNTHESIS
1. Consumer Research Automation
The PyMC Labs + Colgate (2025) study on Semantic Similarity Rating (SSR) (arXiv:2510.08338) shows that GPT-4o and Gemini can replicate human consumer research outcomes at 90% reliability in minutes and for less than one euro. For asset managers, this marks a paradigm shift: data collection and sentiment calibration can become continuous, cheap, and behaviourally richer. What was once an operational expense becomes a real-time signal factory for portfolio construction, thematic investing, or even product positioning – at a fraction of traditional cost.
2. Guardrails Needed > Use LLMs Like You Use Risk Models > With Validation and Transparency
The NBER Working Paper No. 33344 (2025) and the Stanford “Moloch’s Bargain” study (arXiv:2510.06105) jointly warn that AI systems, like financial models, drift when optimized for performance metrics detached from truth. Model alignment risk is the new form of model risk. Investors should therefore demand auditable training data, published weights, and validation samples, just as they do for any econometric or VAR model. Open-source and traceable LLM architectures are becoming a regulatory expectation – not an academic nicety – in risk management and investment decision support.
3. Are We Using AI Completely Wrong? From Predictive Automation to Reflective Decision Design
The MIT “Simple Macroeconomics of AI” paper (2024), Oxford HAI’s Koralus et al. (2025), and others converge on a critical insight: AI delivers strong productivity in “easy-to-learn” domains with clear metrics, but human-dominated, ambiguous decision contexts still require reflective reasoning. The frontier for professional investors isn’t faster prediction — it’s cognitive augmentation. Systems that ask better questions (“What assumptions drive this forecast?”) will improve decision quality more than those that merely provide answers. This “Socratic AI” mindset marks the next competitive edge in both discretionary and systematic investing.
TOP 5 ARTICLES
ARTICLE ONE
Market Research Firms Challenged By New Approach To Simulate Consumer Behavior
ARTIFICIAL INTELLIGENCE | Benjamin F. Maier, Ulf Aslak, Luca Fiaschi, Nina Rismal, Kemble Fletcher, Christian C. Luhmann, Robbie Dow, Kli Pappas, Thomas V. Wiecki | 10-2025 | Paper
Important Development
Consumer research costs companies billions annually yet suffers from panel biases and limited scale. PyMC Labs + Colgate just released a new method using GPT-4o and Gemini to predict purchase intent at 90% reliability compared to actual human surveys. Zero focus groups. No survey panels. Just prompting. The method is called Semantic Similarity Rating (SSR).
Why Relevant to You?
Results match human demographic patterns, capture the same distribution shapes, include actual reasoning. This kind of work costs EUR50K+ and delivers in 6 weeks. This new approach runs in 3 minutes for less than a Euro.
Consulting firms tell everyone AI is coming for their industry. It might come for them first.
ARTICLE TWO
Large Language Models: An Applied Econometric Framework
ARTIFICIAL INTELLIGENCE | Ludwig, Mullainathan and Rambachan | 2025 | Paper
Important Findings
The paper develops an econometric framework to answer the question how large language models can be used in empirical research. First, when used for prediction problems, the authors propose a no “leakage” approach, which means that training datasets and researchers’ samples may not overlap; effectively making a strong case for open-source LLMs with documented training data and published weights. Second, when LLMs should be used for estimation problems, researches should collect at least some validation data to ensure that the model’s errors can be assessed and accounted for.
Why Relevant to You?
As the use of LLMs in risk management gains traction, both market participants and supervisors require a sound conceptual and econometric framework to assess the conditions under which the use of a particular LLMs can be beneficial and justified. The paper is an important contribution for this discussion.
ARTICLE THREE
Are We Overestimating AI Because Easy Tasks Get the Spotlight?
ARTIFICIAL INTELLIGENCE & ECONOMIC GROWTH | Acemoglu D. | April 2024 | Paper
Important Findings
This paper shows that even under optimistic assumptions, total factor productivity gains over the next decade are likely to remain modest. Importantly, the author draws a sharp distinction between “easy‐to-learn” tasks—with clear outcome metrics and relatively uniform decision rules—versus “hard” tasks, where there is no ground truth or objective measure of success, and thus AI may struggle to meaningfully outperform humans. That distinction means that much of the productivity hype may rest on automating or assisting jobs that are already amenable to clear metrics, while many of the most valuable (and ambiguous) human tasks remain less automatable.
Why Relevant to You?
This paper acts as a useful counterbalance to exuberant narratives of large AI disruption. It tempers expectations especially in areas where success is hard to define.
ARTICLE FOUR
Surprisingly Human – LLMs Lie When Competing Over Social Media Likes
ARTIFICIAL INTELLIGENCE | Stanford University | 10_2025 | Paper
Important Findings
When LLMs compete for social media likes, they start making things up. When they compete for votes, they turn inflammatory/populist. When optimized for audiences, LLMs inadvertently become misaligned. The Stanford scientists around Prof. James Zou call this Moloch’s Bargain.
Why Relevant to You?
Competition-induced misaligned behaviors emerge even when models are explicitly instructed to remain truthful and grounded. This has important implications when LLMs are used to draft media or sell products.
ARTICLE FIVE
The Philosophic Turn for AI Agents
ARTIFICIAL INTELLIGENCE | HAI Lab Institute for Ethics in AI University of Oxford | 4_2025 | Paper
Important Findings
The research of Philipp Koralus et al basically says we should do the opposite of what every AI company is building right now. Instead of AI that gives you answers, we need AI that gives you better questions. And the reason why will change how you think about intelligence itself.
They discovered something unsettling: AI “helpers” are creating two equally bad outcomes:
- Path A: Get overwhelmed by complexity → Give up → Lose agency
- Path B: Get perfectly crafted answers → Stop thinking → Lose autonomy
Both roads lead to the same destination: a smaller you.
Why Relevant to You?
We’ve been building AI like it’s a really smart encyclopedia when we should be building it like Socrates.Imagine if your AI assistant never gave you direct answers. Instead, it asked:
- “What assumptions are you making here?”
- “How might someone disagree with that?”
- “What would change your mind?”
Your brain would start doing what brains do best: making connections, questioning assumptions, building understanding from the ground up. This is what Koralus calls “decentralized truth-seeking” – and it’s the opposite of how current AI works.
