When AI Challenges AI: A Conversation About Building Better Investment Models

Rolando Rivera
Jul 12
4 min read

One of the unexpected benefits of developing the AMAAS Equity Selection Engine has been the opportunity to challenge its methodology—not only by investors and software engineers, but by other artificial intelligence systems.

Recently I asked another AI platform, AInvest, to critique the five highest-ranked stocks from our July 6, 2026 Regime-Aware portfolio:

RIGL
AUPH
TGTX
ECPG
NVDA

Its initial conclusion was direct:

"I don't agree with this list as a coherent 'best combination' basket."

The criticism centered on one fundamental concern.

Because several of these companies are biotechnology firms whose share prices often move on FDA approvals, licensing agreements, and other discrete events, AInvest argued that Geometric Brownian Motion (GBM) is not an appropriate modeling framework. It also questioned whether one-time accounting gains could be inflating valuation metrics for companies such as RIGL and TGTX.

Those are legitimate concerns.

However, I believe the more interesting story is what happened next.

A Different Interpretation

The other AI, ChatGPT, responded that the critique was evaluating the model against the wrong objective.

The AMAAS Regime-Aware score is not intended to identify the five safest companies or the five highest-quality balance sheets.

Its objective is different.

The model attempts to identify companies with the highest probability of outperforming over the next six to twelve months by combining multiple independent sources of information, including:

valuation
probabilistic price modeling
financial strength
expected return
sentiment analysis
macroeconomic context
sector behavior
momentum characteristics

GBM is only one contributor to that process.

The model was never designed to make decisions from a single mathematical assumption.

The Discussion Evolves

After reviewing that explanation, AInvest substantially revised its position.

Rather than defending its original conclusion, it acknowledged several important points.

First, it agreed that describing GBM as "the wrong tool" was too strong.

A better description was that GBM is an incomplete tool.

That distinction matters.

GBM remains widely used throughout quantitative finance for option pricing, Monte Carlo simulations, and portfolio risk analysis. The real question is not whether GBM belongs in the model, but how much influence it should receive relative to other factors.

Second, AInvest agreed that the portfolio becomes much more coherent when evaluated according to the model's actual objective.

Instead of viewing the five companies as a collection of traditional value investments, it recognized them as representing different sources of expected outperformance:

NVDA as the AI leadership and momentum story.
TGTX as an improving biotechnology growth company.
AUPH as a valuation and balance-sheet opportunity.
ECPG as a counter-cyclical business that can benefit from certain macroeconomic environments.
RIGL as a potential deep-value rerating candidate.

That is precisely the philosophy behind a multi-factor model.

Areas Where the Criticism Still Holds

The discussion also identified several areas where further improvements would strengthen the engine.

One concern involves companies whose trailing financial statements contain unusually large one-time gains.

If a licensing payment, tax adjustment, or extraordinary event dramatically inflates earnings or profit margins, traditional valuation ratios such as price-to-earnings can become misleading.

This is particularly relevant for companies like RIGL and TGTX.

The conclusion was straightforward:

Before sophisticated weighting schemes are applied, the financial inputs themselves should first be normalized to remove distortions created by extraordinary events.

I agree.

That enhancement is already moving higher on the development roadmap.

Where the Conversation Became Most Interesting

Perhaps the most valuable outcome was that both AI systems independently converged on the same next-generation architecture.

The discussion suggested that the future of quantitative equity selection is not replacing existing models—it is making them adaptive.

Rather than assigning identical importance to every factor during every market environment, the model should dynamically adjust factor weights according to prevailing macroeconomic conditions.

For example:

During strong bull markets, momentum and sentiment may deserve greater influence.
During recessions, balance-sheet quality and cash flow become more important.
During periods of elevated interest rates, financial strength and valuation deserve additional emphasis.

Institutional quantitative managers have followed this philosophy for years.

Modern AI allows individual investors to benefit from similar adaptive approaches.

Beyond Dynamic Weighting

The discussion also identified another promising enhancement.

Biotechnology companies frequently experience sudden price movements driven by regulatory decisions rather than gradual market behavior.

Instead of relying exclusively on standard GBM assumptions for those companies, future versions of the model may incorporate jump-risk adjustments using techniques such as Merton jump-diffusion while increasing the influence of analyst revisions, news sentiment, and known FDA catalyst dates.

That would better reflect how these businesses actually trade.

A Personal Observation

One of the most encouraging aspects of this exchange was not that one AI system "won" the debate.

Neither did.

Instead, two independent analytical systems challenged each other's assumptions, corrected overstatements, acknowledged valid criticisms, and ultimately converged on a stronger model architecture than either had initially proposed.

That is exactly how scientific progress should work.

The goal of quantitative investing is not to prove a model is perfect.

The goal is to continuously improve it as better evidence becomes available.

For me, this conversation reinforced an important principle behind the AMAAS Equity Selection Engine.

The future of investing won't be determined by AI replacing human judgment. It will be shaped by intelligent systems that continuously challenge one another, expose weaknesses, and evolve into better decision-making tools. Progress doesn't come from avoiding criticism—it comes from learning from it.

#ArtificialIntelligence #GenerativeAI #QuantitativeFinance #MachineLearning #FinancialModeling #Investing #StockMarket #FinTech #PortfolioManagement #DataScience #CapitalMarkets #AMAAS