Why AlphaGo, not ChatGPT, will shape major AI breakthroughs in science and finance

In recent years, everyone talks about ChatGPT and language models. They help us write texts, answer questions, summarize documents… and seem almost magical.

But behind this magic lies an important nuance:

ChatGPT and similar models are very good at imitating past patterns, not necessarily discovering new things

If we want to understand where the deepest advances in artificial intelligence are, we need to look at another family of systems: those that originate from AlphaGo.

Here’s a link to the 2016 movie. If you’re even slightly interested in this topic, you’ll learn something and have an entertaining time.

AlphaGo: the system that didn’t just imitate, but learned by playing

In 2016, AlphaGo surprised the world by defeating the Go world champion, Lee Sedol.

How did it learn?

First, it imitated humans: it trained on thousands of professional players’ games.

Then, it played against itself millions of times:

• Tried new moves.
• Saw what worked best.
• Improved with experience.

That is, it didn’t just “copy” what it had seen, but experimented and found new strategies.

This is very different from a model like ChatGPT, which trains by reading texts and learning statistical word patterns.

From AlphaGo to AlphaFold: from a game to a Nobel

The same idea behind AlphaGo was later applied to biology with AlphaFold.

What is AlphaFold for?

To predict how proteins fold, which is key to understanding diseases and designing new drugs.

Its impact has been so great that this type of work was recognized with a Nobel Prize in Chemistry in 2024. Not because the model “reads articles,” but because it helps discover new insights into how life works at a molecular level.

The central idea:

• It doesn’t just repeat what it has seen.
• It learns deep rules of the system (in this case, physics and biology).

Here’s another link about the advances achieved by Demis Hassabis’ team:

The same is happening in other fields: finance, materials, genomes…

These “AlphaGo-style” systems are already used to:

• Discover new materials.
• Interpret the genome and better understand gene regulation.
• Make decisions in financial markets, learning how prices move and adjusting strategies.

For example, some investment funds use systems that:

• Simulate different market scenarios.
• Test strategies.
• Learn what works best over time.

They are not crystal balls—they don’t always get it right. But if they succeed, for example, in 60% of decisions while managing risk well, they can generate consistent returns.

The key: it’s not about predicting everything, but about being right more often than wrong.

In both biology and finance, these systems do not aim for perfection.
Their value lies in:

• Providing good directions (reasonable predictions).
• Repeating this process at scale.
• Learning and correcting continuously.

Also, as scientists remind us, model predictions don’t replace experiments:

• In biology, lab validation is still needed.
• In finance, risk must be controlled and real-world testing done.

AI does not replace human work, but it can accelerate it greatly.

And what about ChatGPT and language models?

Models like ChatGPT are very useful for:

• Writing texts.
• Translating.
• Summarizing information.
• Helping think faster.

But, according to Pawel, head of the AI Investments team at Omphalos Fund:

• They primarily imitate human language patterns.
• They don’t truly understand cause and effect.
• They cannot experiment in an environment and learn from the consequences of their actions, like AlphaGo.

In short, they are excellent conversationalists, but not necessarily the best discoverers.

Imitation vs. Innovation

Here is the main point:

Language models (ChatGPT, etc.)

• Train by reading large amounts of text.
• Are masters of imitation and writing.
• Don’t act in the world or learn from the outcomes of their decisions.

AlphaGo / AlphaZero / AlphaFold-style systems

• Learn by interacting: playing, testing, simulating, experimenting.
• Operate in environments with cause and effect.
• Can discover strategies no one ever taught them.

What type of AI do we need for the great scientific challenges?

If we think about the major challenges:

• Discover new medicines.
• Design solutions against climate change.
• Build safer and more efficient infrastructure.
• Make financial markets more stable.

It’s not enough to have machines that speak well.
We need systems that:

• Learn by doing.
• Explore the unknown.
• Can try, fail, improve, and try again.

Conclusion: the future is not who speaks most, but who learns best

AlphaGo wasn’t just “the machine that beat the Go champion.”
It was proof that:

• AI can surpass human intuition in complex tasks.
• We can build systems that discover rather than just repeat.

ChatGPT symbolizes the present well: AI that communicates, writes, and assists daily.

But if we look at the major future scientific and technological breakthroughs, the article argues that the model to follow is closer to AlphaGo than to ChatGPT.

In the end, in this race, it won’t be the AI that talks the most that wins, but the one that learns from the real world the most and best.

Important Note: This content was produced in collaboration with Omphalos Fund and is intended for educational purposes in AI/Machine Learning applications.

This is an edited version adapted for inbestMe followers of an article published in August 2025: https://www.omphalosfund.com/2025/08/11/why-alphago-not-chatgpt-will-shape-the-next-wave-of-human-progress/

Omphalos Fund leverages state-of-the-art AI technologies inspired by AlphaGo and recent advances in time-series prediction, portfolio optimization, and reinforcement learning. Omphalos Fund’s mission is to provide an unparalleled risk-return profile, following strict risk management protocols and ensuring resilience in volatile markets.

Backed by AI Investments and a top-tier team of experts, it is a systematic multi-strategy alternative fund operating 24/5 in global markets, seeking uncorrelated returns, low volatility, and a high Sharpe ratio.

The creation of Omphalos Fund is linked to a research project ”ALTO” involving inbestMe. It was a project to develop an advanced platform for portfolio optimization and risk control using AI and machine learning (inspired by AlphaZero approaches).

The AI Investments (AII) platform offers:

• More precise asset optimization and better risk management against unexpected events.
• Visualization tools to help investors make informed decisions.
• Usage by SMEs, asset managers, pension funds, large companies, and public bodies to monitor systemic risks.

Additionally, the project aims to train talent in cutting-edge AI techniques, promote research in the EU, and lower barriers for new Fintechs, increasing competition and financial innovation.

The consortium consists of two Fintech SMEs (AII and inbestMe) and two universities (Ulm and Oslo), leading the methodological and algorithm evaluation parts.

This content is for informational purposes only and does not constitute an offer, recommendation, or solicitation to purchase shares in Omphalos Fund.