Final Thoughts on Artificial Intelligence in Finance
So this is the last post. Eight posts later, we have walked through the entire first half of Artificial Intelligence in Finance by Yves Hilpisch. AI fundamentals, neural networks, superintelligence, financial theory. A lot of ground.
Time to step back and answer the question I started with: is this book worth your time?
What the Book Actually Gave Us
Let me recap what stuck with me across these chapters.
AI in finance sits in a sweet spot. The training environment is digital. The feedback is clear. You make money or you lose money. Compared to self-driving cars where mistakes kill people, finance is a safer playground for AI experimentation. That framing from the preface is still one of the best things in the book.
Neural networks don’t need you to guess. Traditional methods require you to know the shape of the relationship in advance. Neural networks figure it out from data. That’s a genuinely big deal for finance, where relationships are messy, nonlinear, and constantly shifting.
Data matters more than algorithms. The overfitting example hit hard. A neural network scored 97% accuracy on training data and 39% on new data. Worse than a coin flip. More data fixed it. This lesson applies everywhere, not just finance.
The old models are wrong but useful. CAPM, portfolio theory, arbitrage pricing. They all rely on assumptions that don’t hold in reality. Returns aren’t normally distributed. Relationships aren’t linear. But these models gave Wall Street a common language. You need to know them before you can move past them.
The superintelligence discussion was surprisingly grounded. I expected hand-waving. Instead, Hilpisch laid out a structured framework. ANI, AGI, SI. Five paths. The control problem. The paper clip scenario. He borrowed heavily from Bostrom, but the summary was solid.
What the Book Does Well
Hilpisch is good at making connections. He links AI concepts to financial applications in a way that feels natural, not forced. The comparison table from the preface (games, cars, finance) does more work than most chapters in other books.
The Python examples throughout the book are a real strength. He doesn’t just explain mean-variance portfolios in theory. He builds them in code. You can run it yourself. For anyone who learns by doing, that matters a lot.
He is also honest about limitations. He admits the book is opinionated. He acknowledges that smart people like Francois Chollet think market prediction might be impossible. He shows that a $6 billion ML hedge fund underperformed a basic index fund. That kind of honesty is rare in a field full of hype.
What Could Be Better
The book sometimes feels like two books glued together. The AI fundamentals section and the finance theory section don’t always talk to each other. You read about superintelligence in one chapter and expected utility theory in the next, and the connection between them is thin.
The math gets dense in places without enough payoff. The expected utility section, for example, spends a lot of time on formal derivations that could have been summarized more tightly. For a book that targets practitioners, some sections feel more like a textbook.
And while the first half of the book builds a strong foundation, it also means you are halfway through before seeing any actual AI applied to financial data. That is a pacing issue. Some readers will lose patience.
How the Book Has Aged
This is the interesting part. The book came out in 2020. A lot has happened since then.
GPT-3 launched the same year this book was published. Then GPT-4 arrived. Then open-source models caught up. Large language models went from a research curiosity to something billions of people use every day. The AI landscape in 2026 looks nothing like 2020.
But here is the thing. The fundamentals Hilpisch covers have held up well. Neural networks still work the same way. Backpropagation hasn’t changed. The overfitting problem is still real. Portfolio theory is still taught in every business school. The efficient markets hypothesis is still debated.
What has changed is the scale. Models are orders of magnitude larger. Training data is measured in trillions of tokens. The hardware improvements Hilpisch hinted at in his DeepMind section have accelerated beyond what most people expected.
The superintelligence chapter reads differently now too. In 2020, AGI felt decades away. In 2026, serious AI labs have it on their roadmaps as a near-term goal. The control problem Hilpisch described isn’t abstract philosophy anymore. It is an active engineering challenge with real funding behind it.
What the book misses entirely is the LLM revolution and its impact on finance. Modern AI in finance isn’t just about neural networks predicting stock prices. It is about language models reading earnings calls, processing regulatory filings, summarizing analyst reports, and generating trading signals from unstructured text. That entire dimension didn’t exist when Hilpisch was writing.
So the book’s foundation is solid. But anyone reading it today should know that the field has moved well past what is covered here.
Who Should Read This Book
You should pick this up if:
- You know some programming and want to understand how AI connects to finance
- You want a single book that covers both AI fundamentals and financial theory
- You learn better from code examples than from pure theory
- You want an honest take on whether AI can beat markets, without the sales pitch
Skip it if:
- You want a pure coding tutorial (there are better ones for that)
- You already know both AI and finance well (the book covers familiar ground)
- You want cutting-edge techniques (the field has moved on significantly since 2020)
Final Verdict
Artificial Intelligence in Finance is a solid foundation book. It connects two fields that often ignore each other and does it with real code, honest limitations, and a clear argument. The core thesis, that AI might break the efficient markets hypothesis, is still relevant and still unresolved.
It is not a perfect book. The pacing is uneven. The math sometimes serves the author more than the reader. And reading it in 2026 means accepting that the AI world has changed dramatically since publication.
But I’m glad I read it. The fundamentals are timeless. The questions it raises are still the right questions. And the reminder that a $6 billion hedge fund got beaten by a passive index fund is worth the price of admission alone.
Rating: 3.5 out of 5. Good for building foundations. Read it alongside something more current to get the full picture.
Book: Artificial Intelligence in Finance by Yves Hilpisch Publisher: O’Reilly, 2020 ISBN: 978-1-492-05543-3
All Posts in This Series
- Artificial Intelligence in Finance: A Book Worth Reading
- What AI in Finance Really Means
- AI Algorithms: Types of Data, Learning, and Problems
- Neural Networks and Why Data Matters
- AI Success Stories: From Atari to AlphaGo
- Superintelligence: Forms, Paths, and the Control Problem
- Uncertainty, Risk, and Expected Utility Theory
- Portfolio Theory, CAPM, and Arbitrage Pricing