The pursuit of a true science of deep learning is not a luxury; it is the only path forward that will save the field from its own momentum. For seven years, this scientist has been working toward exactly that, and their dual foot in industry and academia gives them a rare vantage point. They see what many practitioners refuse to acknowledge: that our current models are powerful but poorly understood, and that power without understanding is a fragile foundation for the next decade of progress. We agree, and we think their insistence on honest foundations is the most important stance anyone in this space can take.
What does this mean for you, the person building with these tools every day? It means that the next time a model fails in a baffling way, the answer is not simply to throw more data or compute at it. It means that the field's current reliance on empirical tweaks and benchmark-chasing is a temporary state, not a permanent solution. The scientist's point is that we need a framework that explains why a network generalizes, why it overfits, and why it sometimes behaves in ways that seem almost intuitive. Without that framework, you are left with a tool that works by accident, and that is a risky way to build anything that matters.
This is not an abstract, academic exercise. It is about the difference between engineering and alchemy. When you rely on trial and error, you might stumble onto something that works, but you cannot repeat it with confidence or teach it to others. A true science of deep learning would give you the vocabulary to describe what is happening under the hood, the ability to predict when a model will fail, and the confidence to push it into new territory without fear of breaking it. The scientist's seven years of work is a testament to the difficulty of that goal, but also to its necessity. They are not asking for perfection; they are asking for honesty about what we know and what we do not.
The practical takeaway is straightforward: start demanding more from the research community, and start holding your own work to the same standard. Ask for explanations that go beyond "it works because the loss went down." Push back against hype that promises more than it can deliver. The path to a true science of deep learning is built on foundations of rigor and humility, and every one of us has a role to play in laying those bricks. The alternative is to keep building on sand, and we have already seen how that ends.