For years, we've treated artificial intelligence like a magic trick. We feed data into a 'black box' of complex libraries and frameworks, and out pops a result. But as AI integrates deeper into our lives, a growing movement of educators and engineers is asking a dangerous question: Do we actually know how this works, or are we just trusting the code?
Back to the Basics
Enter "AI by Hand," a counter-trend spearheaded by educators like Professor Tom Yeh. Instead of relying on high-level abstractions, this approach strips AI down to its bare essentials. We aren't talking about writing thousands of lines of C++; we're talking about calculating neural network weights using a pencil, paper, or a simple Excel spreadsheet. By manually walking through matrix multiplications (Matmul) and deep learning calculations, the 'magic' evaporates, replaced by a concrete understanding of the underlying math.
Why Manual Math Matters
It might seem counterintuitive to do by hand what a GPU can do in milliseconds, but the goal isn't efficiency—it's explainability. When you manually map out three inputs or trace a transformer's logic, you stop guessing why a model is hallucinating and start understanding the mathematical friction causing the error. This shift is gaining traction even at the highest levels of industry, with frameworks like AI by Hand appearing in talks at Google to make complex deep learning more accessible to everyone.
The Future of Literacy
As we move toward more autonomous systems, 'AI literacy' can no longer mean just knowing how to write a prompt. It must mean understanding the flow of data. By championing the manual implementation of AI, we are moving from being mere users of technology to becoming its architects.
Sources
Media
