Alex Graves

Before you could speak to your phone and have it understand you on the first try, someone had to solve a very hard mathematical problem: how do you train a neural network to recognize speech when you don't know exactly when each sound starts and ends? That someone was Alex Graves, a Scottish theoretical physicist who ended up doing a PhD in artificial intelligence and, in 2006, published a paper that changed the field forever.

His solution is called CTC, Connectionist Temporal Classification, presented at ICML 2006 alongside Jürgen Schmidhuber at IDSIA in Lugano. The idea is elegant: instead of forcing each audio frame to correspond to an exact symbol, the algorithm learns to align the sequence on its own. In 2009, an LSTM network trained with CTC became the first recurrent neural system to win international handwriting recognition competitions. Today, Google uses CTC in the voice recognition of millions of smartphones. Every time you say "Hey Google" or dictate a message on your phone, a version of Graves' idea is working in the background.

But Graves didn't stop there. In 2014, alongside colleagues Greg Wayne and Ivo Danihelka at Google DeepMind, he published "Neural Turing Machines": an architecture that combines the power of neural networks with a differentiable external memory, an attempt to give AI the ability to reason algorithmically. The paper was foundational for an entire generation of researchers working on memory and reasoning. Then came the Differentiable Neural Computer (2016), another step in that direction. Graves completed his postdoctoral work under Geoffrey Hinton at the University of Toronto, and spent years as a researcher at Google DeepMind in London, before joining InstaDeep in 2025 to continue exploring generative models.

What I find compelling about Alex Graves' story is that he started studying physics and ended up giving machines their voice. His CTC is one of those inventions most people don't know by name but use every single day. The next time you dictate a message to your virtual assistant and it gets every word right, you now know who deserves part of the credit.