Written in Python, Part 5: What the Language Selected For
A tool built in six weeks by a student avoiding his thesis now underlies how a field decides what is true. Language ergonomics are a research policy, and nobody voted on this one.
A series on how one language shaped artificial intelligence, and how intelligence is now reshaping it.
259 Lines
In 2001 Fernando Pérez was a graduate student at the University of Colorado Boulder, working toward a doctorate in particle physics on a problem involving the quantum vacuum. He had recently learned Python and had been looking, without success, for an open-source interactive tool to analyze his simulation data. What he wanted was the thing he had grown used to in Mathematica and Maple: a prompt that remembered what you had done, that let you poke at a result and then poke at it again.
By his own description it was a thesis procrastination project. Over roughly six weeks he wrote 259 lines of Python. The script is still on GitHub. Its description says that it provides interactive execution with automatic history, and that it tries to mimic Mathematica’s prompt system.
That script became IPython, though not on its own. Pérez has described what followed as a merger: three independent projects, built separately by scientists who had each arrived at the same want, folded into one. This is worth noting because it is evidence about the demand rather than about him. Several people, not in contact, had independently decided that what scientific Python needed was a better way to sit and poke at things.
IPython became the notebook. The notebook became Jupyter, renamed when it grew kernels for languages other than Python and needed a name that did not privilege one of them. Today there are something on the order of ten million public notebooks on GitHub. Nature has listed the project among ten computer codes that transformed science. The first image of a black hole was produced through it. So was the analysis behind the first detection of gravitational waves.
The point of reciting this is not that small things grow. It is that the interface through which a large fraction of the world’s empirical work now passes was designed, in six weeks, by one person, to solve one person’s problem, with no theory of what scientific practice should look like and no expectation that anyone else would use it.
And it changed what scientific practice looks like.
