AI autonomy: DeepCoder can pilfer code and write own programs
It’s the result of a collaboration between two of the most formidable hotbeds of innovation in the world, Microsoft and the University of Cambridge, so it makes sense that the outcome would be impressive. And impressive it is – DeepCoder is an AI tool designed to aid the building of programs for people without any knowledge of coding.
DeepCoder emulates the neural network of the human brain, allowing the system to think for itself instead of its actions being preordained by coding. And it signals a massive leap forward in program building; MIT’s Armando Solar-Lezama said to New Scientist, “The potential for automation that this kind of technology offers could really signify an enormous [reduction] in the amount of effort it takes to develop code.”
Computer programmers, however, needn’t start retraining as plumbers en masse; DeepCoder aims to automate the more mundane parts of programming, actually freeing up developers for more sophisticated and creative endeavours. In addition, the AI tool is only in its nascent stages, capable of solving program challenges comprising about five lines of code. This might not sound extensive, but five lines of code can go a long way in programming terms. Solar-Lezama went on to clarify: “Generating a really big piece of code in one shot is hard, and potentially unrealistic. But really big pieces of code are built by putting together lots of little pieces of code.”
DeepCoder is still a fledgling piece of technology, as its developers’ modest disclosure might imply. It’s awaiting testing by independent researchers, but looks set to make a formal debut at the International Conference on Learning Representations held in France next month. What’s more, the system will get exponentially more intelligent as it goes along, learning from its own mistakes and predicting in advance which parts of code will be useful for a given function. Damn. I, on the other hand, have been known to encounter difficulty decoding a bus timetable. Here’s hoping automation doesn’t encroach on the world of tech journalism, then…
Image: Michael Himbeault used under Creative Commons