A FEW years ago, author Neil Gaiman was given the task of making The Cybermen from Doctor Who scary again. They'd become fairly predictable in their actions and therefore in their weaknesses and so although the viewers wanted to see them, you could pretty much guarantee how it was going to play out.
What Gaiman did was introduce a random factor. The new Cybermen could write their own code to patch weaknesses and then distribute it to the entire army.
Now Microsoft and University of Cambridge scientists have created a machine learning system that can write its own code, by cherry picking bits of existing data. It's not a ridiculous idea
Think about that for a second. A machine than code itself. Erk.
The goal of DeepCoder is far more benign, and it is designed to allow non-coding types to be able to describe what they want done and have a computer do it.
A computer might be more nimble at finding options for the code that a human programmer wouldn't have thought of. Alternatively, you could argue that it's a jobs disaster waiting to happen.
The good news, however, is that at the moment, at least, the machine learning technology can only cope with five lines of code. The thing is, though, machine learning learns. And big chunks of code are just lots of little chunks of code anyway.
However, its creators have pointed out that the real advantages come from removing the dull no-brainer bits of coding, leaving the humans to be more creative, more agile and generally spend less time on the crappy stuff.
We can imagine a few ‘amens' there.
It's not the first time that this has been tried. New Scientist reports on an experiment in 2015 which saw a program that was capable of replacing bugs with working lines of code from elsewhere.
Meanwhile, the Cybermen are still made up. But this is how it starts. Jus' sayin'. µ
And, er, not much else
To serve, protect, and get incredibly hot and dusty
Symantec links attack to prolific Lazarus hacking group
Chinese firms drive global smartphone growth in first quarter