Google has trained its AI to solve computing challenges typically presented to human programmers. AlphaCode, the iteration of DeepMind to solve this category of problems, is moving at high speed. Should IT care about this?
Create 3D images from photos, blast professional players at Starcraft, or even detect breast cancer with 99% certainty: DeepMind can do it all, and generally better than a human. Google detailed the inevitably impressive results achieved by AlphaCode in terms of computer programming in the journal Science. If it does not yet rise to the level of the best programmers, AI knows how to solve the simplest challenges.
Read – Google has created an artificial intelligence capable of detecting breast cancer in 99% of cases
To achieve this result, AlphaCode consumed more than 700 GB of code from Github. Apart from the comments in the programs examined, no information was given to him about the algorithms or the programming structures to be used: it is thanks to machine learning that she was able to “understand” how she achieved the desired result.
AlphaCode Artificial Intelligence sees programming as translation work
The problem is stated in human language and the computer must translate it into a given programming language. The first step for AlphaCode is therefore to convert the description of the problem into a model that is understandable to it. The final phase consists of generate functional code from this internal representation “.
Read – Google: Here’s why the camera translation tool is getting a lot more efficient
AlphaCode does not always present an optimal code for a given problem, far from it. According to Science, more than 40% of proposed solutions exhaust system memory or take too long to produce a response within a reasonable amount of time. Since human programmers are also imperfect, AlphaCode ranks in the top half of the table in competitions. He has a programming level equivalent to that of a “beginner programmer with a few months of training”. It is generally agreed that for AlphaCode to one day reach expert programming level, an exponentially increasing amount of energy and resources will have to be allocated.