Artificial intelligence AlphaCode learned to write code no worse than the average programmer
Miscellaneous / / February 03, 2022
He could have done well in programming competitions.
British company DeepMind created the AlphaCode artificial intelligence system, which, in her words, "writes computer programs at a competitive level." In solving problems from the AlphaCode programming competition bypassed most human developers.
AlphaCode was tested on tasks created for the Codeforces competitive platform. New variants are published here every week, and for their solution, developers can earn points to climb to the top lines of the platform's internal rating.
Ten tasks were offered to AlphaCode in the same way as for ordinary Codeforces users: with textual description, requirements for incoming and outgoing information, examples of test cases and results. The artificial intelligence system generated a large number of possible answers, and then ran a test and selected the best options. About the same, according to the authors of the study, do the majority of human participants.
AlphaCode solutions ranked in the top 54.3% of the choices. This
allowed artificial intelligence to enter the list of 28% of the leading developers of the Codeforces Elo rating, which includes 1238 participants.The founder of the Codeforces platform, Mikhail Mirzayanov, noted:
I can confidently say that AlphaCode's results have exceeded my expectations. I was skeptical, because even in simple competitive problems it is often required not only to implement the algorithm, but also (and this is the most difficult) to invent it. AlphaCode managed to perform at the level of a promising new competitor.
Mikhail Mirzayanov
Founder of Codeforces, Assistant at the Faculty of Information Technology and Programming at ITMO University
DeepMind specialist Oriol Vinyals said that the study is still at an early stage. But its results have already brought the company closer to the creation of flexible artificial intelligence for autonomous creation of code blocks according to the described requirements.
In the long term, we are encouraged by [AlphaCode]'s potential to help programmers and non-programmers in writing code, improving performance, or creating new ways to create software.
Oriol Vinyals
Senior Fellow DeepMind
However, DeepMind acknowledged that AlphaCode skills will not always be useful in solving problems that programmers face every day. At competitions, participants are offered rather puzzles for which they want to find the fastest and most elegant solution. In real projects, this is rare, so it’s too early to talk about the destruction of the developer profession.
Read also🧐
- Botto artificial intelligence paintings sold for more than $1 million
- Artificial intelligence has learned to distinguish between 12 types of plastic
- Scientists have created xenobots that can reproduce
For 10 years in IT, I tried a lot: I worked as a system administrator and tester, I wrote in a dozen different languages programming, led the computer department of the editorial office of a printed newspaper and led news feeds high-tech portals. I can patch KDE2 for FreeBSD - and tell you in detail about all the nuances of this process. I dream about homemade R2-D2 and space flight.