AlphaCode: A New AI Development

 

Introduction

The new Artificial Intelligence (AI) system for writing computer code that was developed by DeepMind is called AlphaCode. According to the researchers who conducted the study, AlphaCode is capable of achieving an average human-level performance when it comes to solving programming contests.

Usage of AlphaCode

The development of an AI-assisted coding platform that is capable of creating coding programmes in response to a high-level description of the problem the code needs to solve could have a significant impact on the productivity of programmers. It could even change the culture of programming by shifting human work to the process of formulating problems for the AI to solve. 

To this day, human beings have been needed to write the code for the solutions to unique programming issues. Although recent neural network models have shown impressive code-generation abilities, they still perform poorly on more complex programming tasks that require critical thinking and problem-solving skills. One example of this is the competitive programming challenges that human programmers frequently take part in.

Researchers from DeepMind present AlphaCode, an AI-assisted coding system that can achieve approximately human-level performance when solving problems from the Codeforces platform, which frequently hosts international coding competitions. AlphaCode was developed by DeepMind researchers. By iteratively predicting segments of code based on the prior segment and generating millions of potential candidate solutions, AlphaCode was able to solve natural language problems that had never been seen before. 

This was accomplished through the use of self-supervised learning and an encoder-decoder transformer architecture. After that, these prospective solutions were selected and clustered based on the validation that they functionally passed basic test cases. This resulted in a maximum of ten possible solutions, all of which were generated without any built-in knowledge about the structure of computer code.When examined using Codeforces' challenges, AlphaCode performed roughly at the same level as a typical human rival. It was able to obtain an overall average score within the top 54.3% of human participants while the number of submitted solutions was limited to 10. However, it was able to solve 66% of the issues with its very first submission.

Conclusion

J. Zico Kolter argues in a Perspective article that covers the benefits and drawbacks of AlphaCode that, "Ultimately, AlphaCode performs surprisingly well on previously unforeseen coding difficulties, regardless of the degree to which it 'really' understands the task."

Written by Chitraksh Mayank