Health Transformation Institute (HTI)
institute for continuous health transformation
Joaquim Cardoso MSc
Founder, Chief Researcher & Editor
December 9, 2022
Is ignoring everything that is known about code the best way to write programs?
Science
J. ZICO KOLTER
8 Dec 2022
Abstract
- Competitive programming problems represent a challenging task for even skilled programmers: Given a short natural language description of an algorithmic problem, contestants must quickly write a program that solves the task.
- On page 1092 of this issue, Li et al. (1) present the AlphaCode system, which represents a substantial step forward in the development of machine learning (ML) models that can synthesize computer programs to solve these types of challenging problems.
- But what is perhaps most surprising about the system is what AlphaCode does not do: AlphaCode contains no explicit built-in knowledge about the structure of computer code.
- Instead, AlphaCode relies on a purely “data-driven” approach to writing code, learning the structure of computer programs by simply observing lots of existing code.
COVER
This image shows blocks with snippets of programs of varying viability (height) created by humans (purple) or by AlphaCode (white).
AlphaCode is an artificial intelligence system that writes computer programs at a competitive level and solves new problems by generating and filtering millions of diverse candidates.
In the future, code-generation systems like AlphaCode may assist and educate programmers. See pages 1056 and 1092
Originally published at https://www.science.org
