Home / Technology / AI Writes Code: Can It Be Trusted?
AI Writes Code: Can It Be Trusted?
12 Mar
Summary
- AI code generators may speed development but introduce bugs.
- New start-ups aim to verify AI-written code like math proofs.
- Concerns remain about defining code correctness for complex systems.
Researchers have highlighted that artificial intelligence technologies capable of generating computer code, while potentially accelerating software development, may also degrade code quality and introduce bugs. This has spurred a new generation of Silicon Valley start-ups focused on developing AI systems that can automatically verify code. Companies like Axiom Math and Harmonic are pioneering this approach, seeking to apply principles similar to mathematical proofs to ensure code integrity.
Axiom, valued at $1.6 billion after raising $200 million, has developed technology that originated from solving complex math problems. Their system, AxiomProver, achieved a perfect score on the Putnam Exam and utilizes the Lean programming language, originally designed for mathematical proofs. The goal is to transfer this rigorous verification capability to computer code.
However, experts express caution, noting that defining code correctness is more challenging than verifying mathematical statements. While AI can verify certain aspects of code, the complexity of modern software, particularly for applications like social media handling millions of users, presents significant hurdles. The distinction between correct and incorrect code is not always clear-cut.
