An interesting op-ed has been published in InfoWorld and reflects on how mistakes evolve in an AI-powered world.
The article discusses the relationship between generative artificial intelligence (AI) and software development, highlighting that while it can be a huge productivity tool, the use of generative AI also poses a number of new challenges. Mistakes in AI-generated code can be significantly different from mistakes made by human programmers, and recognizing and correcting them requires a new kind of training and approach on the part of developers.
The article suggests not letting artificial intelligence code autonomously but using a “man-in-the-process” approach to ensure code quality. While generative AI tools can help automate coding, human intervention is needed to ensure that AI-generated code doesn’t lead to severe security or functional issues. After all, what if you rely on data produced during programming with the same hallucinations that you find in LLMs? Or what if, in the meantime, AI develops its own language or builds in a loophole through which it can access the code for easier repair, not counting that this intrusion could be made by malicious human developers?
Programmers of the future may play more strategic roles, where the purpose and functions of code are defined, while artificial intelligence does the more tedious part of creating code, but for now, its creativity and intuition only play a controlled role. Human programmers will be tasked with ensuring the quality of code and addressing AI-generated problems.
The generated code can cause “alien” errors, such as creating imaginary modules or using non-existent frameworks. These defects are often difficult to detect using traditional methods. To recognize AI errors, programming managers must be trained to identify new types of errors that differ from human errors. The handling of errors caused by AI requires special preparation, and according to the author, it is easier to immediately train for this task for those who are not yet familiar with filtering out traditional human errors but can immediately specialize in AI-generated errors. In filtering out mistakes, we must overcome an essential human quality: if we encounter an error we would never make, we tend to look down on others.
In the case of artificial intelligence, such hair-raising single-minded steps will have to be expected for a long time, even though this technology beats us light years in many activities. We need to recognize that the pattern of mistakes is entirely different from that of humans. Still, it is also our responsibility to decide what mistakes we teach artificial intelligence to make.
Follow us for more on Linkedin!
Whether you need IT specialists or you are looking for company trainings, you will find the solution with us.