Decoding the Dangers: AI's Role in Software Engineering Under Scrutiny

As we venture down a new unknown path in Ai driven software development, there are numerous dangers lurking ahead that businesses need to take urgent cognisance of and as more and more software code is either generated by, or assisted via Ai, there are credible risks that are starting to emerge.

Angus Norton, in his tech blog, distinguishes between current Ai hype and what he terms applied Ai. Here he provides a definition where, “Applied AI is the practical implementation of artificial intelligence technologies to solve real-world problems and enhance various aspects of business and daily life”.

When it comes to software programing, the goal is the same – to provide real-world solutions by providing systems that process and collect data and translates this into useable information or to run complex systems efficiently and accurately. However, when using AI to enhance or generate software, it becomes a complicated process that needs to be closely managed to ensure accuracy. 

A recent investigation by a code analysis company, found that the quality of code generated, since the introduction of AI coding assistants, has actually become poorer. 

Another study considered refactoring – meaning the process of the improvement of existing code without changing its behaviour, and the study found that existing AI solutions only deliver functionally correct refactoring’s in 37% of cases. 

Birgitta Böckeler, Global Lead for AI-First Software delivery at Thoughtworks recently commented that “The underlying models they(Ai’s) use are quite generic and based on a huge amount of training data that is not always relevant to the task in hand. And large language models also make things up—they ‘hallucinate’.” 

The mistakes AI software coding or coding assistants tend to make, are reported as often being quite subtle, and therefor difficult to detect. The code might even seem correct on the surface and can be quite convincing while still containing errors. 

The development of solid best practices, going forward, is going to be paramount in the future use of AI assisted software coding. Developers and software engineers are going to, of need, be tasked with diligently checking code generated by Ai assistants in fine detail, as not doing so, can result in pervasive errors in software. 

The future of coding with AI, would at this stage paint a picture of people without much skill or experience creating software but this could equally result in catastrophic disasters as these AI generated coding models really don’t know what good code is, and there is a grave danger that companies and systems end up getting a lot of poor quality code with a multitude of bugs, that are not easy to spot or correct. 

A recent report by Wired Consulting states that, “academic research has found that coders using AI assistants, not only write significantly less secure code than those not using them, they are simultaneously more likely to believe they are writing secure code”. “AI can hallucinate fixes for common vulnerabilities and leave off essential validation checks, exposing systems to hackers”.

Tech and advanced software development, is moving forward at an ever increasing pace and there may be the desire to replace caution with speed. The evidence though, indicates that there should be more caution and that protocols are put in place that ensure code engineers and programmers, are vigilantly checking code developed by Ai.