AI (as in Artificial Intelligence, not ‘augmented’ or ‘automated’ intelligence) has rapidly become a transformational factor for dozens of markets, including software development itself.
Even if we were aware that generative AIs like ChatGPT can ultimately generate bullshit, as my colleague Jason Bloomberg says, and we know they are getting overhyped across social networks and overvalued by vulture capitalists, nobody could have predicted the excitement the public and press would feel once they glimpsed the future.
We’re seeing AI work its way into everything—almost every platform or software product has put some form of generative AI magic into the mix. No vendor wants to be late to the party.
Governing AI in development
IT executives are wrestling with how to safely govern the use of AI in their development groups. Maybe they don’t realize it’s technically already being used by most developers in their IDEs, for starters, in the form of autocomplete and code co-piloting.
Then further, we are seeing natural language chatbots being used as ‘no-code wizards’ that can automatically generate the ‘best usual’ code for many component aspects of an application and its infrastructure-as-code specifications from a starting prompt.
After all, no developer is interested in coding their own permissions or horizontal autoscaling, if there’s already something ready to go – they would much rather focus on the differentiated application functionality they are incentivized to build.
Even Stack Overflow is incorporating a ChatGPT natural language prompt feature for generating code based on code that was contributed by other developers. What could go wrong with this recursive loop?
If there are bugs, or security flaws in the generated code, we’d expect human developers to thoroughly check for them. But we all know that’s easier said than done when teams just want to move faster.
The IP proprietary cliff
The entertainment and publishing industries are already launching lawsuits and even worker strikes to buck against the use of ChatGPT as a virtual writer and visual AIs like Midjourney, which are trained on bodies of existing copyrighted works.
Early proponents of AI-generated writing attempt to handwave this concern away, saying something about a new job description for ‘prompt writer’ being a new role that is simply the human using the AI as another creative tool.
So if a company has AI-generated code in their codebase, who should be credited as the author? Do they have the rights to use that IP in their product or business? I would say the future here looks murky. Automated routines have been thoroughly scanning codebases for patent infringements ever since MS-DOS was a thing. Now imagine them armed with the latest AI patent trolls, which can automate this job a thousand times faster than the old search tools, uprooting even more hidden liabilities, and filing threatening settlement letters.
Even if partially generated code doesn’t contain a clear patent violation, a vendor would have a very hard time defending generated code as proprietary work – and it is likely that the code they actually authored could be used for machine learning in someone else’s AI model.
Will AI displace developers?
It’s the most frequent question we are asked by developers in 2023. My answer right now is that AI will settle in and mostly replace the kinds of development work that no developers want to do.
For instance, an AI makes quick work of helping data scientists become productive without having to massage and meta tag data. For software training, an AI can act as an infinitely patient tutor and pair programming assistant for bringing junior devs up to speed.
The quality of documentation many tools can instantly create from almost any application, or even a stack of paper, is mind-blowing. Still, that doesn’t mean we’ll stop needing technical writers and educators – they’ll just continue to move their value upstream to the newer aspects of making better decisions about using our application estates.
Because of the known factual hallucinations and IP issues of an AI model trained on codebases scraped off internet repositories, I don’t think AI will ever fully displace development work in most sectors of the economy. In fact, it is likely to bring even more of a new class of low-code developers into the market who have less technical skill but more business acumen.
The Intellyx Take
Even I’m starting to wonder, as our most miserly clients tell us: “Why don’t I just get ChatGPT to write your analysis for free instead?”
Sure, maybe you should use a chatbot to replace me—if you aren’t afraid of the reputational risk that might occur when your audience realizes they are reading auto-generated content, filled with silicon dreams.
No chatbots were used to write this content, as the editor told me he could tell if I did. At the time of writing, none of the vendors mentioned here are Intellyx clients.
