We can do better
Between 2009 and 2015 Volkswagen sold 11 million cars with engine control software that would emit up to 40x less NOx when tested than during normal driving (Wikipedia). At VW scale it meant very real consequences including an increase in birth defects and premature deaths.
It was not an act of a corporation in the abstract, it took multiple people, managers and engineers, to make that decision, implement it, and even more who knew that the resulting measurements in test conditions did not match reality who chose to stay silent.
If you are a software engineer: would you have done it? If your PM came to you with the task to help in hiding the actual emissions of those engines, would you have delivered it? Worked out the details, maybe through some back and forth with others, implemented it, possibly ended up getting a promotion and a bonus for delivering something that affected the bottom line? Would you have approved pull requests sent by others to this effect?
If you are a manager: would you have done it? Would you be receptive to this idea, circulate it in the company, find the resources for it? Oversee the delivery, see it work as intended, maybe write a nice enthusiastic summary to some Slack channel somewhere?
Would it affect your decision if it was something that the majority of your industry has accepted as the cost of doing business? Or if it was technically not illegal, just unethical?
The tools we use, the systems be build, what we enable or push the users of our software to do – they all matter, and right now a large part of the industry decided that generative AI is the way to move faster. It is impossible to make that decision without taking part in the normalization of consequences:
- the massive theft of other people’s work – books, articles, blogs, pictures, films, virtually anything and everything that can be scanned, downloaded, or scraped, without the creators consent but undercutting their income,
- made worse by the increase of publishing costs through higher hardware prices and relentless scraping and re-scraping of online pages,
- the human rights costs,
- the unsustainable data centers,
- the acceleration of the climate crisis,
- the feudalisation of software engineering as a discipline through the transfer of control to a handful of corporations,
- the financial bubble,
all while undermining the quality of software being built. That quality is not an abstract concept either, it maps directly to higher maintenance costs, lock-in through increasing the amount of code that needs to be maintained (LLMs make extruding large amounts of code cheap enough to make people believe it does not matter anymore; at the same time, skill atrophy is a thing), lower security (partially through increased complexity, partially through bad code moved to production with rushed or non-existant human reviews) and higher error rate.
Participation in this system means being complicit in allowing generative AI providers to claim they are too big and too important to have to obey the rules. All for the carrot of AGI on the one hand and the promise of moving faster on the other.
Is this really the best we can do, as an industry and as a profession? I firmly believe we can do better. We must do better.