Insights Into Artificial Intelligence

Commenter Alex has written a “rant” regarding AI that I want to share with all readers. The great virtue of this rant is that it explains in language that non-specialist can, er, grok what it’s all about, what the uses and misuses may be. Thanks to Alex:

Context, I’ve been working as an enterprise software engineer for almost 30 years (not specialised on AI).

The current LLMs [Large Language Models] are next word prediction machines. They aren’t AGI, and it’s very unlikely imo that they can be. That’s not to say that researchers won’t come up with a better tech, given the massive resource investment.

That is not to say that AI isn’t useful, or that it’s not _intelligent_, it’s just a very different kind of intelligence. They’re not conscious, and have no actual conceptual understanding, although there is work on world models. They have no morals or ethics. The best I would describe their intelligence would be as “functional” or “alien”. A calculator can give you the answer for 1+2, but has no understanding what that means. LLMs, due to their natural language basis, can easily masquerade as “human-like”, exploiting our natural tendency to anthropomorphise.

So, is it all just a bubble? No. You do not need intelligence to fill out forms, or look for patterns in data, or find references in vast seas of prior art. You don’t need to understand the essence of a horse, if you’ve seen a billion pictures of a horse, to draw a horse. You don’t need intelligence to copy code snippets from StackOverflow. You don’t need to pay someone to determine if the building you’re blowing up is full of schoolgirls or not, nobody cares if you’re wrong 20% of the time.

It’s not that the AI we currently have is intelligent, in a human sense. But the jobs it replaces don’t necessarily require much creative or critical thinking, just learned knowledge. The second aspect is cost-benefit. AI doesn’t have to be as good as the human it replaces. It only needs to reach 80% of the output at 1% of the cost. Humans are really, really expensive. API tokens are not. Enterprise cares about time-to-market, turnover and profit margins. Enterprise does not give a rat’s arse about quality. “Quality” is the crappiest product the customer will pay money for.

Software is a vast, abysmal collection of bugs and security issues, held together by code some guy once wrote for a hackathon on a weekend 5 years ago. The guy doesn’t work here anymore. The guy he handed the code over to doesn’t work here anymore. “Production” is where you put code so customers can pay you money to test it. Bugs get handled by overworked teams of new starters, provided enough customers bother to raise them. Customers pay $$$ for shit software that should morally be illegal to sell. But they pay anyway.

Enterprise loses all your data to l33thaxx0r69 from North Korea twice a year, who then sells it to call centres in Cambodia, where kidnapped slaves scam elderly westerners out of their retirement money. Nobody will hold enterprise accountable, ever.

AI can write shit software cheaper than a human. AI can create a Ghibli animation of Donald Trump eating babies in minutes, that AI agents can then spam onto slop channels across social media platforms, and earn you 10x or 100x of what you paid for the API usage from outraged Karens on Facebook. AI can summarise a meeting as good as a junior product manager who only understood half the implications of what the engineers said.

We don’t need AGI to replace jobs. We don’t need AI to replace all jobs. If there’s profit, enterprise will do it, because shareholder value defines C-suite compensation.

Sorry for the long rant. I’ve quit the software industry two years ago, because I came from a time where we prided ourselves in craftsmanship, and those times are gone. Just like the tailors who used to sew clothes got replaced with children working 12 hour shifts in sweatshops in Bangladesh.

AI will replace jobs, and back to Luke’s [Luke Gromen’s] point, it just needs to replace a few white collars who are leveraged over their ears to collapse the whole house of cards [our current societal structure] very, very hard.