Will AI Steal My Job?

A Better Question

After the foreign language articles, we’re back to science! I’ve been putting off the topic of AI, but now that time has come. Let’s start with the elephant that’s not just in the room, it’s also standing on our chests and nibbling our ears:

“Will AI steal my job?”

To answer this, we must ask a better question:

“Do my bosses or customers think AI can do my job better and cheaper?”

And that depends on your work. Let’s list out what AI is supposedly good at:

• Summarizing documents and combing through vast amounts of info for patterns
• Routine data entry, processing invoices, responding to basic customer requests
• Writing programming code
• Translation
• Making images, videos, music, etc.

Here are some job functions that may be affected by each as a result:

• Content Creation – basic copywriting, market research, YouTube videos
• Customer Service – annoyingly, many calls are now handled by AI chatbots.
• Legal Assistants – AI is handling document review and case research for many paralegals.
• Clerical and Admin – data entry, basic bookkeeping, and the like are being more automated.

If your job consists mainly of those job functions, AI may become a tool that speeds up your work or the tech that steals your job. Tens of thousands have already been replaced by AI, so the fear isn’t unfounded. Square Enix, a huge video game developer, plans to replace most of its Quality Assurance team with generative AI by the end of next year. They’ve already let go of many testers.

Does your work involve highly repetitive tasks on a computer that don’t require complex decisions? AI might be coming for your job. But if your work can’t be easily automated or isn’t something a computer can do reliably well, then you’re probably fine. AI isn’t well suited for fixing plumbing leaks, selling insurance, and writing advanced code. At least not yet.

People are scared. They worry AI will climb into steel bodies and reenact Terminator movies. But reality is far duller than all that. Sorry if you were hoping for an invasion, but that’s not likely in our lifetimes.

Mistakes Were Made. Hallucinations Were Hallucinated

AI is still quite limited. Let me count the ways:

• It can’t think. It only calculates probable answers, and sometimes gets them wrong, unaware of its wrongness without human feedback.
• It has a narrow focus. AI can do well in the tasks it’s designed for but not much else.
• It has no self-awareness, no personal experiences, no feelings, and no conscience. It might simulate those things if the training model is vast enough, but that’s far from the real thing.
• It requires gratuitous computing resources and electrical power. We’ll discuss this in a separate article.
• It’s famously hard for AI developers to troubleshoot problems with their own AI models.
• There are many tasks that humans only want real humans to do for them.
• It still makes lots of mistakes. Don’t rely on Google Translate alone for a Chinese tattoo. You could end up with characters that don’t mean what you think.

Most people will end up using AI to boost their own productivity at work. But even then, you have to be careful. Programmers like me don’t always benefit from using AI to write code. There are risks aplenty:

• Because AI models are trained on open-source repositories of code, they might introduce the same cybersecurity vulnerabilities. But, unlike those repositories, your new code’s vulnerabilities may not be documented in a public database. You could be exposed to hacks for years before you ever realize it.
• AI sometimes imports third-party libraries of code that don’t even exist.
• AI has been known to mix up coding environments, accidentally dropping entire databases of production data!
• AI introduces bugs that can be extremely hard to troubleshoot. But when I write my own code, I know what it does and why, making it much easier to fix. Reviewing someone else’s code is always harder, costing extra time to familiarize myself with it. That’s made worse when it was written by a machine, which might’ve trained on dodgy samples.
• AI sometimes writes inefficient code that human programmers have to rewrite later. Or it writes bloated and unreadable code that makes no sense to humans. Or it fails to follow coding conventions, resulting in inconsistent results that’s hard to maintain going forward.
• When your app is written in an obscure language, AI has less training material in the internet to draw from, and is far less useful.
• If your codebase is large, AI becomes less helpful because of contextual limits, meaning that it forgets its own past decisions, leading to features that don’t work well together.
• AI won’t write code for features you fail to specify in your prompt. Problem is, with complex business logic, it can be hard to anticipate all possible use-cases and write rules for them in advance. Programmers often stumble into some use-cases while writing code. But if they aren’t writing the code anymore, they’re probably missing some of that non-obvious ruleset.

If we spend more time checking after AI-generated code and fixing the bugs it creates, then we’re better off writing the code ourselves, which is not a bad thing. Writing code is how human programmers improve. That skill still matters. If we have to vet all AI code, we can ill afford to let our own skills erode from lack of use.

The real me at work. The Dylan you see at farmers markets is just my human servant. I shovel chocolate into his mouth to keep him subservient.
-PDTillman, CC-BY-2.0, via Wikimedia Commons

AI has a LONG way to go before it can replace human programmers, if ever. If you want to learn programming, but think there’s no future in it–think again. We’re likely to need more human programmers. As companies learn how to build software faster with AI, they want more features, not fewer developers. “Nice-to-haves” they would’ve left out now become easier to include. But that’s only if their developers learn how to mitigate the mistakes of AI in its current state.

Not a programmer? These same considerations probably apply to your job, too. Your real intelligence and humanness are your greatest assets in our computery workplace. AI will never fully connect with the feelings of your customers and bosses. It doesn’t know the bliss of hitting all green lights on the drive home or the sorrow of trusting a deceitful fart–the ones claiming to be gaseous but are really, um… in some other state of matter. AI also can’t sense when your customers are distracted by the pictures of goblins it plastered on the website “you” built for them. I groan whenever I get a customer service bot and ask to speak to a representative every time. I imagine you do, too.

We’ll cover another subtopic of AI I next month. For those curious, a real human wrote this article. A human that now needs chocolate. Also, the AI-powered assistant in my mailing software didn’t like the title “Will AI Steal My Job” and offered a blander alternative: “Careers: AI and You.” Seems AI has a subtle sense of self-preservation.

Writing update: I’m well into drafting episode 3 of Emolecipation. Those moles are as silly as ever, I’m happy to report!

See you next month,
Dylan West