Skip to main content

AI is inconsistent and you have to adapt

· 4 min read

Ask anyone on the street what they think about AI, and you’ll probably hear that they’re concerned. That’s to be expected since AI is blowing up today’s headlines, and not always in a good way.

For instance, one thing on many people’s minds is how AI will affect their jobs. That’s a valid fear. After all, technology has changed how we work before. Look at what happened after the industrial revolution and the rise of the personal computer in the 1980s. An estimated 3.5 million jobs disappeared and entire industries went out of existence due to computing and the internet.

Still, don’t go updating your resume just yet. Remember that there are many things that AI can’t do well. It still lacks the creativity and innovation that humans excel at and can’t make the same ethical and moral judgments that we do.

Even among the jobs that AI can do, it’s inconsistent at best. There’s no way to completely replace any given task from start to finish with AI at this point, regardless of the AI tool you’re using — ChatGPT, Stable Diffusion, or FaceSwap.

What employers — and the rest of us — need to remember is: You can’t rely on the results of AI. It still requires human supervision.

Case in point: a lawyer in New York used ChatGPT to research legal citations and judicial decisions for a legal brief submitted to a court. The lawyer didn’t check the authenticity of the cases before submitting the brief. As it turns out, ChatGPT made them up, and the lawyer may face sanctions because of it.

This may be one of the first high-profile instances of ChatGPT lying and getting caught, but it’s certainly not the only one. As anyone who’s used ChatGPT knows, it can “hallucinate” or spit out false responses to prompts. The phenomenon is common enough that tech firm Nvidia developed NeMo Guardrails, an open source tool to add constraints to LLM systems to keep them from returning inaccurate information.

Of course, AI is capable of providing plenty of factual information. It just needs a human to evaluate the results and parse out the hallucinations from the reality. However, even when it gives you accurate outputs, it might not deliver what you’re looking for. You may have to test many different inputs to get output that matches your request.

That will remain true even as AI tools advance. For instance, look at AutoGPT, an open source app that automates multi-step projects, like back-and-forth prompting with ChatGPT. If you’re writing a blog article, you can ask AutoGPT to feed tasks to ChatGPT to help it focus the results. But, at the end of the day, that article is still not going to be as good as an article edited by a human.

There are ways to improve your results, of course, by tuning the data and limiting what AI can do. That’s how FaceSwap is able to work for all users. FaceSwap works well because it’s tuned to be as consistent as possible, and also because it’s limited in what it can do — it’s not able to overstep its constraints and make things up. It is programmed to stick to what it can do, and it does it well.

The overall lesson isn’t that AI is bad; it’s inconsistent, and it needs monitoring. That’s where humans come back into the equation, to be there “looking over AI’s shoulder” to verify results, making sure it’s not only accurate, but also the result you’re looking for. Luckily, unlike humans, AI doesn’t mind being micromanaged.

By the way, there’s a bright side to all of this. Remember the 3.5 million jobs lost in the 1980s? Well, in the decades following, technology added an estimated net gain of 15.8 million jobs in new industries like computer hardware manufacturing and software development.

The same thing could happen today. There’s already a new wave of AI-related jobs that were unheard of before, like data tuners and prompt engineers.

Just like any previous revolution, some jobs will go away due to AI, others will change, and new jobs will arise. The main thing to remember is that we’re not in danger of AI taking over just yet, because it’s far too inconsistent to rule the world. But we will all need to adapt and accept the changes that come to our work and personal lives due to the power of AI.