Image by Author
When I heard AI could write decent code, I’ll admit it: I got a little shiver down my spine. Every day it seems like AI threatens a new occupation. SEO bloggers are quaking in their boots. Artists watch with horror as users flock to Midjourney’s Discord server. And now this.
Screenshot of Chat GPT
I’ve always been a proponent of the argument that technology does not remove jobs, it adds them. Yes, word processors made typists obsolete. And before that, cars made carriage drivers obsolete. But the overall number of jobs – good jobs, fun jobs, engaging jobs – has increased over time, despite the rapid influx of new technology.
The data agrees with my intuition. The World Economic Forum estimates that technology will add 12 million jobs by 2025. Economic historian James Bessen suggests that part of the jump in wages over the past two centuries is thanks to the rise of production, which in turn is thanks to automation and technology.
But this was the first time that I’d been faced with my replaceability. ChatGPT can write code.
Does this mean that AI will, ultimately, replace us?
I believe that it’s not only unlikely but impossible, thanks to the way AI is trained. There are skills (for e.g. these data science skills) AI will never be able to replace, no matter how advanced.
AI will never be creative. AI will never be innovative. AI will never be able to reason, think, or argue.
Right now, all it is is a very capable mimic. AI can pattern match, and it can do it incredibly well. But it can’t do the things that make human workers and human brains so invaluable.
Why is this? Think about how AI generates images to use as an example. Midjourney has been trained on millions of images taken from all parts of the web. To each image, there is some text associated. An image of a woman smiling will be described as a “woman smiling.” A Raphaelite work of art will be described with the name, the style, and the contents of the painting.
When you ask Midjourney to generate a painting of a woman smiling in the Raphaelite style, it is putting all its training data into a blender and spitting out the composite result. That’s not innovative or creative. The AI is not “thinking” or “reasoning” to create that image. It is just using pattern recognition to follow your prompt.
Image created by Midjourney | Prompt: a woman smiling, raphaelite style, painting
Nick Cave was disgusted by a song created by ChatGPT in his style. In explaining why, he showed understanding that what ChatGPT does is ‘mimicry, or replication, or pastiche’. And this is everything writing a good song isn’t. Then he explained what it is:”It is an act of self-murder that destroys all one has strived to produce in the past. It is those dangerous, heart-stopping departures that catapult the artist beyond the limits of what he or she recognises as their known self.” In that, I think Nick Cave explained the difference between a human and an AI. The way I read it is we should distinguish between generating and creating. All AI does is generate, while humans have the ability to create.
It seems ChatGPT agrees with Nick! I asked it to write a blog post about whether AI will ever be used to replace software engineers. It can chew up thousands of blog posts it’s read on the subject and spit out the best approximation of that. It is not considering your question and thinking about how best to respond. It’s just using the training data to complete your prompt.
ChatGPT says it best:
Image from ChatGPT
This means that AI, as it’s used and trained today, can never truly generate a unique work of art. It can never come up with an original thought or share a unique and personal experience. And it can’t be a software engineer.
All it can do is consume and regurgitate.
This is also why AI sometimes makes mistakes, like adding too many fingers to human hands or providing bad information/made-up reports in blog posts. It can’t know what’s right and what’s wrong, only what it’s consumed.
That all being said, I do find AI’s current use cases impressive and valuable. Aside from writing Python code to generate five random numbers, there’s a lot more than AI can do. And most of these skills fall under the umbrella of “make my job more interesting and less drudgework-based.”
For example, the bloggers we hire to write for our company blog often spend a lot of time optimizing blog posts for search engines. (We do this to help interested readers find our blog posts more easily.) This can be a little mind-numbing. Our writers like to focus on making a blog post interesting, informative, and fun. It’s less fun to ensure a blog post has an optimal number of keywords to reach readers.
AI is already used for spell-checking, too. Grammarly, for example, uses those same pattern-matching abilities to make sure that what you write fits the conventions it’s been taught on.
You can use AI for research, too. It is very effective at consolidating and conveying information because it has already scanned more text than any human could ever read in a lifetime. For instance, ChatGPT can explain to me how a camera works, how planes fly, or even quantum mechanics.
Finally, software engineers and data scientists may rely on AI to debug the code they’ve written. We’ve all been on StackOverflow. Think of AI as a StackOverflow assistant who has read every single question and answer and can consolidate the findings into a simple response to you.
The truth is, if AI can replace your job, it means you were doing something pretty menial, to begin with. AI’s current iterations should free you up to do more engaging, intellectual, or skill-based work.
There’s one area where AI can not only automate basic steps in your workflow but make humans work better. That’s in the medical field.
AI can ingest large amounts of data and spot patterns, so it’s exceptionally good at helping medical professionals make a diagnosis. While most clinicians will still prefer to speak with patients to get a holistic view of their symptoms, an AI assistant can monitor blood pressure and match symptoms with potential illnesses.
The field of AI in medicine is somewhat controversial still, ever since IBM bought the “Watson” AI that could beat Jeopardy! contestants to use it to diagnose patients. Ultimately, the AI had too much trouble with complex patient files, and it was unable to make reliable diagnoses. But as AI gets better at ingesting complicated and varied sources of information, and as clinicians get more used to relying on AI to interpret medical scans and symptoms, this is an area where AI could make a difference and save lives.
I got over my shock pretty quickly as soon as I started interacting with AI technology. The more I used it, the more it became clear that what I do, as the owner of a human brain, is not replaceable by AI. AI will never be able to problem-solve like I can. AI will never be able to recount personal experiences like I can.
In my opinion, AI is not here to replace us. It’s here to help us. And if it can replace you, then you should probably start adding some additional skills to your resume ASAP.
Nate Rosidi is a data scientist and in product strategy. He’s also an adjunct professor teaching analytics, and is the founder of StrataScratch, a platform helping data scientists prepare for their interviews with real interview questions from top companies. Connect with him on Twitter: StrataScratch or LinkedIn.