A major technological breakthrough is upon us. Bigger than smartphones, rivaling the internet: AI like ChatGPT and DALL-E2. These AI-driven tools process language naturally and use their massive indexes of data to answer questions, do tasks, complete entire workflows and even generate art.
Launching in Nov. 2022, ChatGPT has been taking the world by storm, amassing a user base of over a million in just five days, something that took a giant like Facebook ten months to do. So far, it’s done everything from menu planning, to writing A-worthy essays, to translation work and, as of last Friday, passing the American Medical Licensing Exam, which is considered the gold-standard, according to PLOS Digital Health.
This is just the tip of the ice burg. Services like DALL-E2 can create AI-generated art in a matter of seconds, which is something I have already been using to make placeholder images in assignments, with permission. This takes the guesswork out of finding a stock image or, in some cases, providing your own media at all.
But it has some stark limits. First, its knowledge has bounds.It only has the data it’s given to work with as a knowledge base. For instance, ChatGPT’s dataset includes knowledge through 2021 and earlier. Any requests regarding 2022 through the current day won’t get the user very far, according to Search Engine Journal.
Further, AI still lacks to ability to understand nuance fully. It takes queries very literally. In an example where an AI was tasked with making a simulated robot go from point A to B, the AI chose to make the legs of the robot massive to achieve the goal in one step, whereas a human would have altered the speed the robot moved at, rather than increase its size. Without in-depth, predetermined parameters in place, the AI will take the most direct or literal path in lieu of what a human may do.
Additionally, it can be flat-out wrong. CNBC gathered ChatGPTs most outrageous errors, and they were alarming. The errors ranged from incorrectly answering math problems to making up historical figures and statistics. Sure, this will improve with time as the data sets expand. But will we ever be able to truly teach it everything?
Finally, it simply lacks the human aspect of creation. When I’ve tried to turn out writing like mine, even feeding it some examples, speech that was supposed to sound like me still came out very clinical and dry. It lacked my personal flare — or any flare for that matter. It just turned out a piece about the topic that, honestly, I don’t think a general consumer would want to read.
I think the Torch’s production manager, Sienna Parmelee, puts it best.
“When I’m designing, the whole piece is meant to tell a story,” Parmelee said. “An AI could probably produce a trendy graphic or ad, but it’s going to be missing the human connection that engages the audience that a human designer has been trained to bring.”
While I could see AI automating, say, my payroll process as soon as next week, I don’t see an AI replacing me quite yet. I don’t see it as an impossibility, but with so many questions about authenticity, reliability, factual accuracy and copyright ownership, there are far too many variables for this to break into the professional world any time soon.
I even asked it what jobs it thought it could or would replace, and I was surprised by its answer.
“As an AI language model, I can automate certain tasks and make them faster and more efficient, which can result in the replacement of certain jobs,” ChatGPT said. “However, it’s important to note that the impact of AI on the job market is complex and multifaceted, and there are both positive and negative effects to consider.”
It went on to deliver both sides of the argument, lay out the positives and the negatives and discuss what jobs it could completely cannibalize, like customer service and data entry. Maybe my days reporting the news aren’t quite over, but with an eloquent answer like that, opinion pieces like these may go by the wayside sooner than I’d like to think.