AI Is An Oxymoron As Bot Writer Authors Articles Riddled With Factual Errors
Everyone that does creative work, whether it's in visual art
, writing, music, film, or even chemical engineering
, should be concerned about AI overtaking their jobs at some point. It's not happening right now (unless you work at CNET), but it will happen eventually. "Eventually" may not mean "soon," though, and there's no better proof of that than what's going on over at CNET.
The long-lived tech site has started posting search-optimized "evergreen
" articles with basic information about best financial practices and investing. CNET has posted as many as twelve of these in a day, and they have natural-language headlines clearly intended to pull in readers
from web searches.
These posts are not written by a human. They're written by an AI, likely one similar to (if not actually) ChatGPT
. While CNET claims that the articles are "reviewed, fact-checked, and edited by a human editor with topical expertise," Futurism
blog reviewed several articles and found egregious errors in all of them.
One of the articles in question was linked directly from CNET's post
explaining that these contentious creations were written by AI. That's right: CNET didn't actually explain that these posts were written by AI, initially, which caused an uproar that forced CNET to address the issue. That prompted the "reviewed" statement above. CNET's response to the uproar was to remove "Staff" in the "CNET Money Staff" byline.
The thing is, in this article, the AI writes things like "if you deposit $10,000 into a savings account that earns 3% interest compounding annually, you'll earn $10,300 at the end of the first year." We probably don't even have to explain what's wrong with this sentence, but to spell it out for you, you will *have* $10,300, not *earn* $10,300. Your earnings will be $300.
This is a very dumb error and an obvious mistake to anyone who understands percentages, but unfortunately these articles, with their simple, natural-language headlines (like "What is Compound Interest?"), are likely to reach those who understand the least. Basic errors like this are a real problem, and this is far from the only example.
The enormous correction for one article that CNET posted.
In the case of that particular article, CNET posted a huge correction at the bottom of the page that addresses all of the issues. We do have to credit CNET with admitting the errors instead of quietly correcting them without notice, but the point is that the company shouldn't have been posting AI-written articles without extremely careful editing in the first place.
The thing about current "artificial intelligence" is that they are essentially black boxes that take in patterned input and then can recreate those patterns. They're machines that recognize patterns and can re-create them in a convincing way. However, they absolutely do not "think"; they have no understanding of these patterns. This is why Stable Diffusion makes hands with 16 fingers, and why ChatGPT says
that APR and APY are the same thing.
Until we achieve generalized artificial intelligence that can speak on topics with understanding instead of simply playing "monkey see, monkey do" with data, you'd better leave all of your "what is" and "how do I" questions for another human being.