Feeling dumb? Let Google's latest AI invention simplify that wordy writing for you Exasperated by the prolix verbiage and gratuitously convoluted phraseology that so often permeates technical treatises, philosophical discourses, or the meandering expositions of journalists afflicted by a lack of rhetorical economy? Then Google has a new AI feature for you - provided you use iOS, that is. The Chocolate …
Just for the record, it's pretty much standard practice in most newsrooms that the reporter doesn't the headline or subhed of a piece -- that's the job of the editors.
The "Zog" in the subhed in question is a reference to a Gary Larson Far Side cartoon1.
____________________
It's not just 'Muricans.
One of the thing I've been doing for a living over the past 30 years is write product manuals. Which was never a problem, except that for the past 10 years I've been getting more and more users asking me if I couldn't just give them a YouTube video instead.
And that's a global phenomenon. Reading is difficult. It's hard work. Makes brain tired. Want YouTube on phone. Duh.
So the next iteration of this will probably not just simplify text but provide a link with an AI generated talking head so the Great Unwashed won't have to read anymore.
Idiocracy is just around the corner here.
> asking me if I couldn't just give them a YouTube video instead
I *hate* this phenomenon. I was helping someone assemble a flat-pack drawer intended to be installed within a kitchen cabinet. "First, we'll read the instructions", sez I. "Oh, they're online", sez he.
What was online was a video, showing the unboxing of the kit, followed by long-distance shots of a chap assembling unspecified parts at a speed that indicating he'd been in training for a week.
Back to working out how the damn thing goes together from first principles, with lots of measuring (twice) and "offering up".
A couple of days ago, I encountered - on youtube - someone had made a video where they were reading out the speech bubbles and text from a comic/graphic novel that I remembered from back in the eighties.. no animation, no highlighting which speech bubble (karaoke style), just shots of the cartoon page, advancing as they went. At least the "creator" didnt include their own head in the frame.
I suppose it may be useful for illiterates and non-english readers, but how did they get that far to start with? And how can you understand a comic style layout without knowing which panel is related to the words being spoken (which is why karaoke highlighting would help).
Also, apart from mentioning the cartoon/graphic novel name, they completely failed to credit the original writer/artist.
A youtube of a comic being read aloud, how far the internet has fallen....
Consider Google is responsible for a lot of the AI written garbage currently infesting the web, it seems like they created a problem and then introduced an app to be its solution - with the bonus that they can gather data from iOS users. That's the app isn't for Android, they've already got all the data there is on them so this app is superfluous.
Not only that - there's a huge amount of verbiage online these days that is so poorly written (by humans) that it's meaning, when taking at face value, is exactly the opposite of the message the writer was attempting to display: ie, lots of people are missing out negatives in their rapidly constructed comments.
If, big IF, this works as described, AI may have a real use since editors are becoming a rare species if some websites and the remnants of printed and TV media are any guide.
Something that can turn manglese from manglement (to coin a phrase I think) into information is a fantastic achievement. However I suspect output would be an empty page.
OTGH, turning turgid academic verbiage into information accurately is great. Merely translating from insider jargon to plain language is not dumbing down.
Merely translating from insider jargon to plain language is not dumbing down.
One man's jargon is another man's shorthand.
The audience articles like the one quoted is not the general public. It is aimed at specialists in the subject and they know what the "jargon" means. It is concise, accurate and unambiguous.
As for avoiding prolixity look at the number of words used in each example. The original is ten lines, the "simplified" version is twelve.
So, a tool that loses some of the more subtle meanings and precision to make a technical piece of writing palatable to the uninitiated.
This post has been deleted by its author
I could mostly understand the original† and could compare it with the zogified version.
As far as I could determine the translation was accurate enough but the original was more balanced in tone, vocabulary and comprehension required.
The machine version less so - I would love to run it through Grammarly to see what it makes of it.
"Changes in the composition of lung immune cells" I would read as changes in the types‡ and number of immune cell found in the lung rather than internal changes within the individual, presumably already resident immune cells. The translation "the immune cells in the lungs change" for me implies the rather different second sense.
As it is I suspect most radiology and pathology reports are routinely "machine assisted" or at least copy and paste.
In most most specialised contexts such translations are pretty pointless as the target readers don't have the minimal contextual knowledge to make any sense of the translation. In this case the polloi have no or wildly inaccurate knowledge of human anatomy and definitely none of pulmonary physiology.
Basically the effective translation is "Mate, your lungs are fucked. If you don't stop smoking they will be fuckeder."
† long story. ‡ eosinophils, neutrophils, lymphocytes etc etc
As far as I could determine the translation was accurate enough
MD here. Honestly? Errmmm... Not really.
Reading them side by side, a lot of the meaning got lost. Just to mention a few: the original text describes a contiously ongoing process, where the dumbed down version presents it as an end result, something stationary. Furthermore, it described damage, thickening of the tissue *between* the alveoli, the small lung sacks where oxygen is taken in, and CO2 expelled. The dumb text looses this distinction, throws everything "in the lung proximity" on one heap. Imagine this as a process over a membrane: if the membrane is thicker, this process becomes more hard. The "scarring" (different process, but hey) getting thicker of that "membrane" is because of a continuously ongoing infectious process against a "trigger". This trigger can be, yes, cigarette smoke, but also many other things like for example coal dust, asbestos, diesel fumes, pigeon fungi, solvents in paints and glues, and loads more, you get the idea. That oxygen needs to get in your blood, so it can be distributed across your body. But the walls of your vessels also get thicker because of this "out of control" infectious process. So your "sats" go down (measured with that clothes peg thing on your finger), you feel out of breath more and more. Now, did that text give you that?
I do understand this is a dumbing down tool, but then we also come into the topic of what is dumbing down? How to dumb down. And most important, why to dumb down. This is also an issue what I have seen (lost) when people use LLMs for for example making a quick abstract of a full text. Meaning is lost. And more scary, meaning is interpreted differently, and thus changed. Why this IMHO is worrying is...
For example: globally a physician has the legal obligation to get patient consent before treatment. For that (s)he has to explain what is going on, what the (best) intervention is, and ask the patient whether (s)he would like to proceed as suggested/ advised. Of course, it is critical in this process that patient understands what is said, so (s)he can make an educated decision. That's why you are always asked (I hope) if you are happy, and whether you have any questions.
The text here could not be used for that, because a lot of its meaning is lost. So yeah, sure it "broadly" describes what is written. But is it properly "translated"? Personally I would not spend the Wattage/ energy on it. But hey, this is Google trying to sell you something, so you know what that means.
As it is I suspect most radiology and pathology reports are routinely "machine assisted" or at least copy and paste.
Yes, you are right there. And if not "on the machine" when writing it down, then also in the head of the diagnosing physician. We all do something (again, I hope) called DD, differential diagnosis, which basically is a SOP for diagnosing, making sure you tick all boxes and consider options and alternatives. Just like any problem solving process I suppose...
So one use of AI is to generate screeds of verbiage from a simpler prompt. Now there is another use of AI to simplify screeds of verbiage down to a simple summary
I reckon Google iis actually doing a search on the verbiage and when they recognise it as their own to send back the original prompt that it was based on.
Obviously just publishing the simple version in the first place is out of the question, as that wouldn't involve Google.
..."Simplify," which is meant to dumb down complicated writing without losing its meaning....
FFS! I've long been of the opinion that using AI will lead to a large reduction in human capacity to learn and improve, ultimately leading to a loss of the ability to think. Now Google come along with something that will make it even easier to stay dumb:-) *1
When it comes to preserving meaning during simplification, that is a hard task for a human let alone for a machine. Machines don't have the ability to pause and say 'Oh, I see what the author meant' and start again. We've all written a sentence that needed revision a few times before it accurately reflected what we - the original author - intended it to. Good luck getting a machine to do a good job of that! Also, you can't go back to the AI and ask for clarification:-)
There is a big difference between 'easy to read' knowledge and 'easy to understand or comprehend' knowledge. Also, this does nothing for the spoken word which IMO is at least important.
IMHO, people should be constantly challenging themselves to learn and improve themselves more. If you don't understand a sentence or a particular word, then do what I do and look it up i.e. *learn* something. I still love doing that. They say - and I believe that - 'Every day is a school day!'.
If we start to blindly trust machine output without thought, I fear the result won't be good for us. Read Frederik Pohl and Cyril Kornbluth's "The Space Merchants" to see what happens when society believes machine output.... badly translated output, badly edited output or maliciously produced output, it makes no difference if it is believed.
Yes, that book is a work of fiction. Here's hoping it doesn't become a prophecy!
*1 Also known as shift even more power to Google/the technocracy.
The "simplified" version is both longer, and less accurate. Human editors at least have a chance of making sure the accuracy remains as the word count (of simpler words and sentences) goes up. AI still doesn't have a clue about meaning.
See also Yahoo! Mail's new AI summary feature, which frequently gets it wrong, meaning if you want to be sure about an email's meaning, you wind up reading BOTH every time. Saves no time.
I'm put in mind of something I recall reading about the 1950s Raleigh RM-1 mo-ped. Keen to diversify and get on the contemporary mo-ped craze the company did its market research and determined that potential customers were after simplicity. Of course what they really meant was easy to use. The engineers interpreted this as mechanical simplicity and produced a machine with no gearbox and no clutch. It was virtually impossible to ride.
-A.
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2025
