I knew this day was coming…
Anthropic recently said that it would watermark text generated by its various models in efforts to comply with European regulations, specifically the EU AI Act’s Transparency Code. They are trying to address requirement from the Act that requires all AI companies to somehow identify AI-generated or edited content in a way that can be identified.
From the very first hallucinogen of the various LLMs (the best example is the AI that was convinced that it was Abraham Lincoln that chopped down the cherry tree, not George Washington), people, politicos and regulators have had concerns about the content created by AI not being identified as AI generated.
And with good reason. In many ways, AI has made us dumber: we have forgotten basic fundamentals in programming, writing and artistic design, all now substituted for whatever computer generated “slop” the various AI models can come up with (never mind that some of it wildly diverges from expected outcomes or is riddled with errors or omissions).
I recently saw an AI generated map of the Las Vegas Strip. If you are in the tech industry, chances are that you have been to a conference or two somewhere on the Strip, and have frequented the numerous casinos that can be found there. But for those that have been, you will see how incredibly incorrect this map actually is. Casinos that no longer exist, some casinos located on the wrong side of the street, some that just are incorrectly placed. And yet: is it good enough? No one was (probably) using it for land navigation. But following this guide, you would expect that the MGM Grand was located next to the Mandalay Bay (where Black Hat was just hosted). And you would be very wrong.
For the quasi-educated observer, this map was immediately identifiable as AI generated. But as AI improves, it gets tricker to tell the real images from the AI generated.
Which brings us to text generated by AI. There is plenty of anecdotal evidence showing that our writing skills are rapidly devolving as we use automation and AI for those tasks. Heck – as I write this, there are red lines ALL OVER THE PLACE for the words that I have typoed (it doesn’t like that word, BTW) and my poor grammar.
Should I have included a disclaimer that this blog used the continuous grammar and spell check feature while writing it?
Let’s take the next example: from time to time, I use several of the AI tools to do proof reading of my text. There are times that it finds things that need to be corrected. And you – the reader (yes, I used dashes here and it was not AI generated) – are likely thankful for that effort, as it makes things more presentable, if not readable. Do I / should I include a disclaimer that states this blog went through AI editing?
What about using AI for outline creation? What about some basic statistical analysis? What about searching if someone else wrote something similar? Yet if I do the same search manually using Google or your search engine of choice, it is somehow OK?
It is silly that regulators and those that seem to know what is better for you would tell you what tools you should use to make your lives easier. Should you be forced to use quill and inkbottles instead of a ballpoint pen? What about the now defunct writing quill industry? I’m sure they have an interest group. Somewhere…
I have a colleague that is an extremely literate and excellent writer (it is a major part of his life, both professionally and personally). His work is constantly flagged as being AI generated, and literally none of it is. Sometimes, mine is flagged, sometimes it isn’t. I guess it depends on how good of a writer I was at that particular moment.
I will continue to use AI to assist in my work. But not complete my work for me, as I personally believe that I add value to the work that I do. I would be foolish not to take advantage of all of the tools available to me.
My point: We (analysts, pundits, news media, and everyone else) will continue to debate this. But just like you probably use a calculator to complete long division or use that ball point pen to sign a check (written a lot of checks recently have you? Another one of those dying “technologies that we rarely think about), people will continue to utilize the tools that are available to them. And to somehow penalize those that do is ridiculous and practically impossible: those that want to find a work-around will likely do so, and those that don’t find a work around for having their paper edited by AI will somehow be ostracized.
______________
This blog was written (poorly) by a human, and not edited by AI (or really anyone). Spell checker was used (somewhat) and the grammar detector was also used (marginally).