Think Faster This Time! Are We Ready For The AI-Plaque Tsunami?
Take a moment to remember this — never again in your lifetime will you live on a planet where all the content you see, read, and hear is ‘made by human’. As of February 2023 that phase of your life is over. This month marks a fundamental shift from just a month ago or especially this time last year.
Back then everything on the web, in the media you consumed, and in your daily digital interactions was ultimately written or vetted by a human, with a few exceptions. But now that’s no longer the case and the effects of this step change are being amplified quickly and may result in AI-plaque poisoning the digital well of human knowledge and communication.
We’ve faced similar transitions before so by now we should be aware of the potential pitfalls, know what to watch out for and proactively demand big and small tech do a better job this time because one way or another, this will impact the rest of our lives.
First up, this phenomenon isn’t just because everyone is suddenly using ChatGPT to write everything. That exponentially increases their productivity and so on its own that’s a good thing! This risk is a level below that where the underlying data consumed by these AI helpers is already no longer natively human and less so as each day passes. That’s because the exploding amount of AI-generated content will be fed back into AI content generators and then regenerated over and over recursively, every day, for the rest of your life.
Using tweets as an example, one AI bot generates a tweet based on a trigger event and then another bot takes a bunch of such tweets as inputs when generating the next message, and so on. These AI-generated inputs and outputs accumulate over generations and like Theseus’ Paradox where a ship is replaced plank by plank until the original has disappeared, the human inputs that started it will fade to nothingness too. And that new AI-generated content will become the corpus of data our internet is based on.
Is this actually a problem? Machines have been writing tweets and specialised texts for many years already, but the magnitude, quality and scope of automated writing has just transitioned to a new level. It’s surpassed industrial level already and has gone nuclear, so the proportion of content based on a machine-only input will skyrocket while the human-led inputs that got us this far will soon diminish and be relegated to distant memory.
We’ve already seen alarming or amusing cases of ChatGPT and Google’s Bard generating content which is uncomfortably weird, wrong or even dystopian. Many of those reported examples are arguably just teething problems and improvements to the biggest AI performance issues will be made quickly. However, the point where the AI outputs consistently feel wrong or are actually dysfunctional for humans is when we start suffering from AI-plaque. And like plaque on your teeth or arteries, by the time the impact is obvious you’re already far down the track.
A useful comparison for this new reality is a previous turning point that also occurred in your lifetime and see how that turned out. A major inflection point which you may not have noticed is Google launching the autocomplete search function 15 years ago. This feature predicts what you’re searching for based on the first few words you type and it was a game changer. It’s a massive time saver, helps in discovery and leads to better outcomes from your search task. We would all notice if this feature disappeared and we miss it on those apps and sites that don’t have it.
Unfortunately for such an impactful yet seemingly simple feature, when it doesn’t work it’s not only annoying but even leads to bad outcomes. For example, Amazon seems to have a non-functional autocomplete which consistently gives product options unrelated to my actual intent, so much so that I often flip to Google to zero in on the right search vein in a matter of seconds and from there get redirected back to the right Amazon product. That’s absurd but proves that a good idea isn’t guaranteed to work — execution really matters.
More broadly it’s also bad for diversity as predictive search pushes more and more people into the same lane of narrow options which gives incumbents a huge advantage and adds a layer of difficulty if your query is about a niche product or topic. And it’s really annoying or downright useless if the predictions simply aren’t what you want.
So while predictive search is both good and bad depending on how you use it and how it’s implemented, it’s simply so darn useful and saves so much time that you can’t turn back time to when we lived without it. Even if it influences us in unexpected or unseen ways we just need to adapt and manage that.
And here we are today with AI-generated content. Like the game of Chinese Whispers or Telephone where each time a message is passed along it becomes distorted or changed as the output from each person is based on their perception and interpretation of the input from the person before them, AI will do the same except that interpretation isn’t being done by a human trying their best to make sense of things for the next person. So like badly implemented autocomplete this could be done poorly and leave us drowning in an ocean of data garbage.
How fast AI-plaque becomes an issue in our brave new world starting in February 2023 depends on big tech and the digital industry for whom this is a lucrative game-changer. But the responsibility will come back to us very quickly to recognise the signs of plaque which inhibit our communication and understanding of the world around us and the digital world we unavoidably live in. And as the digital world is essential for our lives today it really matters when something gets in the way.
If and when AI-plaque becomes a problem we need to make proactive individual choices which can coalesce quickly at a global community level to incentivise industry to clean up whatever mess has manifested. And we must do so faster than previous messes we’ve lived through already in the digital era from Cambridge Analytica to massive data breaches.
Ultimately I’m hopeful. Although AI can mess this up quickly, humans can fix it fast too if we recognise the problems and act together this time round. Good luck with that!
Before you ask…
Yes, this article is original human work and was not written by ChatGPT.
No, ironically AI couldn’t produce a suitable non-dystopian photo for this article. So you’re getting my reinterpretation of Rodolfo Clix’s “woman biting gray nails”. Ouch.
