Cotton-wool for brAIns.
As soon as the first wintery drafts of December chilled their way into the Shire, the influenza nasties took full advantage of the vulnerabilities of my respiratory system. Pneumonia it turns out is not much fun. Probably the symptom that was most frustrating to bear however was the head fog. A lack of focus, an impeded ability to assimilate information and trouble interpreting clear meaning when reading anything more complex than the back of the cereal box.
Yesterday I was sent an article which transported me right back there…
The article was well formatted, clearly structured and the right length for a short read but for some reason the point of the feature was lost.
I had definitely read something, a few hundred words, I was open to understanding more on the topic and I was interested but I was really none the wiser. It seemed like the copy optimisation had got me to read it, the formatting had made me believe that there was something to learn on the horizon but ultimately I was left wanting. It was like eating that space food ice cream you can buy from the National Space Centre – something goes in, it’s vaguely familiar but ultimately unsatisfying and artificial… and artificial is exactly the point.
The content I was reading, it occurred to me, was clearly the work of Generative AI. It was intuitively obvious. I didn’t exactly know how I knew, but something in the sentence and article structure, the lack of human voice and a cold efficiency was telling me this was AI generated. Something designed, made for a specific goal, not written with real intent or creativity.
Anyone who’s seen stock photography knows exactly what I mean. The component parts are all in place, it’s well composed, and beautifully lit. The smiling group of young people are all real, they’re happy to be in that beautiful office and they’re really happy about the perfect angle of attack that growth arrow is taking. It’s genuine happiness too, why wouldn’t they be happy? Being paid to point and laugh isn’t exactly shovelling shite is it.
All of this is true and yet we know we’re looking at something staged.
Good photography; meaningful, crafted photography has soul. It is created with meaning, context and a human touch that reflects real human experience. Somehow the two are almost always immediately discernible.
That was always the case, I’d contend it just about still is, but the world is changing at a helluva rate.
AIn’t it uncanny?
At this point in time and at the current point in the development of Large Language Models (LLMs) and Generative AI tools, I feel like we humans, despite our computational deficiencies, still have good instincts and can somehow tell when something is un-edited Generative AI produced content once we get close-up. The words are there, they’re in the right order and the content is saying something about something, but somehow we find ourselves in a literary (and still – just about – visual) uncanny valley.
Don’t get me wrong, to my seventies-born-mind, we’re living in the future that eighties sci-fi promised. Generative AI capabilities are astounding and the pace of development, the daily updates leave me a little breathless. However, seeming almost human, isn’t quite enough. At this point in AI-volution, I can’t help but feel we’re in the bottom of the valley, it’s still all a bit unnerving once you get past the first glance.
Now, the frightening thing is, if we are at the bottom of the valley, then the only way is up and that means Skynet! Well, maybe not quite Skynet, but it does mean that as the technology and tools improve, they will become ever-more human and indiscernible – and when that happens then we have a brave new world. Let’s be clear, this is where Generative AI is headed and it will get there. Soon.
Something in the AIr…
So what happens when AI does get there? Well, that’s a government department in itself and for another billion articles and philosophical conversations.
Let’s just agree that at this point in time, seeing is already no longer believing, certainly not on first glance and without really analysing writing style or with a bit of pinch zoom… We can still intuitvely tell what we’re looking at when we delve a little deeper, but tell me honestly, did you immediately notice something wrong in the header image? Did you intuitively pick out the real photograph from the two AI generated images — or just scroll past without noticing something was ‘off’? Can you tell which one is the real image now you know there are a couple of ringers in there?
Freely available generative AI video came along in the last few weeks and AI imagery is already everywhere online and in the media. These technologies have recently generated entertainingly convincing images of the Pope in a Balenciaga jacket and deeply concerning deepfakes of Taylor Swift but this is clearly the tip of the AIceberg.
The potential for anyone with a GPT account to leverage the technology to generate content that could potentially destroy reputations, change the direction of political discourse or even influence elections is still only taking us down to the lapping waves – there are depths to this that we can’t yet imagine.
Optimistically however, Meta announced in early February that it’s working to create common technical standards for identifying AI content and that it will label AI-Generated images on Facebook, Instagram and Threads. OpenAI, which produces Chat GPT and Dall-e also announced that images generated in ChatGPT will include metadata to allow social platforms and content distributors to see when an image was generated by AI.
These threats are clearly on the global radar and the tech world and governments are starting to react to the AI Control Problem. Some may argue it’s already too late but for me we’re not out of Uncanny Valley yet. I just hope that before Generative AI sprouts feathers and soars out of the valley, that an organised global consensus actions a sensibly managed wing-clipping. We humans are famous for our ability to gain and act upon consensus, right?
Maybe it’s in the trAIning?
It’s common with nascent tools and technologies that the potential is outweighed by the current technological iteration and the limitations of the user.
It could be that most of us haven’t yet begun to understand, let alone master prompt engineering to the extent that our commands produce results that would pass muster with the more discerning reader or viewer past an initial glance.
You’ve read or seen iRobot, right? What it, and the University of Oxford paperclip thought experiment illustrate are that AI systems don’t think like us. These new tools are designed to achieve a specified objective, not necessarily what is intended by the user of the tool.
Sci-Fi and in current debate points to bad AI design, uncontrolled AI tools or poor/ill-intentioned human users having the potential to cause disastrous results. This would not emerge from any ‘evil AIs’ or ‘bad robots’, but through optimisation of performance.
On the much more prosaic subject where we started this — throwaway ‘content’ for social media, the stakes are much, much lower. We do however have to understand that, in simple terms, AI will do what you ask but not necessarily what you mean and I firmly believe that we haven’t developed the faculties for articulating ourselves clearly to our AI friends just yet.
At least, most of us haven’t and I suspect this was true of the author of the piece which triggered this article. They have a point of view worth hearing and a skill-set and experience that command respect. However we’re learning to use a new, still emerging tool, the tool is not yet mature, nor are our skills in using it.
As per a previous article on AI, I still believe current Generative AI will only get us 70% of the way there, the future it seems still needs us. For now.
Ready to get acquAInted?
If you’re curious about Generative AI training or how our content creation services can enhance your offer or if you just fancy a chat about the opportunities AI, LLMs and prompt engineering may present for your business, then drop us a line to studio@bulbstudios.com or pay us a visit.
Image Credits:
Photograph by Amir Seilsepour on Unsplash
GenAI images created using CoPilot Designer