This morning when I was opening windows to try to cool a Minergie building that was designed for cold winters that no longer exist I had a thought. That thought was that AI is nothing but a glorified photocopier.
What I mean by this is that at the moment AI companies are buying tens of thousands, if not millions of books, removing the spine, scanning the books and then pulping the remains. In so doing Ai companies are hoovering enormous amounts of data and destroying books that could be donated to libraries, schools, universities and more.
That they do this with books is one thing, but it’s likely that they are doing this with other intellectual property, such as CDs, DVDs and more. Remember, everything has a “no unauthorised copying, lending” and more warning, and yet AI is doing this on an industrial scale.
Ai has no comprehension of a conversation, or anything else. If I say “I want to run to meet new people and it’s hard to choose which event will have more return on investment” AI will give coaching advice that has no link to what requires emotional intelligence, rather than guesswork.
Ai is not even guessing, it’s pattern recognition. If “this” then “that”, but without understanding what this, and that is.
I saw a headline or two that said that young people hate AI generated crap, and that it’s a turn off. I agree with them. There is a very simple reason for it. It’s formulaic, often kitsch, and more than anything else, it’s empty of depth. I see groups use AI to generate “What, when, where, why, how” posters and they all look the same.
No human is an island, but neither is. Everything is derivative of what has come before it but with AI it might be great at imitating an artist, but it will be a while before it generates its own. The Gimli situation is a key example. People like the style and asked AI to gimlify an image. It works well but it is disingenuous. It is a filter, rather than an artistic rendering. It has no creative process, but rather an algorithmic one.
When AI generates an image, or video it is rendering an image from hundreds, if not thens of thousands of images that it has ingested so it is “stealing” from hundreds of people at once. It took data, tokenised it, and then returns those tokens as a derivative “photocopy” of sorts.
Since AI is a probability engine it does not see, hear, or understand what is being asked to do, or what it is doing. If I say “Draw a duck riding a bike it finds ducks, and bikes, but it doesn’t look at Donald Duck comics for ducks riding bikes, and generate something realistic. That’s because it doesn’t imagine such a scene. It sees a human bike shape, and renders that, but it doesn’t render duck feet. It doesn’t render human hands or limbs well.
AI can’t even, or couldn’t, recognise typographic text, and that is very simple to see, for humans, but not so easy to reflect. That’s why handwriting training takes weeks, months, or even years, and why we start with pencils, before moving to pens.
In the case of AI plagiarising it is clear when the style for Gimli, Disney or others is copied, but with photorealistic images the challenge is greater because an AI generated photorealistic image might have used hundreds, or thousands of images to render so it is hard to determine copyright ownership.
One of the key issues with AI when used to generate photos, books, blog posts and more is that it floods the market with even more noise to filter through to get to read something new and interesting.
It might take 6 prompts to write a novel now and a few minutes but books take hours to read, so we’re generating in seconds something that takes hours to read. AI generating podcasts have no value, for humans, because they’re slow and time consuming to get through.
Anything generated by AI, in Europe needs to be labelled as such, in order for people to avoid it. It’s not about stigma, it’s about humanism.
Two decades ago we worried about content farms creating collections of blog posts to game search engines. Today the problem is amplified by AI because AI can generate blog posts, podcasts, books, images and more and saturate social networks automatically.
Linkedin allows people to flag AI content and more networks should do the same. Human generated content has value. AI generated content doesn’t.
As I have said before, if you use AI to generate content, then, with a similar premise I could generate the same content without using a middle person. That’s why AI should be used for processing big data, but not for generating content for human beings.
I want original content rather than the AI equivalent of an algorithmically altered photocopy.
Leave a Reply