3 minute read

Dali or DALL-E? Questioning the true creativity of AI

AI can talk back to you, but did you know it can create unique artworks too?

GARY HENDERSON

,

director

of IT at Millfield School and ANME ambassador, follows up his ChatGPT insight with a look at AI and creativity

Artificial Intelligence (AI) has been in the news a lot in 2023, with ChatGPT regularly hitting the headlines - but ChatGPT isn’t the only AI model out there. While ChatGPT deals with providing natural language responses to user prompts, other AI solutions, such as DALL-E, provide visual responses to a prompt. Solutions like DALL-E allow users to give a textual description of a desired image, from which the AI creates an original artwork based on its image training data.

I have seen several people raise concerns about AI solutions’ creativity, or lack thereof. ChatGPT, for example, was described as tending towards providing ‘average’ responses. The dictionary definition of ‘creativity’ is, ‘The use of imagination or original ideas to create something’ so it seems clear that AI solutions cannot be considered creative. Their responses are based on the ingested training data and probabilistic modelling of content, so they can hardly be regarded as original.

HUMAN CREATIVITY?

But what is human creativity based on? Or does it just come from the ether?

My understanding is that a lot of what we, as humans, create is based on personal experiences, including past events, upbringing, experiences and more; creating something

‘original’ doesn’t just come from nothing. The challenge is that - unlike AI models which have an accessible record of all the training data provided - we often cannot access the training data we, as humans, have experienced in our creativity. We don’t consciously remember many of our previous experiences, and we may suffer unconscious biases which play into our creativity.

SO, ARE WE THAT MUCH DIFFERENT FROM AN AI?

Maybe we aren’t. ChatGPT displays a notice stating the risk of bias in its responses, bias that comes from the training data it was provided with. Again, how is this so very different from the decision-making of a human, where bias - often unconscious - may play a part? So maybe AIs are more creative than we think, or perhaps we just think we humans are more creative than we are, at least based on the above definition of creativity.

Homogenisation

I mentioned the ‘average’ nature of ChatGPT responses earlier, hinting at another challenge associated with AI-based outputs. If we are looking at the creation of images, AI solutions will tend, through their probabilistic nature, towards that which is common. Therefore, if AI is heavily relied upon, this runs the risk that all outputs will tend towards this average and away from art which contains the uniqueness associated with human-based creations. We may start seeing art losing some of the randomness and individuality which is critical to creativity.

Copyright And Attribution

Another issue when looking at the products generated from AI models is copyright and attribution. DALL-E, for example, will use image data it was trained with to generate its response - image data that has been ‘scraped’ from the internet; this, therefore, includes copyrighted materials. If the AI uses a copyrighted image in generating its output - where this is clearly without the permission of the copyright holder - is this not a copyright infringement?

And then there is the issue of attribution; to whom do we attribute a piece of AI-generated artwork based on a user prompt and a set of training data? Does it belong to the user, to the AI, or to its creators? Or does it, perhaps, belong collectively to the individuals who created each piece of training data? How do we even know which parts of the training data the AI used, given that AI models are generally black box solutions where, although we can see the input (the prompt) and the output (the response), we are unable to see the process with which the AI creates its response?

SO, WHAT DOES ALL THIS MEAN?

AI is here to stay, so we must make the most of it. I believe that AI has flaws and limitations, but so do we humans. The key potential lies in humans and AIs working together, using the benefits of each to reduce our collective limitations. AIs can quickly produce images based on a prompt, presenting different options far more quickly than humans. This can save the human artist time, focusing on adding the human element, randomness and, dare I say it, ‘creativity.’

As to the challenges, we will need to work through them. We must decide whether AI-based content is attributable to the user proving the prompt, or the AI. We will need to consider the ethics around the use of AIs which have ingested copyrighted works and then use them in their outputs. We will need to think about the risk of artistic output, be it written, visual or even audio content, being stripped of the uniqueness, randomness and originality which we believe is a key part of human-based creations.

These are interesting times!

This article is from: