A framework to commercialise (or not) GAN-based outcomes
About trust, creative expression, and AI Ethics
Imagine you are at this spanking new restaurant in the city. You made it on Day 2. The reservation system timed out. There was so much build-up into the opening.
You are ushered in. The ambience is outstanding, ethereal. The decor — abstract and Van Gogh-esque.
The menu items are classy, but the range is limited — kept intentionally so, as you have been told.
The meal is brilliant. You had ordered a Greek-Vietnamese fusion with an egg-noodle base. The fragrance of oregano, mint, and lemongrass was a perfect blend.
The chef checks in with you mid-way. You tell him it is great!
You write a great review online.
A week later, you hear that this restaurant 3D-printed the noodles and the mint leaves using natural ingredients. The restaurant justifies that the menu did have a footnote that said the chef exercises the right to use creative freedom to enhance the customer experience. (The decor was a clue: got it?)
Are they right in their stance?
Do they appear less credible? or the same?
I have the same questions about GAN-based creative products. The creative products may be images, fonts, 3D objects, portraits or videos.
GAN or Generative Adversarial Networks (thanks for the link Jason Brownlee) leverage deep learning techniques to generate realistic entities/outcomes across a range of use cases. They do need to be told about the problem context they need to operate in; and input entities (photos, videos, fonts, vectorised language).
GAN-based outcomes are already realistic, but will get even better. They would reach a stage where humans are unable to distinguish the real from the real.
The questions then are, as a creator do I have an obligation to inform my customers about the creative process; and as a customer do I have a right to demand additional information?
There is enough said and done about being fully aware and conscious of the source and processes of contents that satisfy one of our senses (what we ingest as food and drink).
Why would we not extend the same logic to what we see, hear and touch?
While creators are fully entitled to their expression, customers are fully entitled to the information as well.
Here is a potential framework that I have used to wrap my head around GAN-based entities.
On the X-axis is the rather discrete choice for the creator; was the entity created with an intent of purely creative expression or with an intent to monetise?
On the Y-axis is a continuous spectrum of how real the input variables are/have been (100% real to 0% real)
The framework proposes four actions based on the nature of inputs entities and creator’s intent.
Educate: If the creator has chosen to commercialise his/her/their creation; and the base inputs are real places, people, objects, entities they have the responsibility to ensure that the customer has understood the process of creation and what they are seeing is a worked-on version of a real input. The customer then has the choice.
Inform: If the creators have chosen to commercialise their creation; and the base inputs are fully imaginary or an n-th generation of a GAN-based image, the customer needs to be informed of the nature of reality — meaning is the input entity a first-generation GAN product or second or Nth (if you know what I mean)
Evolve: There is no greater joy than pure creative expression. The freedom of experimentation and its impact on brain chemicals and human well-being is immensely positive. I am choosing the word “responsibility” deliberately — Art creates Art and like any creative movement, needs to be evolved and built on.
Enjoy: This usually is Step 1 for the beginner, isn't it? This is where the choices of whether to monetise or use entities that are not real and made. This is place where one needs to enjoy one’s creation, show it off (if you like), and feel proud.
Dear reader: I am keen to know what you think
