From UPSC perspective, the following things are important :
Prelims level : Latest developments in AI
Mains level : ChatGPT, AI generative models, limitations and challenges
- 2022 had an unusual blue-ribbon winner for emerging digital artists; Jason Allen’s winning work Théâtre D’opéra Spatial was created with an AI Generative model called Midjourney.
Crack Prelims 2023! Talk to our Rankers
What is Midjourney?
- Midjourney is an AI based art generator that has been created to explore new mediums of thought.
- It is an interactive bot, which uses machine learning (ML) to create images based on texts. This AI system utilises the concepts and tries to convert them into visual reality.
- It is quite similar to other technologies such as DALL-E 2.
The journey of AI generative models so far
- Midjourney generator: Midjourney is one of the rash of AI-generated Transformer or Generative or Large Language Models (LLMs) which have exploded onto our world in the last few years.
- Earlier models: Models like BERT and Megatron (2019) were relatively small models, with up to 174 GB of dataset size, and passed under the collective public radar.
- Composition skills of GPT3: GPT3, released by OpenAI with a 570 GB dataset and 175bn parameters was the first one to capture the public consciousness with some amazing writing and composition skills.
- Models that creat images or videos based on texts: The real magic, however, started with Transformers which could create beautiful and realistic pieces of art with just a text prompt OpenAI’s DALL-E2, Google’s Imagen, the open-source Stable Diffusion and, obviously, Midjourney. Not to be left behind, Meta unleashed a transformer which could create videos from text prompts.
- ChatGPT, a latest and more evolved, like real communication: Recently in late 2022 came the transformer to rule them all ChatGPT built on GPT3, but with capabilities to have real conversations with human beings.
Are these models ethical?
- Ethics is too complex a subject to address in one short article. There are three big ethical questions on these models that humanity will have to address in short order.
- Environmental: Most of the bad rap goes to crypto and blockchain, but the cloud and these AI models running on it take enormous amounts of energy. Training a large transformer model just once would have CO2 emissions equivalent to 125 roundtrips from New York to Beijing. This cloud is the hundreds of data centres that dot our planet, and they guzzle water and power at alarming rates.
- Bias; as it do not understand meaning and its implications: The other thorny ethical issue is that sheer size does not guarantee diversity. Timnit Gebru was with Google when she co-wrote a seminal research paper calling these LLMs ‘stochastic parrots’, because, like parrots, they just repeated a senseless litany of words without understanding their meaning and implications.
- Plagiarism, question of who owns the original content: The third prickly ethical issue, which also prompted the artist backlash to Allen’s award-winning work is that of plagiarism. If Stable Diffusion or DALL-E 2 did all the work of scouring the web and combining multiple images (a Pablo Picasso Mona Lisa, for example), who owns it. Currently, OpenAI has ownership of all images created with DALL-E, and their business model is to allow paid users to have rights to reproduce, paint, sell and merchandise images they create. This is a legal minefield the US Copyrights office recently refused to grant a copyright to a piece created by a generative AI called Creativity Machine, but South Africa and Australia have recently announced that AI can be considered an inventor.
Do you know ChatGPT?
- ChatGPT is a chatbot built on a large-scale transformer-based language model that is trained on a diverse dataset of text and is capable of generating human-like responses to prompts.
- A conversation with ChatGPT is like talking to a computer, a smart one, which appears to have some semblance of human-like intelligence.
What are the other concerns?
- Besides the legal quagmire, there is a bigger fear: This kind of cheap, mass-produced art could put artists, photographers, and graphic designers out of their jobs.
- Machine does not have human like sense: A machine is not necessarily creating art, it is crunching and manipulating data and it has no idea or sense of what and why it is doing so.
- As it is cheap, corporate might consider using it at a large scale: But it can do so cheaply, and at scale. Corporate customers might seriously consider it for their creative, advertising, and other needs.
- Legal and political leaders across the world are sounding the alarm about the ethics of large generative models, and for good reason. As these models become increasingly powerful in the hands of Big Tech, with their unlimited budgets, brains and computing power, these issues of bias, environmental damage and plagiarism will become even more fraught. Such AI models should not be used to create chaos rather a harmonious existence.
Q. Name some of the models of AI based art generators. Discuss the ethical concerns of such models.
(Click) FREE 1-to-1 on-call Mentorship by IAS-IPS officers | Discuss doubts, strategy, sources, and more