[Paper Review] Can There be Art Without an Artist?

Avijit Ghosh, Genoveva Fossas|arXiv (Cornell University)|Sep 16, 2022

Art History and Market Analysis18 citations

TL;DR

This paper investigates whether art can exist without a human artist in the age of generative AI, arguing that while current AI models exploit artists through unauthorized training data and profit shifting, they can become a legitimate new artistic medium if deployed responsibly. The authors advocate for ethical AI integration via consent-based datasets, unlearning tools, and artist-centric legislation to preserve artistic integrity and equity.

ABSTRACT

Generative AI based art has proliferated in the past year, with increasingly impressive use cases from generating fake human faces to the creation of systems that can generate thousands of artistic images from text prompts - some of these images have even been "good" enough to win accolades from qualified judges. In this paper, we explore how Generative Models have impacted artistry, not only from a qualitative point of view, but also from an angle of exploitation of artists -- both via plagiarism, where models are trained on their artwork without permission, and via profit shifting, where profits in the art market have shifted from art creators to model owners. However, we posit that if deployed responsibly, AI generative models have the possibility of being a positive, new modality in art that does not displace or harm existing artists.

Motivation & Objective

• To examine the ethical, legal, and economic tensions arising from generative AI in digital art.
• To analyze how AI models are trained on unlicensed artist work, leading to exploitation and profit displacement.
• To investigate the potential for AI-generated art to become a distinct artistic modality without harming human artists.
• To propose regulatory and technical safeguards ensuring ethical AI use in artistic creation.

Proposed method

• Analyzing case studies of AI-generated art, including Jason Allen’s Midjourney-winning piece Théâtre D’opéra Spatial.
• Surveying the training practices of state-of-the-art vision models, including DALL·E, Stable Diffusion, and Midjourney, focusing on data sourcing and consent.
• Examining the use of datasets like MS-COCO, ImageNet, and CelebA for their lack of proper attribution and licensing compliance.
• Reviewing the phenomenon of style mimicry, where AI models clone artists’ styles and even logos without permission.
• Proposing technical solutions such as unlearning mechanisms to allow artists to revoke consent post-training.
• Recommending policy frameworks that prioritize individual artists over corporate model owners in AI regulation.

Experimental results

Research questions

• RQ1 Can art exist without a human artist in the context of generative AI, and what defines artistic value in such cases?
• RQ2 How do current generative AI models exploit artists through unauthorized training on their work?
• RQ3 To what extent has profit in the art market shifted from creators to model owners and infrastructure providers?
• RQ4 Can generative AI become a distinct artistic medium akin to photography or digital art, without displacing human artists?
• RQ5 What regulatory and technical measures can ensure ethical and equitable use of AI in artistic creation?

Key findings

• Generative AI models like Midjourney and DALL·E are trained on vast datasets scraped from the internet without artist consent, including images under restrictive licenses such as Creative Commons with attribution or non-commercial use.
• A significant portion—around 95%—of images in the MS-COCO dataset from 2017 were distributed under licenses requiring attribution or prohibiting commercial use, yet these were not properly honored in the dataset.
• AI models such as Stable Diffusion can replicate the styles of living artists with high fidelity, and some have even attempted to clone artists’ logos, raising concerns about identity theft and artistic mimicry.
• The art market has experienced a shift in profits from freelance artists to corporate owners of AI models, exacerbating economic inequity in the creative sector.
• The paper demonstrates that AI-generated art can become a unique artistic modality, similar to how photography and digital art were once perceived as threats but later integrated into the art world.
• The authors propose actionable solutions, including consent-based data collection, unlearning tools for model updates, and legislation that prioritizes individual artists over corporate interests.