Model Collapse, also known as AI collapse or Habsburg AI, refers to a phenomena where machine learning models gradually degrade due to errors coming from uncurated training on synthetic data, meaning the outputs of another model including prior versions of itself.[1][2][3]
Shumailov et al. [1] coined the term and described two specific stages to the degradation: early model collapse and late model collapse. In early model collapse the model begins losing information about the tails of the distribution – mostly affecting minority data. Later work highlighted that early model collapse is hard to notice, since overall performance may appear to improve, while the model loses performance on minority data.[4] In the late model collapse model loses a significant proportion of its performance, confusing concepts and losing most of its variance.
Why does Model Collapse happen?[edit]
Synthetic data, although theoretically indistinguishable from real data, is almost always biased, inaccurate, not well representative of the real data, harmful, or presented out-of-context.[5][6] Using such data as training data leads to issues with quality and reliability of the trained model.[7][8]
Model Collapse occurs for three main reasons – functional approximation errors, sampling errors, and learning errors [1]. Importantly, it happens in even the simplest of models, where not all of the error sources are present. In more complex models the errors oftentimes compound, leading to faster collapse.
![](https://upload.wikimedia.org/wikipedia/commons/thumb/e/ef/Model_Collapse_in_Generative_Models_Can_Be_Avoided_By_Accumulating_Data.png/220px-Model_Collapse_in_Generative_Models_Can_Be_Avoided_By_Accumulating_Data.png)
Is Model Collapse inevitable?[edit]
From even simplest models it becomes clear that model collapse is not inevitable. For example in the gaussian model[1], a superlinearly increasing amount of data can bound is needed. Later work [9] highlighted that it can also be bounded in some settings, yet comes with a significant training cost – requiring accumulating and tracking data over time.
Alternative branch of literature investigates use of machine learning detectors and watermarking to identify model generated data and filter it out.[10]
Impact on large language models[edit]
In the context of large language models, research found that training LLMs on predecessor-generated text—language models are trained on the synthetic data produced by previous models—causes a consistent decrease in the lexical, syntactic, and semantic diversity of the model outputs through successive iterations, notably remarkable for tasks demanding high levels of creativity.[11]
Data poisoning for artists[edit]
Data poisoning is a form of Adversarial machine learning where the data of an image or text is altered so it cannot be trained on accurately by a training model. There are two main types of data poisoning, defensive, where an image's data is alter to protect the integrity of the work by preventing copying and look-alikes, and offensive, where an image is altered to reduce the reliability of generative artificial intelligent image generation.[12] However, it is unknown how much data poisoning affects training data and generative artificial intelligence on a large scale.
References[edit]
- ^ a b c d Shumailov, Ilia; Shumaylov, Zakhar; Zhao, Yiren; Gal, Yarin; Papernot, Nicolas; Anderson, Ross (2023-05-31). "The Curse of Recursion: Training on Generated Data Makes Models Forget". arXiv:2305.17493 [cs.LG].
- ^ Ozsevim, Ilkhan (2023-06-20). "Research finds ChatGPT & Bard headed for 'Model Collapse'". Retrieved 2024-03-06.
- ^ Mok, Aaron. "A disturbing AI phenomenon could completely upend the internet as we know it". Business Insider. Retrieved 2024-03-06.
- ^ Wyllie, Sierra; Shumailov, Ilia; Papernot, Nicolas (2024-06-05). "Fairness Feedback Loops: Training on Synthetic Data Amplifies Bias". Proceedings of the 2024 ACM Conference on Fairness, Accountability, and Transparency. FAccT '24. New York, NY, USA: Association for Computing Machinery: 2113–2147. doi:10.1145/3630106.3659029. ISBN 979-8-4007-0450-5.
- ^ De Rosa, Micholas (May 31, 2024). "How the new version of ChatGPT generates hate and disinformation on command". CBC. Retrieved June 13, 2024.
- ^ Orland, Kyle (May 24, 2024). "Google's "AI Overview" can give false, misleading, and dangerous answers". arsTechinca. Retrieved June 13, 2024.
- ^ Alemohammad, Sina; Casco-Rodriguez, Josue; Luzi, Lorenzo; Humayun, Ahmed Imtiaz; Babaei, Hossein; LeJeune, Daniel; Siahkoohi, Ali; Baraniuk, Richard G. (July 4, 2023). "Self-Consuming Generative Models Go MAD". arXiv:2307.01850 [cs.LG].
- ^ Self-Consuming Generative Models Go MAD. The Twelfth International Conference on Learning Representations.
- ^ Gerstgrasser, Matthias; Schaeffer, Rylan; Dey, Apratim; Rafailov, Rafael; Sleight, Henry; Hughes, John; Korbak, Tomasz; Agrawal, Rajashree; Pai, Dhruv; Gromov, Andrey; Roberts, Daniel A.; Yang, Diyi; Donoho, David L.; Koyejo, Sanmi (2024-04-01). "Is Model Collapse Inevitable? Breaking the Curse of Recursion by Accumulating Real and Synthetic Data". arXiv:2404.01413 [cs.LG].
- ^ Kirchenbauer, John; Geiping, Jonas; Wen, Yuxin; Katz, Jonathan; Miers, Ian; Goldstein, Tom (2023-07-03). "A Watermark for Large Language Models". Proceedings of the 40th International Conference on Machine Learning. PMLR: 17061–17084.
- ^ Guo, Yanzhu; Shang, Guokan; Vazirgiannis, Michalis; Clavel, Chloé (2024-04-16). "The Curious Decline of Linguistic Diversity: Training Language Models on Synthetic Text". arXiv:2311.09807 [cs.CL].
- ^ The Nightshade Team. "What is Nightshade". Nightshade. University of Chicago. Retrieved June 13, 2024.