AI Model Collapse

[Zinjanthropos]

A big problem with no answer at the moment. Ai authoring its own demise? Interesting problem described here :

https://witness.ai/blog/ai-model-collapse/

Potential topics?…..Could we relate this to human existence/experience? Has the human model already collapsed in a similar fashion?

[C C]

A big problem with no answer at the moment. Ai authoring its own demise? Interesting problem described here :

https://witness.ai/blog/ai-model-collapse/

Potential topics?…..Could we relate this to human existence/experience? Has the human model already collapsed in a similar fashion?

On the plus side, it mandates humans having to continue to be involved in knowledge and fact production (along with art and literature). However, students are substantially relying on AI to do their thinking and work for them, a dependence and of lack of proficiency which will carry on into their adult professional careers.

But similar has transpired for decades with the chosen morality narratives and eclectic ideology aggregations of the competing human media and the administrative part of the social sciences. Who use those presuppositions to interpret what's going on with respect to new events and research results (affects experiment setup of the latter, too). That circular process feeds such loaded evaluations back into the interpretative standards to produce yet another iteration of biased diagnosing, which likewise could potentially degrade over time. (Humans can still undergo trends and thought-orientations shifts, though, that shuffle things up unpredictably.)

https://witness.ai/blog/ai-model-collapse/

EXCERPTS: AI model collapse refers to the progressive degradation of an AI model’s performance caused by training on data that has been generated by other AI systems rather than original, human-generated content. Over time, this recursive process leads to a loss of data diversity, accuracy, and meaning...

[...] 1. Loss of Output Quality
As models absorb increasing amounts of synthetic or low-quality data, their ability to produce accurate, coherent, or creative text diminishes. This manifests as nonsensical sentences, factual inaccuracies, or repetitive phrasing — symptoms often described as semantic drift or output entropy.

2. Bias Amplification
Because AI-generated content often mirrors the biases and limitations of its source models, recursive training can reinforce existing distortions in data — from gender or racial bias to ideological skew — compounding societal and ethical risks.

3. Erosion of Real-World Alignment
AI systems trained primarily on AI outputs lose grounding in real-world data. This disconnect weakens the model’s ability to reason about current events, cultural context, and factual accuracy, making it less aligned with human expectations and truth.

4. Degraded Future Generations of Models
Model collapse doesn’t just affect a single system — it threatens the next generation of foundation models. As more models trained on synthetic data are used to generate new datasets, the entire AI ecosystem risks an information feedback loop, leading to systemic degradation across the industry.

5. Economic and Security Impacts
In industries like healthcare, finance, or cybersecurity, degraded model performance could lead to dangerous or costly decisions. Misdiagnosed conditions, inaccurate risk assessments, or false anomaly detections may result if AI models lose connection to reliable, human-validated data.

[Zinjanthropos]

But similar has transpired for decades with the chosen morality narratives and eclectic ideology aggregations of the competing human media and the administrative part of the social sciences.

I guess AI is capable of collapsing much faster. Two models, both collapsed, and the future of the world dependent on them. If all this is known then why can’t we and AI just do the opposite of what the models tell us? Could it be argued that this is the difference between being right or left? Either go with what people and machines are telling you or think about it first?