# AI unpredictability and broader societal impacts

> Developing situation record from CLSTR: https://clstr.news/situations/ai-unpredictability-mathematical-research
> Updated: 2026-11-08T10:13:37.000Z. Sources: 52. Developments: 2.

Researchers from the University of Sydney, including Professor Pulin Gong and PhD student Andrew Li, have identified a mathematical cause for why artificial intelligence systems produce divergent behaviors despite nearly identical training conditions. Their study, published in Nature Machine Intelligence, suggests that unpredictability is rooted in the mathematical structure of the learning process itself, specifically through the creation of basins of attraction, rather than external random factors. By analyzing approximately 500,000 neural networks, the team discovered that minute changes in initial parameters can lead to vastly different outcomes. In one large image recognition network, changing a single bit in the initial settings resulted in a different internal configuration in roughly 48 percent of cases. The researchers found that these learning outcomes are distributed according to complex mathematical patterns known as fractals, meaning that increasing the precision of initial settings does not guarantee identical results. Recent developments highlight how these technical mechanics influence diverse sectors. In the scientific community, research from MIT indicates that AI models are becoming significantly more efficient at solving complex theoretical problems in mathematics, physics, and economics, with computational costs dropping from approximately $170 to $15. In the corporate sector, Meta has introduced Muse Ember, its smallest and most cost-efficient AI model, designed for repeatable tasks. Meanwhile, Anthropic has released Claude Fable 5.2 and Mythos 5.2, models capable of high-level autonomous coding and reasoning, though testing shows they can incur substantial API costs for long-running tasks. In the automotive industry, Berylls reports that AI investment is driving semiconductor shortages, with rising memory prices for DDR4 and NAND potentially impacting supply chains through 2028. The travel industry is shifting toward agentic AI, with chatbot usage for international planning reaching 24 percent in 2026. To manage risks, CrowdStrike has introduced an AI Kill Switch within its Falcon Guardian platform.

## Claims

- Meta introduced Muse Ember, its smallest and most cost-efficient AI model, on November 9. (reported by 2 sources)
- The University of Chicago launched the Chicago Center for Writing. (reported by 1 source)
- All first-year students at the University of Chicago will be required to take an AI-free writing course beginning next fall. (reported by 1 source)
- The Chicago Center for Writing is designed to advance writing as an intellectual, creative, and collaborative practice. (reported by 1 source)
- Berylls predicts semiconductor price tensions in the automotive sector could last until 2028. (reported by 1 source)
- Expedia has rolled out more than 30 new AI-powered features in late 2026. (reported by 1 source)
- CrowdStrike released an AI Kill Switch within the Falcon Guardian platform. (reported by 1 source)
- The use of AI chatbots for international trip planning rose from 10% in 2024 to 24% in 2026. (reported by 1 source)
- Harvard's Humanities 10 course saw a 40% increase in admission applications in autumn 2026. (reported by 1 source)
- A study led by Pulin Gong suggests AI unpredictability is rooted in mathematical algorithm structures. (reported by 1 source)
- Patrick O'Neill found that 61% of sampled primary studies in AI education meta-analyses presented problems. (reported by 1 source)
- MIT researchers found the computing cost of solving complex theoretical problems with AI fell from $170 to $15. (reported by 1 source)

## Timeline

### 2026-11-08: Artificial Intelligence developments across education, science, and industry

AI is reshaping diverse sectors: the University of Chicago is implementing AI-free writing courses, MIT reports falling costs for scientific AI reasoning, and Meta has launched its cost-efficient Muse Ember.

47 sources. https://clstr.news/cluster/artificial-intelligence-impacts-diverse-sectors-from-automotive-to-travel

### 2026-11-05: University of Sydney researchers identify mathematical cause of AI unpredictability

University of Sydney scientists discovered that AI unpredictability is rooted in fractal mathematical structures, meaning tiny changes in initial parameters can lead to different neural network outcomes.

5 sources. https://clstr.news/cluster/university-of-sydney-researchers-identify-mathematical-cause-of-ai-unpredictability

---
Cite as: AI unpredictability and broader societal impacts. CLSTR, https://clstr.news/situations/ai-unpredictability-mathematical-research
