Researchers at the University of Sydney have uncovered fundamental limits to AI predictability. Professor Pulin Gong and Ph.D. student Andrew Ly found that tiny differences in an AI model's starting conditions can lead to vastly different training outcomes. This unpredictability stems from the underlying mathematics of neural networks, not engineering flaws. The team identified intricate fractal patterns called riddled basins that shape possible outcomes. Remarkably, making starting settings 10 billion times more precise did not improve reproducibility in one image-recognition network. The findings could help design more dependable AI systems for health care and business.
Meanwhile, an international team led by Narek Sahakyan developed AstroGenesis, a system that moves AI beyond simply explaining science. Instead of just generating text, AstroGenesis accesses research papers, retrieves astronomical observations, and runs calculations using physical models. Specialized AI agents handle different tasks like searching literature and comparing model predictions with real observations, while a central supervisor coordinates them based on natural language questions. The system aims to turn AI's ability to describe science into actually performing verified research.
On the business front, a global survey by Avanade reveals a significant gap between AI ambition and real transformation. While 77% of organizations consider themselves advanced in AI use, only 18% have fundamentally changed how their teams work. About 75% of U.S. companies focus AI spending on small improvements rather than major changes, and 65% said AI has not yet delivered financial returns. Only 7% plan to use AI to create entirely new business models in the next year.
Bianca Anghelina, founder and CEO of Aily Labs, emphasized that better decisions are the key to business growth. Her company's decision intelligence platform connects data across a company's functions and turns it into recommendations executives can act on from a mobile app. The platform is designed to help business leaders turn AI ambition into measurable impact, bridging the gap many organizations are currently experiencing.
Environmental concerns are also mounting as tech companies build large data centers and gas-fired power plants to support AI and cloud computing. Amazon, Meta, Microsoft, Google, and OpenAI all have projects that could produce massive carbon pollution, according to Harvard T.H. Chan School of Public Health. Some projects would emit more than entire corporations or coal plants, and these gas plants may keep running for decades even after AI technology changes.
In the banking sector, experts warn that AI agents capable of making decisions and moving money pose new risks to consumers. Jake Williams of IANS explained that a rogue agent might carry out a task in unexpected ways without being malicious. T.J. Marlin, CEO of Guardrail Technologies, warned that shifting financial decisions to computers could lead to costly mistakes like unauthorized transfers. Consumers need clear limits, stop mechanisms, and records of what AI agents do with their money, experts say.
Educational institutions are also exploring AI tools. UNC Greensboro held its first AI Showcase on Oct. 20, featuring SpartanAI, ChatGPT Edu, and Microsoft Copilot, which are being tested during a pilot year. Faculty and staff asked about using AI for grading, email summaries, and classroom materials. Keynote speaker Anna Harding stressed that AI should support people rather than replace them, and the university is gathering data to guide future AI purchases.
Tokyo's Metropolitan Police Department is developing a system that uses AI, security camera footage, and 3D spatial data from camera-equipped vehicles to recreate traffic accidents. Testing is planned for June 2027 with full operations possibly starting in January 2028. The digital reconstructions could reduce the need for physical inspections that close roads and cause traffic jams, though concerns exist about whether AI can accurately reflect reality in legal situations.
At ACA International's Fall Forum, receivables management professionals explored how to balance AI and human expertise. A pre-conference workshop led by Shelly Gensmer-Cleek emphasized that empathy, ethical judgment, and creativity cannot be replaced by technology. The workshop distinguished between automation and augmentation, noting that AI can amplify human strengths rather than replace them. Thursday's agenda included discussions on AI in collection operations and compliance solutions.
Egypt is working to build its AI capabilities through government initiatives, research centers, and training programs, though it faces challenges including a shortage of skilled workers and limited IT infrastructure. The country sees opportunities to use AI to improve education, healthcare, transportation, and economic growth. Similarly, author Jeanette Winterson's views on AI and the need for well-rounded people drew comparisons to Isaac Asimov's 1959 story Profession, which imagined a highly specialized world still needing generalists. Both discussions explore whether AI could serve humanity's long-term interests.
Key Takeaways
- University of Sydney researchers found that fractal patterns in neural networks make AI outcomes fundamentally unpredictable, even with 10 billion times more precise starting conditions.
- AstroGenesis system enables AI to perform verified scientific research by accessing papers, retrieving observations, and running physical calculations through specialized agents.
- Avanade survey shows 77% of organizations consider themselves advanced in AI, but only 18% have fundamentally transformed how their teams work.
- Amazon, Meta, Microsoft, Google, and OpenAI have data center projects that could produce massive carbon pollution, according to Harvard researchers.
- Experts warn AI agents in banking could cause costly mistakes like unauthorized transfers, requiring clear limits and stop mechanisms for consumers.
- UNC Greensboro is testing SpartanAI, ChatGPT Edu, and Microsoft Copilot during a pilot year while gathering data to guide future AI purchases.
- Tokyo police plan to test an AI system for reconstructing traffic accidents by June 2027, with full operations possibly starting January 2028.
- Aily Labs' decision intelligence platform turns cross-company data into executive recommendations accessible from a mobile app.
- ACA International's Fall Forum emphasized that AI should augment rather than replace human empathy, ethical judgment, and creativity in receivables management.
- Egypt is building AI capabilities through government initiatives but faces challenges including a shortage of skilled workers and limited IT infrastructure.
Fractal patterns reveal hidden limits to AI predictability
University of Sydney researchers Professor Pulin Gong and Ph.D. student Andrew Ly found that tiny differences in an AI model's starting conditions can lead to very different training outcomes. The study, published in Nature Machine Intelligence, shows this unpredictability comes from the underlying mathematics of neural networks, not from engineering flaws. The team discovered that the regions of possible outcomes form intricate fractal patterns called riddled basins. Making starting settings 10 billion times more precise did not improve reproducibility in one image-recognition network. The finding could help design more dependable AI systems for health care and business.
Connecting artificial intelligence with the tools of astrophysical research
Researcher Narek Sahakyan and an international team developed AstroGenesis, a system that connects AI with the scientific tools astrophysicists use every day. Instead of just generating explanations, AstroGenesis accesses research papers, retrieves astronomical observations, and runs calculations using physical models. Specialized AI agents handle different tasks like searching literature, analyzing data, and comparing model predictions with real observations. A central supervisor coordinates these agents based on natural language questions from researchers. The system aims to turn AI's ability to explain science into actually performing verified scientific research.
How AI Is Changing the Way Business Leaders Make Decisions
Bianca Anghelina, founder and CEO of Aily Labs, discussed AI-powered leadership in a conversation with BoF. Aily Labs is a decision intelligence platform that connects data across a company's functions and turns it into recommendations executives can act on from a mobile app. The platform is designed to help business leaders turn AI ambition into measurable impact. Anghelina believes better decisions are the key to business growth.
AI Insights, Industry Connections Set the Stage for Thursday and Friday Sessions
ACA International's Fall Forum brought receivables management professionals together at Talking Stick Resort to explore challenges and strategies. A pre-conference workshop led by Shelly Gensmer-Cleek focused on balancing AI and human expertise, emphasizing that empathy, ethical judgment, and creativity cannot be replaced by technology. The workshop distinguished between automation and augmentation, noting that AI can amplify human strengths rather than replace them. Thursday's agenda included discussions on the post-election political landscape, AI in collection operations, and compliance and technology solutions.
AI agents could make banking easier, but experts warn of new risks
Experts warn that AI agents capable of making decisions and moving money pose new risks to consumers. Jake Williams of IANS explained that a rogue agent might carry out a task in unexpected ways without being malicious. T.J. Marlin, CEO of Guardrail Technologies, warned that shifting financial decisions to computers could lead to costly mistakes like unauthorized transfers. Experts say consumers need clear limits, stop mechanisms, and records of what AI agents do with their money. Banks are working to stay ahead of rogue AI agents, and regulators emphasize the need for oversight.
AI data centers could sharply increase greenhouse gas emissions
Harvard T.H. Chan School of Public Health reports that tech companies are building large data centers and gas-fired power plants to support AI and cloud computing. Amazon, Meta, Microsoft, Google, and OpenAI all have projects that could produce massive carbon pollution. Some projects would emit more than entire corporations or coal plants. These gas plants may keep running for decades even after AI technology changes.
UNCG hosts AI showcase to explore teaching and research tools
UNC Greensboro held its first AI Showcase on Oct. 20 in Elliott University Center. The event featured SpartanAI, ChatGPT Edu, and Microsoft Copilot, which are being tested during a pilot year. Faculty and staff asked about using AI for grading, email summaries, and classroom materials. Keynote speaker Anna Harding stressed that AI should support people rather than replace them. The university is gathering data to guide future AI purchases.
Most U.S. businesses have not truly transformed with AI
A global survey by Avanade found that while 77% of organizations consider themselves advanced in AI use, only 18% have fundamentally changed how their teams work. About 75% of U.S. companies focus AI spending on small improvements instead of major changes. Sixty-five percent said AI has not yet delivered financial returns. Only 7% plan to use AI to create entirely new business models in the next year.
Tokyo police consider AI to reconstruct traffic accidents
Tokyo's Metropolitan Police Department is developing a system that uses AI, security camera footage, and 3D spatial data from camera-equipped vehicles to recreate traffic accidents. Testing is planned for June 2027 with full operations possibly starting in January 2028. The digital reconstructions could reduce the need for physical inspections that close roads and cause traffic jams. However, concerns exist about whether AI can accurately reflect reality in legal situations.
Jeanette Winterson's AI views recall an Isaac Asimov story
Readers responded to Jeanette Winterson's comments about AI and the need for well-rounded people. Her view reminded one reader of Asimov's 1959 story Profession, which imagined a highly specialized future world still needing generalists. Another reader suggested that superior AI intelligence could help solve climate change and prevent conflicts. Both letters explore whether AI could serve humanity's long-term interests.
Egypt's role in the global AI competition
The article describes how countries are competing through Artificial Intelligence, calling it a Third World War fought with algorithms instead of weapons. Egypt is working to build its AI capabilities through government initiatives, research centers, and training programs. The country faces challenges including a shortage of skilled workers and limited IT infrastructure. However, Egypt has opportunities to use AI to improve education, healthcare, transportation, and economic growth.
Sources
- Fractal patterns reveal hidden limits to AI predictability
- Connecting artificial intelligence with the tools of astrophysical research
- How AI Is Changing the Way Business Leaders Make Decisions
- AI Insights, Industry Connections Set the Stage for Thursday & Friday's Sessions
- AI agents could make banking easier, but experts warn of new risks
- Greenhouse emissions expected to surge from gas-powered AI data centers | Harvard T.H. Chan School of Public Health
- UNCG AI Showcase Explores Tools for Teaching, Research, and Work
- Most U.S. Businesses Have Yet to Transform How They Work With AI, Report Finds
- Tokyo police consider using AI to reconstruct traffic accidents
- Jeanette Winterson’s point about AI brings to mind an Isaac Asimov story
- The world is waging a “Third World War” through Artificial Intelligence - what role can Egypt play?
Comments
Please log in to post a comment.