Collaborative Intelligence Week in Review - 07Oct2024
Please note that beginning with this issue, the publication day of this newsletter changes from Friday to Monday and includes items published on the weekend.
Best business article(s) I read this week…
How Machine Learning Will Transform Supply Chain Management
The article discusses how ML can improve supply chain management by addressing traditional planning challenges such as inaccurate demand forecasts and inefficient inventory optimization. It introduces a new approach called Optimal Machine Learning (OML) that uses historical data and supply chain inputs to make decisions rather than just improving forecasts. The OML model combines a decision-support engine, a digital twin, and an end-to-end data architecture to improve agility and resilience in supply chains. The authors argue that this approach can help companies reduce costs, increase efficiency, and enhance customer satisfaction by optimizing inventory and managing disruptions more effectively.
Best technical article(s) I read this week…
Intelligence at the Edge of Chaos
This article may not appeal to many. It appealed to me because I have extensively researched the application of chaos and complexity theory to strategic management. The paper describes how complex rule-based systems can foster the emergence of intelligent behavior in artificial models, specifically through training large language models (LLMs) on elementary cellular automata (ECA). The study shows that as the complexity of the ECA rules increases, the models exhibit better performance on reasoning and prediction tasks, suggesting that complexity is essential for developing intelligence. However, overly chaotic systems hinder performance, highlighting a "sweet spot" of complexity—referred to as the "edge of chaos"—most conducive to intelligence. The findings indicate that exposing models to structured yet unpredictable systems can help generate sophisticated, generalizable behaviors without requiring inherently intelligent data.
Other items I found valuable/interesting…
Microsoft’s AI Story Is Getting Complicated
This WSJ article describes the challenges Microsoft faces in maintaining its lead in AI. For example, Microsoft spent $55.7 billion (23% of revenue) in the recently ended fiscal year on capital investments and leases, much of which supported AI. This is up from 14% of revenue just five years ago. This level of expenditure affects other financial measures, such as free cash flow (FCF). Investors and analysts are trying to gauge the long-term impact of increased investment on the company’s financial performance.
AI assistants are ratting you out for badmouthing your coworkers
I included this as a “buyer beware” piece. It describes how AI assistants capturing and transcribing online meetings continue to record the meeting as long as it is active. It describes several instances where people who had left a meeting got a transcript that included co-workers or prospective business partners bad-mouthing them after they left. Remember: your AI assistant hears and records everything.
Why the OpenAI-to-Anthropic pipeline remains so strong
This article details the history of Anthropic (started by former OpenAI employees) and describes why so many people from OpenAI (co-founders, executives, and staff) move to Anthropic.
National Science Foundation Funds Development of Digital Twins for Healthcare
The NSF grants will fund seven projects, including the development of models for virtual clinical trials of cardiovascular medical devices, digital twin-based studies of neurodegenerative diseases, and AI-informed decision-making linked to glucose metabolism.
Coolest thing I saw…
MIT researchers have developed a new system that helps robots quickly map a scene and identify items they need to complete tasks (such as “get a first aid kit” or “move a rack of magazines”). There’s no demo, but the description and their approach are cool.
A company that caught my eye…
The company recently partnered with Google to accelerate the deployment of AI agents built on its “neuro-symbolic architecture.” The company claims its new architecture combines generative AI’s conversational skills with the predictability and traceability of rule-based AI.