AI Society for 9.9.25: Fond Farewells and Two Relevant Essays on China and AGI from the NY Times

Source: ChatGPT

Greetings — I plan for this to be my last daily blog for a while… on to other adventures! But I’ll still post periodically on my data analysis and visualization projects as well as trends that I consider especially noteworthy. For review, my latest projects are here:

Published at AI Advances:

Visualizing Scientific Discovery: How Clustering and Embeddings Reveal Hidden Patterns in Research

https://ai.gopubby.com/visualizing-scientific-discovery-how-clustering-and-embeddings-reveal-hidden-patterns-in-research-78be810c7d04

Published at Towards AI:

Automating Competitive Intel: From 10-Q to Insights in Minutes

https://pub.towardsai.net/automating-competitive-intel-from-10-q-to-insights-in-minutes-beef67d67a74

Published at Generative AI:

Scaling Scientific Discovery: How We Built a High-Performance PDF Recommendation Engine

https://generativeai.pub/scaling-scientific-discovery-how-we-built-a-high-performance-pdf-recommendation-engine-2763cde76e92

Published at AI Advances:

AI Energy Paradox: Powering the Future Without Burning It Down

https://medium.com/p/752396678a6e

Published at Towards AI:

Agents on the Loose? Why Securing AI Isn’t Optional Anymore

https://pub.towardsai.net/agents-on-the-loose-why-securing-ai-isnt-optional-anymore-b0178c8f4690

Published at Generative AI:

An arXiv Search System with 2.2 Million Abstracts

https://generativeai.pub/an-arxiv-search-system-with-2-2-million-abstracts-8da1412962fd

Two recent opinion pieces, both in the ‘NY Times,’ struck me as poignant and reflect our current AI challenges. The first, a guest essay (link shared) by Thomas Friedman and Bill Brink dives into the current AI schism between the US and China, and potential disastrous consequences if global AI bifurcates. Thomas notes:

There’s only one way to manage this, and that is if the two A.I. superpowers — China and the United States — collaborate together on a system for controlling A.I. and ensuring that every A.I. device that either of them makes or sells to the other has embedded in it a set of ethical normative controls to ensure that their A.I.s can only be used for the advancement of human well-being, and not for any nefarious purposes.

We’ve grown up in a world where the only ones who had agency were God and God’s children — us. Well, we will now have a new species with agency. And there is nothing that guarantees that its agency will always be in alignment with human well-being. That’s №1.

With A.I., it could be the equivalent of giving everyone a nuclear bazooka that actually learns and improves on its own with every use. And that’s why I feel so strongly about not only getting these controls in place in the United States, but also not doing what we did with social networks — sitting back and saying, “Let’s just move fast and break things,” which is what Mark Zuckerberg urged us to do. And then he broke society. He urged us to have no controls over what is published on social media platforms. And now we live in a world awash in misinformation, disinformation, and hate speech that is tearing our society apart. Well, if we follow the same advice on A.I. — to just move fast and break things — this time, we could break the whole world.

The second, penned (link shared) by Gary Marcus, looks more at the peak of inflated AGI expectations, top-of-mind based on recent LLM developer missteps (i.e., GPT-5, Lllama 4, Grok 4). He writes:

GPT-5 is a step forward but nowhere near the A.I. revolution many had expected. That is bad news for the companies and investors who placed substantial bets on the technology. And it demands a rethink of government policies and investments that were built on wildly overinflated expectations. The current strategy of merely making A.I. bigger is deeply flawed — scientifically, economically and politically. Many things, from regulation to research strategy, must be rethought. One of the keys to this may be training and developing A.I. in ways inspired by the cognitive sciences.

However, as I warned in a 2022 essay, “Deep Learning Is Hitting a Wall,” so-called scaling laws aren’t physical laws of the universe like gravity but hypotheses based on historical trends. Large language models, which power systems like GPT-5, are nothing more than souped-up statistical regurgitation machines, so they will continue to stumble into problems around truth, hallucinations and reasoning. Scaling would not bring us to the holy grail of A.G.I.

Source: Gary Markus

Many of generative A.I.’s shortcomings can be traced back to failures to extract proper world models from their training data. This explains why the latest large language models, for example, are unable to fully grasp how chess works. As a result, they have a tendency to make illegal moves, no matter how many games they’ve been trained on. We need systems that don’t just mimic human language; we need systems that understand the world so that they can reason about it in a deeper way.

We need a new approach, closer to what (psychologist Daniel) Kahneman described. This may come in the form of neurosymbolic A.I., which bridges statistically driven neural networks (from which large language models are drawn) and some older ideas from symbolic A.I.

To build A.I. that we can genuinely trust and to have a shot at A.G.I., we must move on from the trappings of scaling. We need new ideas. A return to the cognitive sciences might well be the next logical stage in the journey.