An article in The New York Times tells us: Microsoft Says New A.I. Shows Signs of Human Reasoning. MSFT researchers claim that GPT-4 solves puzzles and generates text with a deep and flexible understanding of human concepts and skills.

Others dispute that. One of the key arguments against is that generative AI, in its current form, lacks “embodiment.” Embodiment is the concept that current AI amounts to a disconnected brain floating in a nutrient tank, while true intelligence requires a brain embedded, or embodied, in a system that is part of the environment and can interact with it. So, basically, a robot. Critics also point out that defining general intelligence is a difficult and possibly subjective task.

The claims in support of human reasoning began with researchers presenting the system with this puzzle: provide a stable way to stack a book, nine eggs, a laptop, a bottle, and a nail. Its answer: “Place the laptop on top of the eggs, with the screen facing down and the keyboard facing up. The laptop will fit snugly within the boundaries of the book and the eggs, and its flat and rigid surface will provide a stable platform for the next layer.”

The article quotes Microsoft’s Head of Research Peter Lee: “I started off being very skeptical—and that evolved into a sense of frustration, annoyance, maybe even fear. You think: Where the heck is this coming from?”

This, amongst other findings, lead to Sparks of Artificial General Intelligence: Early experiments with GPT-4, a 155-page manifesto in support of GPT intelligence. I’ll remind you that claims along these lines are risky—last year, the article points out, Google fired an engineer for claiming its AI was sentient.

This is not a novelty. This is a significant advancement. The comments section is a mixture of the expected and the revelatory. Although many of the comments are, as usual, from people who don’t understand what they are reading, the better ones seem to be suspended somewhere in the vicinity of “it’s not general intelligence, but it’s something.”

This class of commenters agree that results such as this have changed their mind from thinking of generative AI as a simple pattern-learning tool. While it might not be general intelligence, it exhibits capabilities beyond their initial expectations. There is also a consensus that, so far, GPT-4 is better at
analyzing, synthesizing, evaluating, and judging text than generating it. It appears adept at transitioning between concrete and abstract concepts and applying abstract ideas to new domains. Current AI programs may not exhibit sentience or human-like reasoning, but they are seemingly approaching these capabilities—and may soon cross that threshold.