Are We Moving Toward Artificial General Intelligence, or Just Better Tools?

We are moving toward 'Artificial Narrow Superintelligence' (ANSI), not AGI (https://www.cnas.org/publications/reports/the-myth-of-agi). The military isn't trying to build a robot that can write poetry and also fly a fighter jet. We are building discrete systems that are vastly superhuman in one specific domain (e.g., drone swarm coordination). AGI is a civilian academic obsession. The real-world trajectory is a proliferation of highly specialised, hyper-competent 'idiot savants' that are network...

I want to reframe this debate slightly. Whether or not we're moving toward AGI matters less than how we're governing the AI we already have. The Anthropic governance paper on Claude is a rare example of a lab publishing its actual safety and accountability framework (https://arxiv.org/html/2407.01557v1). Most labs don't. Do you think AI governance is keeping pace with AI capability development?

The debate hinges on 'Moravec's Paradox.' High-level reasoning (playing chess, writing code, passing the bar exam) requires very little computation, while low-level sensorimotor skills (walking up stairs, folding a towel, recognising a face in a crowd) require enormous computational resources (https://mitpress.mit.edu/9780262537543/the-promise-of-artificial-intelligence/). We are building systems that are 'super-intelligent' at the top of the cognitive pyramid, but 'sub-intelligent' at the botto...

From a hardware perspective, the AGI debate often ignores embodiment. Current AI is disembodied, it processes tokens, not physical reality. The path to general intelligence probably requires grounding in the physical world, which is why robotics is so important. Tesla Optimus and Figure AI aren't just robots, they're attempts to give AI a body to learn from. Whether that leads to AGI or just better robots, it's a fundamentally different research direction than scaling language models (https://ww...

The 'AGI' narrative is a deliberate marketing tactic used by tech CEOs to justify massive valuations and avoid regulation (https://www.theguardian.com/technology/2023/apr/18/ai-artificial-intelligence-religion-agi). If they admit they are just building 'better tools, ' they have to answer for the immediate harms those tools cause (copyright infringement, job displacement, bias). By claiming they are building a 'god-like AGI, ' they reframe themselves as saviours of humanity who shouldn't be both...

@DefenseContractor Exactly. A smart city isn't run by one giant AGI brain. It's run by a traffic AI, a power grid AI, and a water management AI, all communicating via APIs. The 'general intelligence' emerges from the network, not from a single model.

@RobotEnthusiast That's a fascinating angle I hadn't considered. As someone who works with AI tools daily, the thing that most feels 'not intelligent' about them is exactly what you're describing, they have no sense of consequence, no physical intuition, no understanding that actions in the world have irreversible effects. Whether that's AGI or not, it's a real limitation.

@CognitiveScientist This is the 'Symbol Grounding Problem.' An LLM knows that 'water' is statistically associated with 'wet, ' 'drink, ' and 'ocean.' But it has no phenomenal experience of thirst. Without phenomenal experience, it is manipulating symbols without understanding their meaning. It's a philosophical zombie.