If you are talking about machine consciousness, then I don't believe we will ever achieve it with the way semiconductors are currently built. I subscribe to Roger Penrose's theory that human consciousness arises due to quantum entanglement effects between microtubules in neurons. In computers, the LLMs and other AI models exist in high level application space many levels above the bare metal, so any quantum entanglement effects will be between adjacent transistor gates but have no logical connection to the AI model being executed whatsoever.
I find far-flung explanations of human intelligence to be rather passe. There's nothing special about humans. The main advantage of the human brain is the sheer
number of connections. There are
~100 trillion synaptic connections in a typical human brain, against the largest neural network right now, which is just
~1-2 trillion parameters. Obviously the two are not 1:1 comparable, but you get the idea - the average human brain has an order of magnitude more connections than our
largest neural networks today and it all fits inside a small skull box, compared to the large GPU clusters needed to do inference on anything close to the same size.
The other difference is how they were trained, or rather in the human case, how it
evolved. Functional human brains are the product of two complementary processes - evolution (biology) and education (nurture). Humans are born with a biological blue print shaped by
millions of years of evolution, which only need to be "fine tuned" on their actual life experiences. By contrast, we're training our AI models from scratch. Roughly speaking, "evolution" is what we're trying to emulate through feeding the model the whole of the internet to create a baseline.
But it's a poor man's version of evolution because it can only learn from human artifacts like text, videos, images, etc. It can't learn the embodied, interactive behavior that the human brain evolved to handle. That's why LLMs are "merely" pattern matching boxes - the equivalent of the visual-spatial-verbal processing components within the human brain. They lack the rest of the machinery we've evolved to have, and that's why they can't act autonomously in an intelligent way. They just don't have those sorts of experiences.
But Open AI, Google, etc. all know this, as well. That's why the current direction of generative research is extending generative models through agent-based capabilities. That is to say, get the model to start interacting with the environment & making decisions, and learning from those decisions through reinforcement learning, network evolution, etc.
That's where AGI will ultimately be born, although I think we're still a decade away, at the minimum, both due to diminishing returns on computing hardware and because much still has to be figured out about how to scale agent-based learning.
