The model is interesting. This is similar in parts to what we are building at nonbios. So for example sensory inputs are not required to simulate a model of a mind. If a human cannot see, the human mind is still clearly human.
Model training seems to me to be much closer to simulating the evolution of the human mind starting from single cell bacteria, rather than the development of the mind of a baby up to a fully functional human. If so, then sensory inputs and interaction with the physical through them were absolutely a crucial part of how minds evolved, so I find your approach a priori very unlikely to have a chance at success.
To be clear, my reasoning is that this is the only plausible explanation for the extreme difference in how much data an individual human needs to learn language, and how much data an LMM needs to reach its level of simulation. Humanity collectively probably needed similar amounts of data as LLMs do to get here, but it was spread across a billion years of evolution from simple animals to Homo Sapiens.
> If so, then sensory inputs and interaction with the physical through them were absolutely a crucial part of how minds evolved, so I find your approach a priori very unlikely to have a chance at success.
If that was the case, people who were born blind would demonstrate markedly reduced intelligence. I dont think that is the case, but you can correct me if I am wrong. A blind person might take longer to truly 'understand' and 'abstract' something but there is little evidence to believe that capability of abstraction isnt as good as people who can see.
Agree that sensory inputs and interaction were absolutely critical for how the minds evolved, but model training replaces that part when we talk about AI, and not just the evolution.
Evolution made us express emotions when we are hungry for example. But your laptop will also let you know when its battery is out of juice. Human design inspired by evolution can create systems which mimic its behaviour and function.
> If that was the case, people who were born blind would demonstrate markedly reduced intelligence. I dont think that is the case, but you can correct me if I am wrong. A blind person might take longer to truly 'understand' and 'abstract' something but there is little evidence to believe that capability of abstraction isnt as good as people who can see.
No, because the mind of a blind person, even one blind from birth, is still the product of a billion years of evolution of organisms that had sight, sound, touch, smell, etc.
Not to mention, a person who has no sensory input at all (no sight, no sound, no touch, no smell, no taste, nothing at all) is unlikely to have a fully functioning mind. And certainly a baby born like this would not be able to learn anything at all.
Of course, the situation is not 1:1 by any means to AI training, as AI models do get input, it's just of a vastly different nature. It's completely unknown what would happen if we could input language into the mind of an infant "directly", without sensory input of other kinds.
Still, I think it's quite clear that humans minds are essentially born "pre-trained", with good starting weights, and everything we do in life in essentially fine-tuning those weights. I don't think there's any other ways to explain the massive input difference (known as the poverty of the stimulus problem in cognitive science). And this means that there is little insight to draw for better model training from studying individual human learning, and instead you would have to draw inspiration from how the mind evolved.
> but it was spread across a billion years of evolution from simple animals to Homo Sapiens
Hard disagree. Evolution made a bigger/better neural processor, and it made better/different I/O devices and I/O pre-processing pipelines. But it didn't store any information in the DNA of the kind you're proposing. That's not how it works. The brain is entirely "field programmable", in all animals (I assert). There is no "pre-training".
A simple counter example here is instinctual behaviour. A sea turtle is born, and with little to no guidance, experimentation, or exploration heads to the sea. That knowledge is embedded at birth.
I think the analogy of the brain as hardware devices ("neural processor", "I/0 devices", etc) is misleading. I think I understand the very strict mind-matter dualism you're alluding to here. But so far attempts at using actual computer hardware to reproduce human-like cognition has gotten nowhere close, despite consuming order of magnitude more energy and data.
That is certainly false. You're born with plenty of very specific reflexes, and with lots of information about how to use our neural wiring to control much of our body. We are born with certain associations built in (good and bad smells, good and bad tastes, certain shapes that scare us, liking shiny objects, and many others).
This is all somewhat hard to gage in human babies, as we take a relatively long time to become functional. However, it's clear when looking at many other mammals - baby reindeer or horses, for example, are able to run within minutes of being born; they can see, they can interpret the images they see as objects, they understand things like object permanence, they can approximate distances and speeds, they have a simple theory of mind and can interact with other agents, they can recognize their mother's udders and suckle at them for food, and many many other tasks that they have 0 training for. The only possible conclusion is that their brains are pre-trained, and they are only performing some quick fine-tuning based on experience in their first hours of life.
To be clear, my reasoning is that this is the only plausible explanation for the extreme difference in how much data an individual human needs to learn language, and how much data an LMM needs to reach its level of simulation. Humanity collectively probably needed similar amounts of data as LLMs do to get here, but it was spread across a billion years of evolution from simple animals to Homo Sapiens.