>>12319322>You can just let the AGI control some entity in a simulation if you want to see how it behaves in a physical environment. That would be much cheaper and easier, and it would also give us way more control over what kind of environment we put the AGI in.That's debatable whether or not it is cheaper.
You can just strap a camera to a platform with motorized wheels and that could be enough for some purposes.
Also real life images are extremely more detailed than virtual environments, so if you want your AI to engage with real world data in the end, using sensors in the real world is a necessary step.
>The idea that an AGI needs to have a built-in module for understanding 3D space is hilarious.>What's so special about understanding 3D space?Neuroscientists have fond that sections of our brains are homeomorph to our 3D visual surrounding. That is, spacial directions and distances have equivalents in those neural fields.
Your idea that we should build an AI top-down is not a bad idea. In fact, I suspect real breakthroughs will come from there (because bottom-up approaches require too much processing power for now), but contrary to what you said, we won't have a choice but delve into finer and finer details as we go on, because the brain is a very complex machine, not uniformly made, that possesses specific mechanism in specific places.
There was a bunch of retards a few years ago who claimed that they cracked intelligence and found a formula for it. I'm afraid there's no single equation for what we define as intelligence, but a multitude. And even if you implemented all of them, you would still need to find the original data your brain has to start with.
Like the video said, this amount of initial data required is bounded by genetics.
