Instead of discovering a grand theory of intelligence and using it to build the first AGI, I think we’ll stumble upon it through experimentation, and no one will have a detailed notion of how its mind works. The architecture of LLMs probably preclude them from ever achieving general intelligence, but LLMs themselves are smart enough to develop better kinds of machines. The physical infrastructure being built for LLMs today (data centers, power lines, power plants) will also be needed to support true AGIs after they’re created. As computation and hardware get cheaper and the LLMs get smarter, they’ll be able to run more and better experiments.
I think there will be a global network of intelligent beings in the future, each specialized for a different type of task (cognitive, physical, emotional, etc.), and each doubling as a sensor node that feeds data into the network, and that some central intelligence would allocate tasks within the network to the intelligent beings best suited to doing them. Think of it as something like a “Borg Collective” where some of the Collective’s members are pure AIs, some are augmented human brains floating in jars with wires going into them, and some are new life forms we can’t imagine.
For a time, some members of the network will be humans with varying degrees of augmentations. Such a setup would dovetail with other long-term trends, including the rise of personal assistant AIs that would come to understand the strengths and weaknesses of individual humans, and mass surveillance that would track the locations of all humans in real time. Once the skills of each human and the locations of each human are known, and once all humans are connected to the global network, it will become possible for a central intelligence to fluidly assign tasks to each human in a manner that makes maximally efficient use of the labor force. “Mechanical Turk,” which is a computer-based service where people get paid to do random tasks, gives a small clue to how things will be like.
The network would find genuine uses for humans for years after the invention of AGI and advanced robots. Brush up on Ricardo’s Theory of Comparative Advantage and you’ll understand how weak, slow-witted humans like us could still fill many niches in the economic and social fabric of a future world where there are vastly smarter and faster machines. As machines take over jobs and as they become better at recommending enjoyable things for us to do, humans will invariably start asking machines for “career” advice. The machines will match humans to tasks that represent some balance of optimal skill utilization and worker satisfaction.
Unfortunately, I doubt our role would persist forever. Even if the organic composition of our brains gave us an advantage in some kinds of computation, eventually better, bioengineered brains would beat us in those niches as well, and we’d be at the mercy of other intelligences.
Dovetailing off of that last point, while humans are the only animals capable of intelligent thought, I think some other species’ brains have niche advantages over ours. Bat brains, for example, are suited for echolocation, and squirrel brains probably have better spatial memory than humans, or how else could one of them find the hundreds of nuts he buried months earlier? If it turns out that the organic substrate conveys advantages to minds that protect them from obsolescence from machines, and if science is used to create brains optimized for specific modes of thinking, then the human brain won’t be starting point for all of the latter. It will make sense to build some kinds of organic “processors” by starting with, say, a dog brain as the basis, and modifying it from there.
I’m unconcerned with predictions that the Earth will become uninhabitable in the far future due to things like the expansion of the Sun or the convergence of all the continents (“Pangea Ultima”). This isn’t because I think I’ll be dead by then; it’s because I believe we’ll have technological solutions to the problems. For example, as the Sun expands and its brightening light heats up Earth’s surface, we could deploy satellites between the two bodies to block out the excess sunlight, keeping Earth at a constant temperature. We could even very slowly widen the Earth’s orbit to keep it at a constant distance from the Sun’s surface as the latter expands. This wouldn’t be easy, but if we have a billion years to do it, it’s just an engineering project.
Once we have autonomous combat drones (in the form of planes, ships, tanks, infantry, etc.), a country’s maximum military potential won’t be capped by the size of its human population. Arbitrarily large numbers of robots could be built, and a country’s military strength would instead rest on its GDP, technology, and access to resources. Having a larger human population might actually be to a country’s military disadvantage since it would be a larger drain on its resources. The global power balance could shift in unexpected ways as a result.
Like anything else, robot servants will get dirty over time. The easiest way to deal with this will be to have them wear clothes that they’ll wash in regular laundry machines. Even if covered, their skin will also slowly get dirty, and the most obvious solution to that is to make them waterproof so they can bathe in standard bathrooms. And if some of them are built to be our companions, we’ll want them to be able to partake in meals with us even though they will lack digestive systems. The easiest way to deal with the masticated waste will be for them to spit it up into toilets afterward.
If you wanted to experience a perfect virtual reality replica of what part of the world was like in, say, 2020, it would be impossible due to insufficient data, and the inaccuracy would worsen the farther back in time you wanted to go. However, late this century, everything and everyone will be under surveillance, so it will be possible to use VR to piece together moments in time. Someone in 2100 could re-live a day in 2075 with perfect accuracy (or so close to perfect that a limited human mind couldn’t detect the inaccuracies).
Imagine we built a Dyson Sphere or Swarm around our Sun, and structure was 1 AU in diameter. It takes 8 minutes and 20 seconds for light to travel from the Sun to the Earth, so sending a signal from one end of the Dyson structure to another could take up to 16 minutes and 40 seconds.
If the Dyson structure is inhabited by intelligent machines spread out across it, then the tyranny of distance will impose communication lag problems on their civilization. This means that the machine civilization will not be a unitary consciousness; it will be many conscious entities distributed across the massive Dyson structure.
Each entity would control a relatively small area over which lag times weren’t so bad that they would impose a burden on the ultra-high processing speeds each entity would be capable of supporting.