Nowadays everyone wants to build a godhead. I content myself with 3D printers, but if I did want godhood, I think I would buy some birds and tokens instead of a promethean datacenter. I think there's a ripe opportunity to be the progenitor of a new sentient(?) species. Here's a list of the most accomplished English-speaking parrots:

These birds all started training in either adolescence or adulthood, and they were immersed in relevant-to-their-interests English conversation for just a few hours per day while being trained. I don't want to minimize the work of these researchers, training a bird that will probably never exceed the intelligence of a dumb child is peak tedium, research grants are small and finite, and it's not clear that more of the same humans-as-teachers could have pushed the birds much further than the results above. Fortunately, we've recently invented machines of infinite patience and superavian language capabilities.

My proposal is to get a few families of newborn African Greys and raise them with a robotic feeder and game-engine animated parrot 'parents' on a screen adjoining their nest. Every waking moment their 'parents' will coo at them, offer food, warn of (fake, rendered) threats, and converse amongst each other in front of the chicks, all in a pitched-up chirpy accent of English tailored to parrot chick vocal cords. I don't know if this would work, but it seems like a much fairer shake at teaching birds real structured human language than anything done so far. It seems plausible to me that all English speaking parrots to date have essentially learned English as a second language, and as a sort of hobby. The 'model-rival' training described above involves a human teacher offering some sort of reward-for-interaction scheme, and a second human playing the part intended for the parrot in the hope it will get competitive and imitate to win the prize. This analog approach clearly works to a point, but I don't think anyone would claim it's recruiting the base survival drives of the bird to motivate learning, nor reproductive drive. Immersing the birds in English speech while the robotic parent-beak feeds them can pull the survival drive in without crossing any ethical lines: the normal course of a chick's early life is a little eating and a lot of chirping for more food while its siblings get fed. If we add a few virtual siblings chirping for food in English and getting fed, maybe we get somewhere?

This would be a years long project, so the AI interaction system would have plenty of time to evolve as a nights and weekends project for a motivated hacker. Persistent non-parent character parrots should be introduced on the screens over time so in the lead up to the birds' age of sexual maturity, they can see more charismatic parrots get laid more. Show the lady parrots pairing up with the most charming boy parrot. Parrot breakups. Parrot apologies. Throw everything we can at maximizing the salience of English speech. Our true goal here is to bootstrap goal-oriented parrot to parrot conversation in English, and ideally seed a talking parrot culture that rewards wit to start sexually selecting on English fluency in subsequent generations. This is also why we need to start with a few nests, not just one family. It might also be worth trying this on budgies (way faster generations) or monk parakeets (way more communal social structures) if you're really excited about the sexual selection angle.

Corvids

The other thing AI-automated training enables is 'longshot' research no sane human would be willing to spend years on. For example, common crows and their corvid relatives have more documented intelligent behaviors than parrots and are capable of fantastic vocal mimicry, but nobody in academia has tried to train them for complex human language conversation. As far as the LLMs can tell me, this is mostly a historical accident. Something like availability bias led to the big bird speech research going to charismatic parrot species, while corvid research focused on tool use, puzzle solving, etc.

I think that's a missed opportunity! Crows have immensely complex social lives with long-running alliances and grudges between individuals and families, and they can remember individual humans for years. They clearly model the agency of other living things to a greater degree than most animals, their brains are as big or bigger than African Greys, and they're comparatively easy to entice into puzzle solving and other goal directed behavior to win treats. It seems like fertile ground to test how much of oral communication capabilities are 'baked in' to brain structure by evolution vs adaptive response to a language-rich infant environment. The crows have every incentive to communicate effectively amongst themselves, maybe we just have to teach them grammatical structure and they'll run away with the ball. Probably not, as with every good outcome in this scheme, but what if!

Godhead

We should be clear eyed here: the overwhelmingly likely outcome is that we get some particularly great talking parrots, they don't learn complex bird to bird conversation, and whoever gets sniped into trying this by this post has some very cool pets. BUT! Let's imagine how fun it would be if it worked.

Suppose the 2nd generation of talking parrots learns English from its real flesh and blood talking parrot parents, and with continued access to the ever-improving virtual parrots the 2nd generation is a bit smarter than the first. We're off to the races! Give them more habitat, let them grow a culture! The things we'd stand to learn about Saaper-Whorf ish questions of 'does language structure cognition?'! Imagine being the human who created a self sustaining population of talking parrots. We're deep in a flight of fancy now, but perhaps their distant descendents would remember your name.

And how smart might they get? They hatch from eggs, their brain size / cranial capacity might be fairly unconstrained for rapid evolution. To say nothing of the genetic tools we humans can bring to bear - if the experiment succeeded and we found ourselves with sapient but not very sharp parrots, gengineering them bigger brains seems like the kind thing to do right? This scheme is ripe for execution, and it will only get easier as the years go by and the LLMs can build more of the software infrastructure to train the birds. Whose name will be sung from the trees?