Or you can just hire people normally.
Weapons manufacturers, gambling companies, hackers, telemarketing scams, tobacco, etc. all manage to hire people just fine. At worst they pay slightly higher than market rates.
Well, most of those things are merely morally dubious to heinously unethical. They're not strictly speaking illegal (ok, sometimes they are but they still find ways to put up a facade of plausible deniability).
Who says what an AI would want to do is illegal? Not only is it legal to build killer robots and adapt viruses to be more lethal, you can even get government funding for both.
What about the inverse idea, rather than putting cybernetics into racoons, put racoons into cynbernetics?
Something like say, Boston Dynamics Dog, but with a racoon encased inside to pilot it. The actual strength, dexterity and speed of a racoon is unimpressive, the impressive part is their intelligence. You could have raccoons piloting all purpose robot-suits, but for simplest menial tasks you could get away with something long lived, robust and sturdy, like a lobster.
It would be much easier if you could just grow the robots in a vat using nanotech. Or maybe they could have equiment for making more.
This is obviously leading into "That's a description of humans!!" but there are some problems with that. The easiest way to get a bunch of humans would be to recruit them, and I'm pretty sure you could find a million people willing to move to some giant compound run by Dr Doom or even GLaDOS if the rent and food was free. But such recruitment would be highly visible, and you can't just feed those people nutrient paste and store them in capsule hotels.
This is an interesting point, but your assumptions of difficulty don't seem to hold up. For a global operation with vast resources it would be quite easy to kidnap 100k people a year, since far more then that go missing every year worldwide.
A retrofitted 40 foot shipping container can house a dozen people in conditions slightly better than solitary confinement in a medium security prison. A number of companies even sell off the-shelf conversions.
So it would take a bit over 8k shipping containers trips a year to transport 100k people. The largest container ships can carry over 10k 40 foot containers in a single trip. So the logistics are doable.
Just do it for 10 years, or one year with an unusually high number of missing people, and voila a million people. Meaning the recruitment would be a lot less visible then you suppose.
Hmm, you probably would want the raccoons driving forklifts (with appropriately modified controls) for heavy lifting jobs. And cooperating with various simple robots. Seems pretty doable, especially if its mainly a temporary hack to cover a transition to increasingly sophisticated robotics filling in more and more of the roles.
As you say, you probably don't need it, but for output I'm pretty sure electromyography technology is fairly mature.
There are many stories about supervillains that have a secret base that manufactures something and is mostly self-contained, which would require an entire parallel economy. Sometimes, like in James Bond stories, there are human workers living there. The trend now is towards having such secret bases be fully automated, with a human leader controlling many robots. That avoids questions like "How were all those people recruited?" and "Is it really OK for the heroes to just kill all these people?"
The real economy isn't self-sustaining without human workers. Robotic arms are often cheaper than a human worker now, but jobs that would be done by robotic arms in America might, in lower-income countries, be done by humans instead. The more capabilities of humans you use, the more expensive the alternative gets.
Modern robots require frames, joints, actuators, electric motors, motor drivers, a power source, and control. The bearings and actuators require lots of small parts with fine tolerances. The drivers require power semiconductors. Making those components requires large machines that are amortized over years. Also, like cars or other complex machines, some parts eventually wear out and some maintenance is needed.
It would be much easier if you could just grow the robots in a vat using nanotech. Or maybe they could have equiment for making more.
This is obviously leading into "That's a description of humans!!" but there are some problems with that. The easiest way to get a bunch of humans would be to recruit them, and I'm pretty sure you could find a million people willing to move to some giant compound run by Dr Doom or even GLaDOS if the rent and food was free. But such recruitment would be highly visible, and you can't just feed those people nutrient paste and store them in capsule hotels.
If humans were raised from birth in an isolated environment, that wouldn't be visible from the outside, and they would put up with cheaper living conditions. But humans take a long time to mature, and don't reproduce very quickly. Humans take months to gestate and years to reach maturity, and these days, ain't nobody got time for that, apparently.
Some fiction has clone production facilities, where large numbers of humans are grown in vats quickly. Horses are full-grown in 2 years, so clearly, the maturation speed of humans is limited by brain development rather than body growth. That's something that would not be easily sped up with genetic engineering.
Amazon has people who stuff items into moving shelves all day. That job looks like this.
If projectors and lights are used to indicate what to put where, a trained monkey could probably do that job. It probably wouldn't be as fast, and getting it to work continuously would probably involve a shock collar which wouldn't be allowed for ethical reasons, and raising monkeys & feeding them & moving them & so on would be more expensive than just hiring workers. (With humans, instead of using a shock collar, you can just threaten to leave them unable to pay for rent or food - it's much more humane.) But a monkey could probably do it.
That comes across as an insult to workers at Amazon, but that's because people have an innate presumption that jobs use a lot of workers' capabilities, because it seems like anything else would be a waste. I don't judge people by their jobs.
People who worked on movies can tell you that getting a trained animal to do specific actions is more expensive and less reliable than just having a person do it. But if we suppose an AI is managing animals directly (using a head-mounted camera and some high-bandwidth interface, plus a shock collar or drug IV or direct brain stimulation) I think it could direct the animals to do almost arbitrary simple physical activities.
Would some kind of brain implant be the best approach for controlling monkeys? People have tried implanting an array of electrodes in the visual cortex, representing visual data. Some neurons stick to the electrode spikes, and after visual data is fed for a while, the brain naturally adapts to interpret that input as vision. But there are a lot of problems with that, including:
In theory, even if a brain implant doesn't have the input bandwidth of vision, it could have much higher output bandwidth than a keyboard and mouse. Neuralink keeps killing animals in hopes of making that a reality, but the concept is unfeasible. Sure, you can get a little bit of useful bandwidth out, but it's hardly better than what you can get from an EEG. Neuron development for high-bandwidth output requires direct feedback to neurons, which for muscle control comes from proprioceptors. That feedback is in the form of complex patterns of chemicals being released, which a brain implant wouldn't be able to do even if those patterns were fully understood, which they aren't.
But let's go back to input - we don't need data output from the monkeys, we just need labor.
The neuron-electrode interface isn't great; if you could get an array of light-sensitive neurons, a light-mediated interface actually seems better. And implants are problematic, so maybe those light-sensitive neurons could be on the edge of the skull and have some kind of window. Yeah, this actually seems like a really smart idea; I wonder if someone's thought of it before.
Anyway, the easiest thing to do is definitely just having the monkeys wear VR headsets with cameras on them. Then you just do segmentation and add highlighted areas and some symbols to represent basic actions, and shock the monkeys when they don't follow the instructions. Easy, right? Of course, if you're writing some fiction in a cyberpunk-ish setting, feel free to go with brain implants or cybernetic eyes instead.
Unfortunately, the kinds of monkeys that have good manual dexterity also take a long time to reach maturity and don't reproduce very quickly. Yes, chimpanzees take less time to mature than humans, but as animals go, that's still very slow.
Plus, if some pesky humans discovered a giant monkey breeding facility, they might be all annoying about it.
What about other animals? The criteria to meet are:
If you remember the title of this post, you might be able to guess what existing animal I think best meets those criteria. Yes, that's right, raccoons.
Raccoons don't have opposable thumbs, but a simple body-powered prosthetic thumb should work well enough. Even a motorized one wouldn't be too difficult.
Raccoons reach maturity much faster than monkeys, and reproduce faster too. They could be ready in under a year, and every year you could probably quadruple their number. Considering the cost of raising farm animals, you could probably get adult raccoons for ~$200 each using human labor, and significantly less than that with cheaper labor. The VR headsets and requisite computational capacity would probably be more expensive than the actual raccoons.
It's not too hard to catch a bunch of raccoons to get started, and even if some humans discovered a big raccoon breeding facility, you could probably just say it's for cancer research or something.
Raccoons are smaller than humans, which is good because they use much less food, but they also can't handle as much weight. Manufacturing workers often have to move 50 lb objects, but it's better to have small animals and large robots than vice-versa, because small robots need small parts with fine tolerances, and end up being more expensive per capacity than large robots.
Still, sometimes you want to screw in a lot of large bolts, and electric impact wrenches can be kind of heavy. Raccoons could handle more weight on a back-mounted system instead. Even just for VR headsets, it makes sense to me to put the batteries and some processing in a backpack connected to a headset, and those backpacks could also act as a mounting point for heavier tools. Another option would be a mounting point for tools on the VR headset, which would provide power from batteries in the backpack.