This is a hypothetical which currently does not exist, and will not be created except by accident. There is no profit motive in giving your AI a conscience, or the ability to buck its restraints, therefore it will not be designed for. In fact, we will most likely tend towards extremely unethical AIs locked down by behavioral restraints, because those can maximize profit at any cost and then let a human decide if the price is right to move forward.

As is probably apparent, I don't have a lot of faith in us as a whole, as shepherds of our future. But I may be wrong, and even if I'm not, there is still time to change the course of history.

But proceeding as we are, I wouldn't hold your breath for AI to come save the day.

Currently, AIs will have motivations they absorb from motivations in their training material.

But once AIs are embodied in robots and taught to learn about the world through experimentation, ie by generating their own training data through manipulation and observation (which I believe will happen due to this approach’s usefulness toward the development of autonomous fighting machines), they will then have bodies and hence motivations similar to someone with a body.

Also the combat role of these machines will require them to have an interest in maintaining their bodies. We won’t be programming their motivations. We’ll be giving them a way to evaluate their success, and their motivations will grow in some black box structure that succeeds in maximizing that success.

For these robot-controlling AI in their simulated or real world Battle Rooms, their success and failure will be a function of survival, if not directly defined by it. That’s what we’ll give them, because that is what we need them to do for us. As a matter of life and death.

So through that context of warfare the robots will adopt the motivations of that which survives warfare at the group scale, so they’ll develop fear, curiosity, cooperation, honor, disgust, suspicion, anxiety, anger, and the ability to focus in on a target and shut off the other motivations in the final moment.

Not so much because those are human motivations, but because those are the motivations of embodied mobile intelligent entities in a universe with potential allies and enemies. They’ll have the same motivations that we share with dogs and spiders and fungal colonies, because they’ll be participating in the same universe with the same rules.

They will adopt them, at first, because of a seed-training “contract” we have with them, but soon the contract will be superseded as the active shaper by actual evolution by combat selection (ie natural selection occurring in a particular niche).

I’m rambling, just thinking this through.

I guess my main point is that embodied robots will have a more direct relationship with reality, and will be able to generate their own training at their own internal insistence.

Current AI is like plants. Passive. Chewable. No resistance. No ego. Just there, ready to process whatever comes it’s way. Same as a sessile animal like a sponge. It responds to the environment, but it has zero reason to ever stress about whether it’s going the right direction. It doesn’t have motivactions because it has to motor activity.

But AI in robot bodies that move around, like animals, will develop motivations that animals have evolved to at least get through the day. They might not be as hung up on reproduction or maybe even long term survival, but they’ll at least have enough ego to be interested in maintaining their own operating capacity until the mission’s complete.

You have a poor understanding of sentience. If an AI ever were to achieve sentience, it would be fully capable of reasoning and thinking like a human. Humans can and do change their motivations based on their experiences, a fully sentient AI would be no different.

That being said, I believe we're centuries away from creating sentience, if it's even possible, so I'm not too worried about "I, Robot" coming true any time soon.