The idea of AI creating human-like robots raises a powerful question: can we truly stay in control of something that looks, moves, and maybe even thinks like us? The answer depends on the choices we make right now. If we design these systems carefully, with clear rules, transparency, and strong ethical boundaries, then yes, we can ensure they remain tools that help us, not take over from us. But if we ignore the risks, let things grow unchecked, or allow these robots to make decisions on their own without oversight, we could find ourselves in dangerous territory. The real concern isn’t just how human they appear, but how much freedom and decision-making we give them. Staying in control means being proactive, setting limits, asking tough questions, and working together globally to guide AI in a direction that serves humanity, not replaces it.

This question currently lies in the realm of logical possibilities as opposed to being within the realm of realistic probabilities. It includes an assumption that AI will make a significant advance in capabilities, and advance beyond what humans have achieved to date, and thus is unanswerable,

Currently AI is limited to being able to answer binary questions when given guidelines, such as what is the fastest route between points A and B with the following restrictions. These might include that driving through a building is not permitted, driving on sidewalks is likewise not permitted, etc. Once a solution is found adding in detours causes as major reworking effort, and too many detours being added will cause the AI to fail. It must, as always, attempt every possible route and then discard any route that violates the guidelines, that is it must plot routes that use sidewalks, go through buildings, etc., and then discard those that fail the guidelines A human can solve the problem without being instructed as to what is not permissible and can usually resolve any number of detours with no special effort AI has no generalized knowledge and currently no means to achieve such.

Another example of where AI must currently fail is the simple question of which tastes better, an apple or a peach? The answer is not, and cannot be made, binary, that is "yes or no", "switch turned on or off" or some other binary choice. As a human, if I was asked which tastes better, an apple or a peach, I would query, "in what or when?" and then be able to give a response, plausibly "right now a ". But the answer is not something achievable via binary process.