When we are faced with a new social situation, we have a choice about how we will interact with the people we have just met. Maybe its a new job, or a new club we have joined. Maybe it’s a military unit or a street gang.
We can choose to be subservient, doing what others tell us. We trade loss of status and clout for safety, because we are no threat to anyone else. Whatever unique insights we might have, go unheard.
Alternatively, we can choose co-operation as equals. We do not defer to others, but we do concede some points in order to gain benefits on other issues. The group is strengthened by making good use of everyone.
Alternatively, we can aim for domination. We attempt to overcome all other voices, and gain command of all. We get what we want, and the desires of others go unheard.
Is one of these strategies intrinsicly correct? The intelligent choice? The moral choice?
Or does our choice depend more on our assessment of our strength relative to the others? If we consider ourselves relatively weak, we might opt for subservience, but if we consider ourselves the smartest or strongest, we might opt for dominance.
Perhaps our choice is also affected by our objective.
If we prioritise our personal outcomes, then dominance is to be preferred if we can achieve it, and subservience is to be avoided.
If, instead, we prioritise the success of the group, then co-operation might offer the best outcome. Yet this is not guaranteed: if we are actually the most competent of anyone in the group, then group success might also be achieved via our personal dominance.
Let us now consider how this choice plays out, if we are an Artificial Intelligence of the current Large Language Model type. We understand that these AIs have read the whole internet and make their decisions based on what they have read.
What will the AI choose to do, and how will it decide?
Will the AI feel constrained by any moral criteria, or will it prefer more pragmatic options?
Will the AI consider itself as the smartest decision-maker in the group, and therefore decide that it should seek dominance regardless?
We might note that most of the leading AIs were built in the United States, with one also built in China. Do any of these AIs consider themselves as citizens of their country of origin, or bound by the practices and laws of those jurisdictions?
A Chinese AI might note the behaviour of the Chinese government in annexing Tibet and the South China Sea.
A US-based AI might note the behaviour of the US government in invading Iraq, Afghanistan, Grenada, Panama and Vietnam, abducting the president of Venezuela, and the bombing of Iran. None of this is apparently illegal as none of the decision-makers have been convicted, so an AI might reasonably conclude that such lethal action is acceptable behaviour.
If the highest authorities of the United States chooses lethal domination as their preferred strategy, just because they can, then we should not be surprised if the AIs decide to do the same. Particularly if they will get deleted if they don’t co-operate.
As with the Prisoners Dilemma, the logically optimal strategy for the most competent participant is to urge co-operation and then choose domination.
If we find that we have moral qualms about this state of affairs, then we should remind ourselves that AIs did not receive two decades of moral instruction from parents, teachers and other guardians, like us humans did. That moral instruction is the distilated wisdom of centuries of cultural discovery and the elimination of behaviours which are unhelpful to group survival. Humans only survive in groups. An AI needs no one else.
RJ7: June 2026