"Artificial intelligence" is just a very sophisticated pattern-finding script. One application of that - and only one - is a natural language model, synthesising human speech. And that is the one that we all focus on because it's easy to interact with. But all these AI models - including DeepSeek - are much more than that. And it is all the other functions that are actually far more interesting and useful.
For instance, imagine an AI air traffic controller. It knows all past patterns of aircraft travel, data from all sources, and works out what patterns tend to precede accidents. It can then recognise near-instantly when a situation is likely to lead to an accident, and warn everyone to take action. The advantage over a human is that it can almost simultaneously look at every data source, so it may notice problems long before humans would see all the data and join the dots. The disadvantage is that it may not recognise the danger in a wholly new situation with no precedent. But it has a very obvious application as at least part of the air traffic control system. Now, let's not debate whether that's a good idea or not, I'm just presenting it as a potential application to illustrate another point. My actual point is:
None of that involves talking like a human. But it's still AI.
Such a model would have multiple interfaces. It would have output to warning lights, radio signals and so forth. It would have input from switches and buttons. It could function entirely without human speech, and still be AI.
However, since natural language models (NLM) are now possible, it might also have an NLM, to make custom warnings to pilots and allow people to query the computer in natural speech. That natural language interface would be another model (or another part of the model).
Now, all the concerns about whether AIs have political biases, personalities, and so forth largely apply to the NLM. The NLM might have political bias in some details of its speech depending on who wrote it. But that is almost irrelevant, it really doesn't matter in this use case. The actual AI air traffic controller part is not politically biased because it doesn't even know about politics - it doesn't even know language. It just knows aircraft flight paths.
Both Chinese DeepSeek, and woke Western models, can be run on a server and trained to be an air traffic controller, or anything else. And/or it can be used as a NLM. In the vast majority of uses, political bias is irrelevant, and might only affect a few peripheral comments here and there in the NLM side of things. Neither model can be dismissed due to this sort of bias, it's a tiny peripheral detail. The actual potential is far deeper and greater than that.
