As AI becomes more ubiquitous is it possible to train it so that the biases ever present in the news media and on the internet are minimized? Is it possible to minimize that bias without injecting our own biases into the training, even if the information dominated by western media is highly biased toward western culture?

It may be possible.  

If it is possible to understand bias from a purely mathematical perspective, then there may be a purely mathematical solution to the problem of removing bias. A solution that evades the problem of injecting personal bias into the process of correcting observed bias in the news media.

To understand this, consider the weights in an artificial neural network are just a series of matrices that are applied sequentially with some non-linear activation between each matrix application. But the bias is contained in the matrices. The number of non-degenerate principal components of these matrices represents the real amount of useful information that the matrix is trained on and the eigenvalues say something about the quality of that information. When the eigenvalues of these matrices are dominated by a few eigenvalues, then we say that the information content is highly biased. When the eigenvalues of the non-degenerate eigenvectors are comparable in value then we can say that the model is less biased. As part of the machine learning training program we can enforce the eigenvalues to be nearly equivalent, thus enforcing anti-bias even when training on highly biased datasets.

Recently researchers at MIT and Microsoft have come up with a similar idea in a paper called "The Truth is in There: Improving Reasoning in Language Models with Layer-Selective Rank Reduction", published in Dec 2023, by getting rid of the noise or least significant eigenvalues of certain of the matrices in neural networks for large language models. My solution, which was described in early 2023, is a different form of regularization, but the end result is the same, that certain matrices have no insignificant eigvalues.

. . . . . . 

Is the future of AI to be conditioned by our basic human nature, from evolution itself, to fight or flight? This is the most basic bias or prejudice that we must remove from AI; otherwise, not trusting us, it may destroy us if it becomes capable of doing so.

This is not a philosophical discussion about whether AI is conscious but rather an exploration of the nature of evolution itself.

It is fun to see what we can infer about the universe using only the simplest possible observations and ideas. What can we infer by thinking about the night sky and conducting some simple thought experiments?

The night sky is darker than the surface of the sun because either the universe is not infinite in time, or it’s not infinite in space, or neither, or the rate at which the universe is expanding is faster than the rate at which light from distant sources can affect the brightness of the night sky. From such a simple observation, so much can be inferred.

The next crucial observation is that we actually see the universe expanding. This makes it much more likely that what we call our universe had a beginning, although this observation alone does not rule out the possibility that it is continuously growing in size. In fact, observations of the cosmic microwave background strongly support a particular theory of General Relativity called Lambda-CDM, in which the universe was once hot and dense. However, this theory does not explain everything. Added to it is a theory called inflation that accounts for the universe's homogeneity on the largest scales. Nevertheless, these results require quite a lot of physics to understand. But the key point is that all of that physics is somewhat intuitive and expected, given the darkness of the night sky.

There is another observation, perhaps just as simple as observing the night sky, that leads to ideas even more profound than the notion that the universe had a beginning.

We live in a universe with only three physical dimensions. Why is that? Nothing in physics or mathematics or any theory we know rules out the existence of universes with any number of physical dimensions.

One principle of physics is that if some physical situation is not prohibited, then that physical situation must occur. If universes with any number of dimensions can exist, then why do we find ourselves in a universe with three physical dimensions?

Through some relatively straightforward physics calculations, we find that the orbits of planets around stars are only stable in three physical dimensions. Stable planetary orbits are a requirement for life forms like ourselves, which take billions of years of evolution on the surface of watery planets to come into being.

So we find ourselves in a universe with three dimensions exactly because all kinds of universes actually exist, but we can only exist in ones with three physical dimensions.

What kind of process could have caused our universe to come into being?

Some physicists like Stephen Hawking have proposed that our universe came into being by quantum tunneling out of nothingness. I think that is a great idea, but I don't think it applies to our universe. I think that's a little like Christopher Columbus thinking that America is Asia—it’s the right idea but the wrong continent. Our universe is too complicated and messy to have come into being in that way. It’s probably the result of a more complicated process, the components of which may have come to be in the way Hawking thought.