[jack_jennings]

This is called "Optical" kerning in Adobe products. In general it's a good time saving device if for some reason you're using a typeface with no kerning pairs defined, but the result often ends up looking robotic and either too tightly or too loosely spaced (as I believe it ends up kerning every pair rather than only those that need kerning). If you have a typeface made by a reputable foundry there's usually no need to turn this feature on.

On the type design end of kerning, automated kerning does exist, though it's always tuned by hand afterwards. The most complicated part of kerning is that there are variables that need to be tuned based on the purpose of the typeface. If the typeface is intended for screen use or print use, or if it's made for large headlines or tiny captions, the amount and method of kerning changes, and so would the algorithm needed to generate it.

[legulere]

That sounds like a perfect match for using machine learning. The exact connection what we think is the perfect kerning isn't obvious and the model used by Adobe seems to be arbitrarily selected. Machine learning probably can approximate how a human decides which kerning is best better.

[jack_jennings]

Right, though I'm skeptical that the program would also be able to correctly identify the motivation for the typeface, which drives all of the decisions about how it behaves. That component of the program would end up being far more complex than generating the actual kerning pairs themselves.

[chrisseaton]

I can't imagine how kerning could look 'robotic'. Can you give an example?

[jack_jennings]

Not an easy phenomenon to describe or even illustrate. I believe that some legibility come through slightly uneven textures that are created by varying shapes that words create and the "color" (density of light and dark) that is created by different sequences of letters. Different words become identifiable by their shape and the visual space they occupy. Kerning somewhat counteracts this by attempting to make sequences appear more regular, so "robotic" kerning, might be described as an extemely even appearance to the detriment of legibility.

Granted were talking about a very micro adjustment. In the end he word will still be word and will still be read.