> The bias problem is also relatively easy to solve
No, its not.
> (Midjourney has already made massive improvements)
Maybe. But it hasn't come anywhere close to solving it even in its domain, which is probably the least concerning domain of AI bias, and not necessarily transferrable.
Actual ML systems that are deployed in production by governments in important roles have massive bias problems, as do SOTA LLMs (including, very much, GPT-4.)
> The AI ethics people have had shockingly little to say on copyright/job loss issues.
The AI ethics people have a lot to say about the first (well “ethics of sourcing without consent” is probably more to the point than copyright, copyright is a component, but that’s more legal than ethical).
They have some to say about the second (well, again “job loss” is the wrong framing; “skill devaluation” is probably more on point), but that’s frankly not an AI issue, its an economic system issue and effects all technological change the same way. Solving capitalism is mostly out of scope for AI ethics. EDIT: Specifically, increasing material output for labor input ought to be, in and of itself (and leaving aside ethical questions of how you do that, which the sourcing issue addresses) a good thing; if there are material losers in that, it is because the economic system to distribute output well, which is not an ethical issue of the system which enables the productivity gain but an ethical issue of the economic system.
> Large organisations systemically overestimated the risks from bias
No, they didn’t. Those aren’t even really risks any more, they are massive current costs of existing adoptions.
> and underestimated the actual AI risks that society actually cares about.
That society (as weighted by social power) doesn’t care about the kinds of bias problems AI manifests is exactly why it is an ethical issue.
> I think the answer is also simple, because accused of racism, will cost any executive in a large company their job.
The large companies that have supplied systems which manifestly suffer from racial and other class biases and which are, in fact, being used in production to implement government policy around the world have not lost their jobs, and neither have the government officials procuring them and responsible for the programs they are deployed in, so this is clearly false.