Deflation reflects the value of the currency. This reflects the value of some products and services relative to others. Obviously some things decline in cost over time -- compute is the obvious example, but just look at your own examples:

> For example, it's entirely practical right now to launch and run a greeting card company without ever hiring an artist.

In other words, the cost of commodity art has gone down.

Whether the long-term result is net deflation depends on whether the government is concurrently printing up any new currency, but in general they are, because currency deflation is bad but easy to offset by doing just that.

> mistakenly assumes that consumer products make up the bulk of discretionary spending.

Where does it assume that? What it assumes is that labor is the primary component of the cost of living -- which it is. There are some things with true scarcity in theory, but those are rarely the bottleneck in practice. It's not that we can't grow enough food to feed everyone, or build enough housing etc. -- it's that those things take labor to deliver, so if we automate that labor the cost of living goes down.

> First, that's not AI - the subject we're discussing.

We're discussing automation, of which AI is a subset. The Luddites weren't upset about AI and nobody really expects AI to somehow automate mining, but that doesn't mean mining automation isn't possible. You might even use AI to devise new technology to automate mining.

> People whose income shrinks drastically or dries up due to AI are not going to have their rent or food costs reduced,

The overall economy no longer has to pay them to do something the machine will do basically for free, so that thing will cost less. Now, this might mean that one specific person loses a $50,000 salary and in exchange the average person (including them) has their costs reduced by $50/year, because that job was only done by one in a thousand people.

But that $50,000 in total is still out there in the pockets of those 1000 people, and they're going to spend it on something instead, and that something is going to create some other new job for that person.

The person who was previously making $50,000/year may not like this. They may even end up with a new job paying $45,000/year and the other $5000 goes to someone else, even though their cost of living only went down by the same $50/year as everyone else. But each $50/year, each occupation that gets automated, adds up. And when you automate more of the old jobs it turns into $5000/year per person or more and outweighs the cost not just on average but for even the people who had to change careers.

> they're just going to have to find some other form of work to pay their bills.

But that's just what they'll do, unless there is no other work that needs to be done. Which would imply that everything should be really cheap.

> The existence of automation and industry does not mean that manual labor under terrible conditions ceases to exist.

It ceases to exist to the extent it has been automated. If you automate half of something but not the other half, the problem is not the half you've automated, it's the half you haven't yet.

And it's not as if no one is trying to increase the level of automation in general. If your objection is that nobody has automated parts of agriculture yet then forget about AI and focus your efforts on accomplishing that.