Drunk on power this afternoon, I commanded my demonic robot slave Claude to make some hexmaps. And I was impressed with what it came out with. I told it to produce a single 24-mile per hex hexmap of Japan for use in an OSR sandbox, then a 6-mile per hex map of each of Japan's regions (Kyushu, Shikoku, Chugoku, Kansai, etc.), and then 1-mile per hex maps of every prefecture. And it duly spat out amazingly detailed, usable, and - one must conceded - aesthetically pleasing stuff.
Here is the 24-mile per hex map of Japan (it even apologised that since some of the sea channels in Japan are less than 24 miles wide it was not able to map them accurately, and at its own initiative placed special sea borders between hexes at such locations):
Here are examples of two of the regional 6-mile hex maps (for Shikoku and Tohoku):
And here are two examples of individual prefectures (Kochi and Ishikawa) at the 1-mile hex level:
 |
Now, you can tell these are AI-produced (note for example how the little insert map, showing where what is depicted fits into the bigger picture, just depicts a random, sliced-off corner of Asia and then sea all around - a human being would never do that). But still, you have to grant that, with about three simple prompts and 30 minutes it is a wonder that it was able to come up with all this. And it is incredibly useful; it would potentially completely revolutionise the way I prepare a campaign, and also scale-up my ambitions for what I could create in terms of campaign settings by several orders of magnitude. I would just love to get cracking keying in those hexes.
And yet, and yet, it is funny how something so marvellous can at the same time fail to satisfy. Last weekend, I happened to have created my own map of a campaign setting, the idea for which had popped into my mind almost fully formed. I won't talk about the substance here now, but here at any rate is the (unfinished and unlabelled) map I did on Worldographer:
It took me a good few hours across two evenings - let's say, at least 6 times as long as it took Claude to map out Japan in beautiful 1-mile hexes in painstaking detail. And it looks nothing like as nice as what Claude came out with. But I was much more pleased with it, because it was mine. OK, ok, so I used a software programme to do it - I didn't hand draw it. But it was my effort and labour that went into it. And doing the coastlines for those islands, not to mention the rivers (have you tried to draw rivers in Worldographer?) was fiddly and took some time to get right. I felt good about what I had done, and satisfied. It's not perfect, and it will take more work to fill everything in and get everything how I want it. But it - forgive me being pompous - feels good in my soul to have made it.
I have over the years made a number of posts riffing on the fundamentally Satanic nature of AI art and writing. But here I come back to the main observation I made several years ago. Like the Angrborn in Gene Wolfe's The Wizard Knight, the problem, the real problem, with what Claude and its ilk produce is that, while technically incredible, lacks the capacity to be loved. As I put it back then:
We examine [what AI creates] as we would a constellation of stars - a thing that, however beautiful, is impenetrable to emotion. We do not love it in the same way that we can love a real painting by a real person, or at least recognise such a painting to be loveable in some sense to somebody even if it's not to our personal taste. We humans can love all manner of things - think of how deeply a child loves a stuffed toy, a hardworking parent loves his or her comfortable bed, a football fan loves his team, or a young person loves an item of clothing. But we do not love AI art. It is inert to our affection.
No comments:
Post a Comment