Experience so far
On the basis of lived experience I think my over all view can be summed up in that I gave using AI tools a shot back in ~2020-2021 and was generally dissatisfied with what they could give me so I chucked it. I had a go at LLMs to boost encounter tables across the pandemic - mostly attempting to increase lists - so very well 'prompted' and the effort required for setting it up and cleaning up the output was not worth it. See examples of villanous escapes, night events, living doors and mythic shoreline encounters.
Since then what is available has become pay-to-access without me seeing much benefit to shelling out that cash. Multiple other digital tools in my life have had AI forced into it along with a price hike - with a net impact of my paying out more money for a worse product. Enshittification is real. I used Meetup for organising a D&D group - they shoved in AI and massively hiked the price so that killed that. I use Evernote (same parent outfit, Bending Spoons) who shoved in AI which I am actually using for better voice transcription so I grudgingly have stuck with them through their enormous price hike.
For counter example, an actual useful deployment of neural-networks is making Voronoi diagrams you can tweak to get city districts or other regional breakdowns. I think there is potentially a very cool integration of AI-tools into helping to animate and lip-sync an NPCs portrait for online play - but would I shell out serious money for it? Probably not, it is a shiny toy, not a priority.
Promises vs reality
Where AI should be useful is as 'super search' - getting you the rules, lore-bits and build ideas that are part of the fixed 'closed system' of the game you are playing. It should be able to fish out and easily provide answers to edge-case rules questions but it cannot currently do that as testified by continuous posts on reddit boards about DMs having to argue against players relying on erroneous chat-gpt rulings. Rules-look-up should be a place for AI to excel and it does not. The solid archetype of how this stuff can be coded up is the D&D Beyond and other character builder - there are codified rules, a right answer is knowable, but "AI" as presented to us today cannot provide it for us.
Where we are getting AI is in the 'open system' - all the creative inputs that are not in the rulebooks - provision of backstories, adventure outlines, locations and image generation. For back-stories and the like "why would I bother to read something someone couldn't be bothered to write" applies. Like a player rocking up having made no effort to learn the rules or create their character, tossing me an AI backstory is just rude. Starting from something AI extruded and editing it to something appropriate and applicable, fine, I'll accept that like I would if you went to a random character generator and then tweaked what you got from that. The application of human brain along the way to fit the outputs to the purpose takes it from trash to acceptable. Creation of a backstory using a spark-list or other random generator would be similar - used raw, it makes no sense but with judicious editting to the campaign at hand, I find that ok as a DM. I do not use it myself as a player since creating a characters background is the fun to me, why would I outsource it?
For the DM side, the shorter the thing needed the more likely AI can give you something useful but it will trend to beige - LLMs generate statistically likely responses so they are going to trend towards common responses unless you spend time redirecting it - at which point if you know what you are looking for why not just draft the thing you are looking for in the first place? My experiences has been that the efforts involved have been no time saver nor have the outputs been particularly amazing compared to just hitting the blogs. I have never failed to find someones inspiring fever-dream that I can cadge for my own use - written by someone who gets the DM's task so they are ready-to-use.
For longer things like adventure paths and storylines, here is where AI falls apart because they are not good at continuity of elements - plot elements, variables, whatever - over a longer form thing. Each scene in an adventure might be ok but coherence over the whole thing is not guaranteed and you will have to put in the time to rewrite it... so why not just use actual DM's tools like those provided by Donjon, Scarlet Heroes, Shadowdark, Fictive Fantasies, Oldskull etc. Even the fact that these generators all have a structure to them shows that a tailored AI should be able to do it... and yet.
Image generation is a place with a lot of hype and promise and so far all it turns up is rarely better than what can be found with a search of pinterest. If you are playing in standard issue high fantasy space where art that would look at home in World of Warcraft is appropriate then fine you may well get some generic outputs that look reasonable - for a PC it might be worth it. But as a DM I have given these image generators a go and I am better off sketching what I want myself or rummaging around on image search.
I canvassed a couple of the discords I am on, pools of players and DMs and I got two use examples and a wish - a fair few people were using image generation for PC portraits and the like, there was one using it for random NPC names and people were interested in the 'super-search' potential of AI that could be achieved by feeding in their game notes and being able to query them. This last was a 'wish' because they did not want to use the current tools which to their mind would just ingest their stuff and throw it on the big heap of all other content to pop up in others searches. For everything except image creation it was viewed as 'not worth the trouble'.
Potential of AI in TTRPGs
I wrote in 'my own personal OSR' that the reason I am here in this corner of the hobby is for the shared creativity and the generous provision of tools and tricks by participants made freely available on their blogs to enable other to benefit from their own flights of creativity and hard-won lessons at table. One might think that therefore I would be all for AI as an aggregator of all these tips and tricks, one that could build further upon them to help us reach hither-to-unreachable heights, no?
Well, so far noone has been able to put such a thing in my hand and I am mostly doubtful that they are ever going to manage to before the wheels blow completely off this AI bandwagon and all these things implode like so many billionaires badly designed submersibles. At time of writing, questions are being asked whether and how generative AI is going to justify the vast investment poured into it.
To get technical - what we have at the moment are a collection of large language models that give you statistically probable strings of words based of ingesting a vast corpus of data. The fact that they fail at math, coding or anything else except by blind luck is entirely in line with what such models can do.
Given that we play games that have rules both written and un-written, one could with enough effort set up your D&D AI - with the big caveat that any given tool would need a huge amout of opt-in/out tuning for any given module given the amount of house ruling and campaign-specific procedures that people, even individual DMs do between different campaigns with different aims.
Note, much in the way that the money blown in the dot-com boom laid the fiber-optic infrastructure on which a later generation of internet rose, I am willing to believe that perhaps a decade hence, exactly the AI tools that I want to see will rise from the ashes and be available on our phones/brainjacks/whatever - but not yet, not by a long shot.
Not for me, not for now
All this is from the ruthlessly utilitarian perspective of 'does AI today make my job as a DM easier' and the answer is no. At best the time I spend wrestling with mediocre tools to get something useful is break-even with what I can do myself. It gives me a route to easily make ok-to-good enough imagery for things but I am not finding that a game-changer and I would not be bothered shelling out for it. I got more joy out of Stuffed Crocodiles public domain token collection than anything I've managed to get the machines to extrude. Your mileage may vary, this is where I have gotten to.
Post script: The Greater Context
All that sets aside the massive theft of materials underpinning AI training of models and the blithe indifference of AI tech leaders to human suffering of job destruction and environmental cost. This also ignores the signs that use of AI atrophies your capabilities to be creative and indeed think.
Work at the interface of automation and engineers has been the majority of why anyone has ever cut me a paycheck and the rules-and-processes elements of D&D means that it should be eminently doable to make a 'DM's assistant' super-search like tool from a technical perspective. I offer no opinion on whether anyone could ever break even on the costs of doing such a thing but it should be doable. All the LLM generalist tools we have available now are just not reliable when you need to be sure of something like a rules call; hallucinations make them no better than your table buddies fuzzy memory - and that is available for free.
So far the use I have seen of AI in the TTRPG space has been to generate mediocre graphics to whack on the front of AI-generated shovelware as part of what strongly resembles those 'passive income' hustles - DrivethruRPG has drowned in them. All AI has done is allow people to jam all the channels we use to talk about the hobby with stuff not worth reading. Thus far, the introduction of AI has inflicted that cost on us while bringing little-to-no benefit. As the saying goes, don't piss on my leg and tell me its raining.
No comments:
Post a Comment