Freelance Traveller

An AI Chat about NPC Motivations

by Timothy Collinson

This article originally appeared in the May/June 2024 issue.

I recently needed not just one but half a dozen or more NPC motivations. One is easy enough but faced with several, at that point I start thinking about random generation to save some brain power and to ensure that I’m not stuck in the same old mental ruts of habit or biasSee for example ‘Beyond our Ken: Describing Research Bases in Traveller’ in Freelance Traveller, no.118, produced for exactly the same reason. .

As it happens, Traveller has just such a thing in Traveller: The New Era with its ‘NPC Motivations’ in the core rulebook (pp.60-61). Here you can find a system that uses a deck of cards to produce motivations in four ‘suits’: Clubs: Violence, Diamonds: Greed, Hearts: Sociability and Spades: Ambition.  The idea is that you draw twice and have a major motivation and a minor motivation. Lower numbers represent less important drives.

I recall thinking it was an excellent idea when I first read it but being unhappy with its execution for various reasons.

Things I disliked:

• the system not being dice based, so I’d have to go off and find a pack of cards when I just wanted something quickly; invariably this would also mean a game of Patience to check that all the cards were actually present
• cards being, for me at least, hard to randomise quickly – it takes a while to shuffle a deck sufficiently such that I’m happy with it (poor shuffling skills, I admit)
• it purporting to generate a ‘scale’ in each suit via the numbers but then using the face cards rather randomly to insert other motivations that might not be at all related to the ‘suit’
• two of the four being ‘negative’ rather than neutral
• only having four suits when the range of human motivation is much wider (although the face card ‘wildcards’ somewhat ameliorated this)
• some of the drives not really being motivations but attributes.

None of them deal breakers perhaps for something that is just supposed to be a quick and dirty way of adding a bit of life to NPCs, but I wanted something more, at least for characters that might be around for a little longer than ‘generic publican’ or ‘bored shop owner’. On one occasion a long time ago, I tried to come up with something based on more traditional Traveller D6 rolls. I wasn’t happy enough with that to keep it (or maybe I have but have no idea where – see below).

Revisiting motivations

I thought I would have another go at my own version and, trying to keep it compatible with the TNE table but expanding it, produced a 2D6 table with six columns of Ambition, Sociability, Physicality, Acquisition, Justice and Honour. I tried to reduce the ‘negative’ connotations and come up with motivations that were neutral such that low numbers would be the negative aspect and high numbers would be the positive aspect. This confused two things: strength of drive and direction of drive so it would need some work on it (or perhaps simply two rolls). Onto this table I then mapped the face card values from the TNE system just for interest to see where they lay. ‘Wise’ was rather hard to place, but then as I said above, some are more an attribute than a motivation. A couple of others were similar. Is ‘violence’ actually a motivation?

I was relatively happy with this and going to replace the TNE version in my ‘referee’s handbook’ of quick reference tools such as this (a document wallet collecting print outs of various rules and tables from Traveller books and other sources). It just so happened however, that I spent a work day in a webinar on AI. At one point we were given a short break and asked to use a large language model to produce a bit of text for later discussion. With motivations fresh in my head I thought I might as well see what ChatGPT 3.5 came up with.

Now, when it comes to creative writing for Traveller I avoid generative AI like this for the pragmatic reason that it tends to be a bit bland and limited and the ethical reason that, like the complaints in the art world, it may well have been trained on material acquired without permission. Not to mention that as a librarian in an academic environment, I’m aware that commercial companies aren’t going to share what data they’ve used for training and how the algorithms work – with the possibility that they introduce (or exacerbate) bias. Plus, for me, half the fun of creative writing is the creative part of it. I’ve not yet struggled with writer’s block – I recently measured my ‘in progress’ folder of possible Traveller ideas and articles at over 12GB. If I’m ever wondering what to write, I’ll just browse through that and pick something up I’ve long since let fall by the wayside.  [Writing this, I’m now wondering what on earth, or not on earth, is in that folder? I really ought to have a look if not a weed.]

A 1980s list

Anyway, my first prompt looked like this:

TC: Hi there, please can you imagine you’re a role-playing game designer in the 1980s trying to come up with some categories of motivations for non-player characters? Say half a dozen to be chosen by a single die?

I was curious to see what it might come up with had this been part of the original (or early) classic Traveller rules. You’ll also note that I don’t seem to be able to resist being polite just in case AI takes over and I want our overlords to remember me fondly…! Clearly, I also take the ‘chat’ in ChatGPT literally. The results can be seen in Table 1. One of the keywords in the prompt is ‘categories’. I didn’t just want a handful for specific characters; I wanted something of broader use.

Editor’s Note: The text of the ChatGPT response has been left as provided; only the formatting has been changed.

I quite liked the AI’s role-playing introduction which was nicely in keeping and I was impressed that it bothered to number the results and stuck to the ‘half dozen’ exactly – but perhaps that is just its literal mindedness. (When I use words such as ‘mindedness’ or ‘role-playing’, I’m aware that ChatGPT is doing no such thing, but is simply very good at predicting next words or phrases having digested large chunks of the internet). I liked the way it divided each entry into ‘description’ and ‘behaviour’ – which I hadn’t asked for but didn’t mind and thought was helpful. I particularly like #5 in a Traveller context. I thought the results were pretty good as a match for the kind of thing you might have seen in White Dwarf or Dragon, or perhaps even JTAS at the time.

Moving on to the 1990s

So, now I asked it a question which I was then curious about:

TC: That looks really good and just as it might have been done back then, thank you. Now can you imagine you are a role-playing game designer from the 1990s when RPGs took a turn for the darker and do the same exercise, please?

On reflection, the last part of that was perhaps a leading question and I wonder if I had not put that whether it would have produced the same results. In any case, the results can be seen in Table 2. At this point I was very taken with its immersion in the role and with the results it offered. I can’t say I was role-playing in the 90s much – having children will do that for you – but this really seemed to capture what I’ve read of games from that era. It was also interesting how little overlap there was with its first set. I had thought that perhaps it would just run ‘darker’ variations on the same themes. Only vengeance directly maps although you might argue that greed and obsession are closely related.

“Contemporary” RPG Motivations

Now I asked it the question I really wanted the answer to (bear in mind that all this took just a few seconds – only as long as it took me to type the text of the three prompts). The answer can be seen in Table 3. The only change I have made to the article from the response ChatGPT gave is the addition of the left most column. I’ve subsequently added the text in italics to make the table’s use a little clearer.

TC: Excellent, that captures the era nicely. Now, how about something that’s up-to-date for the 2020s please.  Ideally a ‘comprehensive’ set of motivations like that, either six or twelve if you prefer.

At this point I really was quite blown away by the response; even though technically it wasn’t a question. Firstly it ‘chose’ to offer twelve rather than six options. I have no idea why it did this unless the prompt pointed it in that direction. Secondly it again seemed original in terms of not rehashing the two sets of six it had already given me. I should say that I have no idea how original in it is being in terms of RPG motivations for games which offer them as I have not until recently read widely outside of Traveller. Thirdly, once again I felt that it had captured something of what I see in newer games or in discussions on social media. It felt to me, perhaps a naïve gamer, that these had a modern sensibility to them which I would be happy to use as given. Fourthly, I was rather taken with the way the three results looked almost as if the AI had guessed what was coming and planned ahead. Not only did each answer seem complete, all three together seemed a coherent whole.

It is tempting to tweak and edit the set of three results or to lump them together for a larger number of options but I think it is actually more interesting to see what ChatGPT presents as its idea of game ‘development’. Again, I appreciate that it is doing no such thing, but simply interpreting my prompts according to algorithms and, in this example at least in my opinion, some excellent ‘next word’ prediction. Its introduction and tag to the set of 12 are also part of Table 3.

Any Sources?

Finally, I had a short conversation with it to see if I could get it to reveal its ‘sources’. This can be seen in Table 4. While I’m not entirely convinced it wasn’t giving me a curated cop-out of “there isn’t a direct influence” to avoid any issues around copyright, it did still name some examples and again, with my limited knowledge, this seemed to be a fair representation. I thought it was a reasonable analysis.

Unfortunately, my ‘play’ time in the AI webinar was over and it wasn’t until later I was able to come back and ask the obvious follow-up question about what might such a table of motivations look like in RPGs in twenty years’ time. Unfortunately, the answers were not as impressive. It gave a reasonable response on what developments there might be in computer RPGs but it seemed to have lost the connection with actual categories of NPC motivations rather than how NPCs might work in the future. It wasn’t uninteresting but despite a couple of attempts at rephrasing the question and being more specific about pen-and-paper games, it wasn’t really what I was after. (Though that may be my narrow focus and missing the value of the results it did give.) I don’t know if this was because my prompting was bad, its ability to ‘predict’ is poor or whether it had lost the ‘thread’ because I was now in a different session of ChatGPT and it had no memory of all the above. It may be a combination of all of those. Perhaps it is worth rerunning the experiment in one go but personally I’d rather be scribbling down new ideas, writing my own words and creating my own things so I’ll leave that for others who are interested.

In any case, I thought this was interesting enough to share and perhaps by including the prompts, any ‘letters to the editor’ Jeff receives might point out where I could have done better for next time. In the meantime, I hope some of this helps with producing more rounded NPCs in your games or provides some inspiration for your writing. I might tuck all three of the motivation tables in my handbook. The ‘dark’ ones might particularly suit antagonists of the nastier kind. Perhaps it might even help players flesh out PCs and add that extra soupçon of flavour to their gaming and motivate them to be that much more interesting.

Postscript

Having written all the above and submitted it to Jeff, our kindly editor then reminded me of his own article ‘It’s In the Cards: Character Motivations for Traveller’ which I’m ashamed to say I must have read but had completely forgotten about. Doubly shaming because it was not only published just two years ago but also a decade previously. (July/August 2022 and March 2011 respectively). (This is one reason I do my bibliography work, to try and keep track of the wealth of Traveller material there is so much of it. But of course, I’m way behind on adding Freelance Traveller entries – mainly because the search feature is so good it’s not a priority).

Jeff charitably pointed out how similar the lines we’d been thinking along turned out to be. I had to point out the other possibility, that my subconscious had done a better job of recalling his previous work. Although it should be said that my very first go at an alternative to TNE was a couple of decades back. Anyway, my apologies to Jeff although I think I’ve perhaps added something novel with the AI generation (as well as moving away from the limits of cards). I’d still warmly recommend his article for some useful and interesting ideas and, of course, yet another set of potential motivations. It also prints very nicely to a four page ‘booklet’ to tuck in my files so hopefully I won’t forget it again; perhaps I can use it as a bookmark in my TNE rulebook.

Finally, two things have just arrived in the post.  Wired for Story by Lisa Cron (Ten Speed Press, 2012) recommended by Jim Vassilakos, mastermind behind the Plankwell Campaign in Alarums and Excursions as well as a denizen of TML. Although it is not Traveller, it really excellent on character motivation and worth reading whether you’re creating a player character, NPC or writing your novel. Also, Anders Backman happened to post this on Vector Movement just in time for me to catch it, and which is particularly interesting because it suggests how characters might respond to certain motivations and also offers some default motivations that Traveller races might have. He also suggests motivations, under his scheme for the Firefly characters and some selected real-world people.