D&D General Deep Thoughts on AI- The Rise of DM 9000

[Hussar]

Of course, there is no way an AI could not Railroad for the simple reason as it is only programed with so much. You sit down to play the adventure Dragon Heist, the AI can only respond to things in that adventure. You can't leave the set game area. Much like video games you would at best get a dead end, where you can walk around in the wild forever and not advance the game....and at worst the AI might just say "BEEP! You can't do that!".....or like really, really worse is if it has Railroading built in and would say "It is raining to hard for you to leave the city" or otherwise just 'block'.

Honestly, I'm not sure that's true. Considering the MOUNTAIN of available information about Waterdeep (for example) on the Internet, I could easily see an AI that could run a canon based open world game in Waterdeep. I mean, good grief, we can do that now practically without AI. And, again, say you start in Waterdeep, fair enough, but, if the AI has access to the FR wiki, and the ability to generate adventures (which isn't all that far fetched, we're seeing the beginnings of it now), suddenly an AI generated Open World Forgotten Realms game isn't all that out of reach.

/edit to add, it's hilarious that @Stalker0 and I both chose the exact same quote to reply to without seeing each other's replies.

 

[OneRedRook]

And sure, you could go to any offical encounter point in the adventure in Waterdeep....but you can't really explore the city or area. You can go to a programed, set location....and maybe there would be a random generator, but that will give you video game results.

And you could only take set programed actions. You use the skill Open Locks on the Locked Door. But if you type "my character climbs in the window" you will get "BEEP you can't do that!" unless THAT move is programed in.....but there is simply no way to pre program the infinite things a player might try or do.

And you can only have things programed into the game. A player that says "I go to the store and buy a spade" can only do so if the AI knows what a spade is, and can only use it if the AI is programed as to what it does.

So......

Unfortunately, whilst I'm sure there are people in this thread who are both more qualified that me to talk about ChatGPT, and know more than I do about Forgotten Realms lore, it apparently won't stop me from commenting here.

The quote above I think shows a significant mis-perception about what ChatGPT is doing. It doesn't need to start with any internal "definition" of objects, actions, etc; instead, it's already been trained on an astonishingly large amount of text, and at the risk of being a bit reductive all it is doing is finding the most likely word to put next, one after the other. That's it.

So the problem that an "AI GM" running with something like ChatGPT as a back-end is not that "my character climbs through a window" hasn't been "programmed" as a one of a series of actions that can be taken, leading to the AI GM blocking - I think as long as you're describing actions that fall within what it can "infer" from its data-sets, you'll get an answer. The problem is going to be that it will sometimes start generating text that doesn't align with with what's been generated so far. And that, I suspect, is a much more tractable problem to fix.

For example, I asked it: "Can you describe a journey from Waterdeep to Baldur's Gate, including the route taken?". The response was:

Certainly! The journey from Waterdeep to Baldur's Gate can take several different routes depending on the mode of transportation and any specific stops along the way. Here is a possible route by foot or by horseback:

1. Leave Waterdeep: The journey begins in the city of Waterdeep, which is located on the western coast of Faerun.
2. Head South: From Waterdeep, travelers would need to head south along the High Road, which is a major trade route that runs through the heart of the Sword Coast.
3. Cross the River Chionthar: After traveling for several days, travelers will reach the River Chionthar, which runs from east to west across the Sword Coast. They will need to cross the river by ferry or by a bridge.
4. Reach Baldur's Gate: Once across the river, travelers will continue south along the Coast Way road until they reach Baldur's Gate, a bustling city on the southern coast of the Sword Coast.

The journey from Waterdeep to Baldur's Gate by foot or by horseback can take around 10 to 14 days, depending on the weather, the condition of the roads, and any delays or detours along the way. There are also other possible routes, such as following the coast by boat or traveling through the wilderness, but they may take longer and can be more dangerous.

On a surface level that's not a bad answer, except that it's wrong: for those unfamiliar with Forgotten Realms as I am, Baldur's Gate is actually on River Chionthar; I think the name of the road is also incorrect. (Notably, it's actually not a bad description of the journey from Waterdeep to Candlekeep). The thing is you can just tell it the correction and it will give a better description.

So I think this sort of thing has the potential to increase in use and viability quite quickly.

As an aside, I did in fact supply that correction - I shall have to see if that correction has affected its global state, or if it was just valid for that specific "chat" instance.

 

[Paul Farquhar]

What I find interesting ChatGPT is what it tells us about humanity. Notably, it illustrates how it is possible to speak with apparent authority about something which you have no understanding of whatsoever. It explains how a lot of humans manage to get promoted.

 

[Oofta]

I think your greatly underestimating an AI's potential ability. Even now ChatGPT can answers hundreds or even thousands of questions on all shapes, sizes, and topics. And this is infant stages. I have full confidence that in the longer run AI could absolutely handle a very wide menagerie of human ideas and actions in a game.

The AI can find and regurgitate answers previously given by humans. It doesn’t really understand the question or the answer. It's just analyzing one pattern (the question) and matching that to the response answer patterns that it's been given to find one that matches.

Given enough statements, enough input and answers AI does a decent job of finding the corresponding output. It doesn't understand anything. That's going to be the limitation.

In a game I understand the act of climbing, what a window is and, just as important, which window. AI can "create" art because it's been fed millions of examples. It's not going to be fed millions of scenarios that we encounter while playing D&D.

 

[Yora]

AI is a marketing term. There is nothing intelligent about it.

 

[Oofta]

Honestly, I'm not sure that's true. Considering the MOUNTAIN of available information about Waterdeep (for example) on the Internet, I could easily see an AI that could run a canon based open world game in Waterdeep. I mean, good grief, we can do that now practically without AI. And, again, say you start in Waterdeep, fair enough, but, if the AI has access to the FR wiki, and the ability to generate adventures (which isn't all that far fetched, we're seeing the beginnings of it now), suddenly an AI generated Open World Forgotten Realms game isn't all that out of reach.

/edit to add, it's hilarious that @Stalker0 and I both chose the exact same quote to reply to without seeing each other's replies.

Generate setting, maps, NPCs? Sure. Actually running the game? That would require broad understanding of how humans perceive and interact with the world around us. I don't see that happening anytime soon.

 

[Bagpuss]

I'm sorry Dave. I'm afraid your character can't do that.

But don't worry, with AI advancing so rapidly, you can always create a new one that's better equipped for the job! Maybe one that's programmed to have better luck or more skill points in computer science.

 

[Bagpuss]

AI is certainly an exciting and rapidly evolving technology, and its impact on our lives is growing every day. As Snarf points out, the rise of AI-generated art and text has been particularly noticeable in recent months, with programs like ChatGPT becoming increasingly popular and accessible.

However, as Snarf also notes, with this increased adoption of AI comes concerns about its potential impact on society. For example, the ability of AI to generate written text raises questions about its impact on the job market for writers and journalists. Similarly, Snarf's anecdote about the use of AI to write a formal protest letter raises questions about the role of AI in personal communication and the potential for misuse.

These concerns are not unfounded, and it is important that we as a society consider the ethical implications of AI as it continues to evolve and become more integrated into our lives. At the same time, it is also important to recognize the potential benefits of AI, such as its ability to generate new forms of art and to assist with tasks that are difficult or time-consuming for humans.

Overall, the rise of AI is a complex and multifaceted phenomenon that will undoubtedly continue to shape our world in the coming years. By engaging in thoughtful discussions about the potential benefits and risks of this technology, we can ensure that we use it in a responsible and ethical manner.

 

[Bagpuss]

AI is a marketing term. There is nothing intelligent about it.

I respectfully disagree with the statement that "AI is a marketing term. There is nothing intelligent about it." While it is true that some companies may use the term "AI" as a buzzword to promote their products or services, this does not negate the fact that AI technologies have the potential to perform tasks that would typically require human intelligence, such as natural language processing, image recognition, and decision-making.

AI refers to the simulation of human intelligence in machines, including learning, reasoning, and self-correction. While current AI systems are not yet capable of replicating the full range of human cognitive abilities, they have already demonstrated impressive capabilities in many domains. For example, AI-powered chatbots can understand and respond to natural language queries, AI-powered recommendation engines can predict user preferences and provide personalized recommendations, and AI-powered fraud detection systems can analyze large volumes of data to identify suspicious transactions.

In conclusion, while there may be instances where the term "AI" is used in a misleading or exaggerated manner, it is undeniable that AI technologies have the potential to perform intelligent tasks that were once exclusive to humans.

 

[Snarf Zagyg]

The AI can find and regurgitate answers previously given by humans. It doesn’t really understand the question or the answer. It's just analyzing one pattern (the question) and matching that to the response answer patterns that it's been given to find one that matches.

Given enough statements, enough input and answers AI does a decent job of finding the corresponding output. It doesn't understand anything. That's going to be the limitation.

In a game I understand the act of climbing, what a window is and, just as important, which window. AI can "create" art because it's been fed millions of examples. It's not going to be fed millions of scenarios that we encounter while playing D&D.

Not to get too philosophical here, but I think you are treading close to the Chinese room argument.

At a certain point, you have to ask, "What is understanding." If an AI can understand natural language, can respond in natural language, and can create art- then how is that materially different than what we do?

Put another way, you say that all the AI does is a decent job of getting input and finding the corresponding output. Which is true! Arguably, of course, that's what we do too.

The idea that certain things (art, natural language, writing, coding, and so on) were the "special sauce" that we had over computers, unlike, say, chess (Deep Thought, get it?) was something I assumed would be the case for a long time- after all, any experience with Siri or Alexa would quickly disabuse you of the notion that there was much there, there. But that's why this is different.

At this point, we are just seeing the beginning of the public-facing aspects of the new technology. But these things happen quickly. Again ... we are already seeing that they can easily write essays that are better than those written by most high school students. They create art better than that of the vast majority of people (who can't do art at all). And unlike us, they are advance quickly and are scalable.

So ... I am less confident than you are. I would be shocked if we don't have AI-capable DMs within five years- even if that isn't the use case for them. And that's not a statement I would have made a year ago.