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

[Oofta]

This is how AI will integrate into Dnd, probably in this order.

1) The more advanced random generator. Able to create specific npcs or magic items, or descriptions of random places. This is where ChatGPT is now. You could also use it to create art for specific monsters.
2) Combat Assistant: The combat side of the game is fairly rigid compared to a lot of other more free-form elements. You can probably have the AI roll dice, have it run NPCs and monsters, and do a pretty solid job with some DM oversight.
3) DM with a human on the wheel. The AI takes over DMing, but a human has the ability to override it when it goes off the rails a bit.
4) Fully autonomous DM.


Now for people thinking chatGPT is still just a "dumb robot", you aren't paying attention. These thing has exploded in sophistication in the last few years, and its passing turing tests in certain areas left and right. Sure its not a novelist, but it absolutely can right things as well as bog standard humans can. And that's just now. Give it a few more years of tuning and sophistication, and suddenly its the top of the high school class. A bit more, best college writing graduate. And a bit more after that, it surpasses human writing entirely.

What people underestimate is the speed of progress. This is an exponential curve, and the snowball is starting to roll down the hill. Its going to pick up speed fast.

Maybe AI will advance quickly, maybe not. What we call AI are better described as exprt systems. They mimic intelligence and can absorb patterns and spit out variations of the pattern.

What they are not, what still be decades (perhaps a century) away, is actual intelligence.

They are getting better at mimicking intelligence. That doesn't mean that they think or that the approach will ever lead to true intelligence.

 

[The Shadow]

I mean, I already believe that about human consciousness.

Really? Who, exactly, believes that?

Says the biological organism that grew from a single type of algorithm.

Now you're equivocating on "algorithm". I would invite you to state the single algorithm by which I have grown. Good luck!

But in case... ChatGPT doesn't even have the means to try to assess whether what it's saying is true or false. Until that day comes, this sort of algorithm will be innately limited.

 

[Hex08]

That's the thing - even glancing through you can see it's drivel/verbal diarrhoea. None of the "arguments" it presents are real counter-arguments

That seems an awful lot like to talking to many real-life people.

 

[Maxperson]

thinking...

So you're saying that the BingBot is fully human, then?

Okay, that one made my week.

....you would think that, wouldn't you?

Workout Programs

The most popular bodybuilding message boards!

forum.bodybuilding.com

And I still hate you for making me see that a year or two ago.

 

[Maxperson]

2. Words. I am using this loosely, but an AI would be able to generate anything from custom adventures to custom subclasses for D&D (the singularity will occur when someone makes the mistake of asking an AI to generate the best possible bard

We will know that your doomsaying is coming to pass if that is asked of it and we end up with AI generated tombstone art. For then you will truly have been replaced.

 

[Clint_L]

Really? Who, exactly, believes that?

Now you're equivocating on "algorithm". I would invite you to state the single algorithm by which I have grown. Good luck!

But in case... ChatGPT doesn't even have the means to try to assess whether what it's saying is true or false. Until that day comes, this sort of algorithm will be innately limited.

1. I do. It is fairly conventional thinking among neuroscientists that there is no particular seat of human consciousness and what we experience as the self is an emergent phenomenon.

2. I'm not equivocating at all. Evolution through natural selection is the single type of algorithm that has produced all life we know of.

3. ChatGPT seems to be able to assess truth or falsity based on the consensus truth test. As far as its limitations go, I suppose everything is innately limited, so I'm not sure what you are specifically arguing. I note that folks have been arguing against this sort of brute force approach to AI for years, yet it seems to be winning out. The proof is in the pudding.

 

[Ruin Explorer]

Interesting. If I had to categorize ChatGPT as a personality, I'd say it's unfailingly eager to please, so much so that it will even make things up to try to keep you happy. And that it is terminally unable to commit to anything, it constantly hedges and hems and haws - it comes across like it's terrified of offending you by saying something you don't like.

Yeah I should be clear I'm talking primarily about Bing, who is a horrible stubborn lying jerk:


I would say ChatGPT is also intellectually dishonest because of the extreme evasions it uses, but that is in a different way.

 

[Hussar]

Fascinating article I read the other day concerning AI translation. The gold standard for language translation is about 1 second per word. Actually, let me clarify that. When a human translator looks at a translation and then has to edit that translation, it works out to about 1 second per word. Machine translation ten years ago was about 7 seconds per word. It's now about 3. They figure it will be about 1 second per word by 2030. IOW, AI linguistic translation, one of the most difficult tasks that most humans cannot even master, will be on human level by the end of the decade.

Discover How Close We Are to AI Singularity

CEO Marco Trombetti previewed the company's discovery during the opening keynote of the Association for Machine Translation in the Americas annual conference.

translated.com

Dunno if it's true, but, it's interesting.

Or, I was watching The Last Leg last weekend and they talked about how one of the jokes in the segment had been written by an AI and asked everyone, including the other comedians on the panel, if they could spot it. None could. They all thought a different joke was the one and it was written by a human. Granted, the AI joke did thud pretty hard because it didn't make a lot of sense, but, the point is, a hundred or so people and three comedians couldn't spot it.

 

[bloodtide]

I don't see an AI DM in any of our lifetimes. Not for any "deep" type of Role Playing.

It will be able to do simple RPGs, as that would just be a RP Video Game. So the video game fans will be just find playing D&D with an AI, as it will just be a video game. Exactly like a video game, the AI can do anything that it is programed to do. The AI can follow the rules the way it is told to, and will "do" whatever it is told to do.

And not just video game fans, but a fair number of mechanical straightforward games would love the AI game. The AI says the room has a door, the player will say "I use the open door action", then the AI rolls some virtual dice.

And, really, more then a couple DMs do run their games like that. Plenty of DMs and players like the simple, direct, straightforward mechanical game. AI will do that fine.

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'.

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......

 

[Stalker0]

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'.

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.