Just How Smart Are the AI Characters in Inworld's 'Origins'?
Вставка
- Опубліковано 27 тра 2024
- Earlier this year, the AI company Inworld released their tech-demo 'Origins': a detective game built in Unreal Engine in which you try to solve a mystery by speaking directly with your suspects.
Powered by Inworld's character engine, each non-player character you meet in Origins is built to support full voice-based conversations using the latest trends in generative AI technology, each with their own personalities, goals and mannerisms.
In this first of a two-part series, I sit down and give my first impressions of Origins, with a future video detailing how it all works, with insights from Inworld themselves.
You can try it out yourself by downloading Origins for free on Steam:
bit.ly/AIandGames-Origins
And if you're a developer yourself, sign up for a free trial of Inworld's tools, and start building your own conversational AI characters
bit.ly/AIandGames-1
[00:00] Intro
[02:15] Who Are Inworld?
[06:35] Analysing the Demo
[10:50] Text Generation
[13:33] Voice Recognition
[15:12] Voice Synthesis
[17:20] Closing
[19:12] Credits
Special thanks to @inworldai for sponsoring this episode. I was compensated by Inworld for time spent producing this video.
--
AI and Games is a UA-cam series on research and applications of Artificial Intelligence in video games.
Follow 'AI and Games' for educational videos and essays:
/ @aiandgames
My second channel 'AI and Games Plus' is dedicated to bonus content, deep-dive analysis, revisiting old topics and more:
/ @aiandgamesplus
Listen to my podcast 'Branching Factor', as I discuss game development and research with my co-hosts and guests:
shows.acast.com/branchingfactor
Watch Livestreams: / aiandgames
Join our Community: discord.aiandgames.com
Buy our Merch: store.aiandgames.com
--
You can support AI and Games via Patreon and UA-cam memberships, and in return, you get bonus rewards!
/ aiandgames
www.youtube.com/@AIandGames/join
--
Follow 'AI and Games' on social media.
Twitter: / aiandgames
Mastodon: mastodon.gamedev.place/@AIand...
TikTok: / aiandgames
Instagram: / aiandgames
Threads: www.threads.net/@aiandgames
BlueSky: bsky.app/profile/aiandgames.b...
Substack: www.aiandgames.com
--
If you're interested in working with us, check out the following:
Business Enquiries: contact@aiandgames.com
Arrange Meeting: calendly.com/aiandgames
#inworld #microsoft - Ігри
![[柴犬ASMR]曼玉Manyu&小白Bai 毛发护理Spa asmr](http://i.ytimg.com/vi/0TsXQ7z2Dh4/mqdefault.jpg)
Origins is out now on Steam: bit.ly/AIandGames-Origins
Plus, sign up for a free trial of Inworld's tools, and start building your own conversational AI characters:
bit.ly/AIandGames-1
I'm already using the tool to create my game! Let me know if you are interested in knowing more about it.
I feel like this kind of tech in its current state might work better in a more visual novel style of presentation. Removing voice acting from the equation and tying the dialogue to a set of varying character portraits rather than full animations seems like it would be a lot easier to sync up properly.
Feels perfect for the next Quantic Dream game.
This! Until they improve on the limiting factors I think that's the way to go.
Then you can have stuff like AI Doki Doki Literature Club.
Very cool.
the response time for speech recognition followed by llm followed by generation seems very impressive. Looking forward to the followup to see how much is local vs remote and if its remote, how they’ll handle a successful game launch traffic
I spent about 30 minutes talking to that officer about breaking bad. She knew everything about the plot, it was just crazy
Seriously??? That is so damn cool
It's probably a good thing there wasn't any burglar alarms going off as a result of the blast.
Especially the purple ones...
A neat if unimpressive game as a player who's mostly in games for narrative rather than "cool that tech can do this"-reasons. From that perspective the stiff delivery is bad enough, but also the game itself feels like it solves a "problem" that's really just a problem in the eyes of money-infusers who wants a product to be generated as quickly as possible, instead of a creative mind who wants to take their time making a good story that people enjoy.
Man I can't wait to see what this will be like in 10 years, then 20.
When the voices sound fluid, when the animations of the NPCs move more lifelike, have perfect mouth/voice synching, it's going to be Soo damn cool.
And especially when smaller studios someday have access to stuff like this, to flesh out a world in a way that usually only big studios are able to.
I did a video on inWorld origins six months ago and the voices were quite good back then. InWorld have gone through a number of voice models over the past few months and they may have had to switch to a cheaper voice engine by the time you made your video.
Great video. Really excited to see where this tech leads
Still lacks Dachshunds, I see.
He's gonna keep looking though. Don't worry. We'll find them.
gotta say while the demo is pretty wonky i kinda found it an AMAZING concept. just IMAGINE what this could be in a few years. and how INTEREACTIVE you can be in games. like a dective story you can actually ask questions and get feedback and what. it would be awesome! and a rpg. where you can connect and bond with charecters or a life sim! i find it pretty amazing!
Very interesting! I'm looking forward to seeing more games utilize this technology to make NPC's more lifelike and interesting to talk to.
Please talk about the work of Chris Crawford on Interactive Storytelling and how it can be applied on today's AI technologies.
I doubt Chris is gonna answer my emails anytime soon.
@@AIandGames Oh that's not true! I was a student of his. He's super accessible.
Can't wait for NPCs in games to start reacting terrified to glitches that happen around them XD.
Oh wait, imagine NPCs that can learn to exploit bugs...
Generative AI (obviously once significantly more developed past its current state) has insane potential for emergent behaviour in games.
Presumably this game is always online? Or is all the language processing done on your own computer?
Yes, as stated in the video it requires an internet connection to interface with Inworld's character engine.
I feel like this will put a real blockade to use it extensively. @@AIandGames
You're correct in that the big issue with language models is the need for online connectivity. The (large) language models are simply too big to store on a local device (and Inworld runs several at once).
While I know that sponsors are like important for getting money, it still feels like a conflict of interest when you make stuff like this and the convai videos while being directly sponsored by the companies that are making the product. I don't feel like I can really trust it, which is disappointing as you have previously had good unsponsored videos on the use of generative AI in video games, covering both the potential and shortcomings as well as their impact on the broader development pipeline while this just feels closer to an advertisement.
I can appreciate that. When approaching topics like this it's important for me to distinguish between the viability of the product on a technical level while also still being cautious of the surrounding issues it raises throughout the industry. After all as I said in my Artifacts episode I am excited by the opportunities generative AI presents, but quite often I find the hype is seldom warranted. So these are opportunities to try out the tech and also ask questions of the developers that sometimes nobody else bothers to ask.
I think the most jarring element for most viewers is I'm giving an opinion on the tech, which is something I seldom do. Typically I leave you to decide your thoughts on the subject at hand, whatever it may be. But as I express to these companies looking to run this kind of sponsored content, they're not immune to criticism. If I have an issue with the tech I'll raise it and many of these companies have then back off given they realise I'm not just going to sing their praises.
This is great tech but the lack of emotion in the NPC voices leaves a lot to be desired. This is something OpenAI seems to have solved already, so I wonder if they've also solved it internally already.
Its pretty cool. What I would like to see in the future is if you could talk to the npc ai in any language and they would respond in the language you talk to them to. Like you could talk to them in french then in english then in spanish then in german... and the npc would comment on it and change language. Seems to me an AI would be able to do that easy. But yeah the voice delivery is still very early... Maybe im wrong but a way that could be improved... all the ``language models`` havent touched on the audio yet... seems to me you could do an AI that basically analyses like 5000 movies from hollywood or something and analyses all the conversations/dialogue and makes a huge databank, the way words are used together, with context and emotion... and uses that data to give us realistic voices. I think it could be done.... If someone would just take a ``language model`` type of AI but do the same with movies or audiobooks or podcasts or hell even youtube videos.
This seems pretty horrible on a few levels.
First, I am not a professional writer. I am a bit of a hobbyist writer, but I do not come to video games to do that hobby. Even if I was a professional quality writer who wanted to write the main character of a game, this kind of situation does not allow me to be creatively engaged with other writers as people in order to produce interesting things for the player character to say. Remember, most video game characters are not the player in a certain hat, they are characters in their own right. Geralt Riveria might fight how I tell him, go where I make him, and choose the dialog option I select, but the options to do that all inform who he is as a guy. Likewise most other player characters that have a name, and even some who do not (Mass Effect is a good example of this). As a player, I cannot know the whole story of what this character is like coming into the situation in order to write this character in a way that tells a story about them rather than a story about me.
Second, this kind of system is going to require some kind of dreadful monetization system. Randomly looking at a CNBC article on this estimates that running a prompt costs something like a half cent per. Currently, venture capital (VC) money is underwriting people getting to play with this stuff for free. This is like how Uber and door dash used to be really cheap and pay okay for its drivers. Literally like it because that too was being paid for by VC funding. When this dries up, there is going to be a wild need to squeeze for funds here.
Finally, this is going to make everything have a real "Bethesda game" vibe imo. The whole thing of LLM work is that they are great at feeling plausible. This is a problem, because it is really hard to tell when the AI is doing some kind of glitch. Is the AI just spitting out random nonsense, if a character is lying, or if they are misinformed. The latter two are cool, but the first one is the most basic text of the game beginning to suffer bugs. There is also no way to robustly test this kind of system. The use of AI in other parts of a computer system has had some work put into figuring out how to reduce the risks of hazardous malfunction, by putting guards around what the LLM can consider and what it can interact with as a result of this consideration, but when the thing to be considered is "a character who has had all of these prior conversations with a player character" that becomes a most likely unknowably complex and chaotic space.
I would really prefer if we were just comfortable having writers and artists make our games, rather than relying on systems that hoover up data and produce something that can plausibly pass for such work that is then trying to be wrangled by some new job of "prompt engineer".
I had this video running on my second monitor when i kept hearing someone saying "N-WORD!" and was like WTF is goin on? lol😂
Anyhow, that's pretty cool and i would be lying if i weren't excited for more studios to go down this path. My largest critique is the way they subtitle it, as it can be sort of confusing when they scroll up one row and pop the next row of text in below. Would be much preferable if they just swapped both rows of text at the same time.
I have a pretty thick Dutch accent and the dachshunds-like issues broke any enjoyment of this demo for me. Plus, I can't get over the lousy voice quality. It's on the level of Ren'Py self voicing.
I recommend the 2kliksphilip video on this demo too
I’m not particularly bullish on this. Firstly, you need an internet connection so outside of online games or live services this is simply a no go for me especially for single player games. More importantly when I looked, their pricing was per conversation or similar, so you’re telling me that if someone talks to the same NPC over and over it will cost the developers more? No go. But beyond this, I don’t want to talk to my computer or game. And their responses just feel far too artificial, I just don’t find that appealing.