AI and how they can be used in game development.

ArilethDeTyr

AI and how they can be used in game development.

• Create novels, create texts (new book from Game of Throns))
o You can use it to create little stories about objects like swords or small towns/villages.
 https://motherboard.vice.com/en_us/article/evvq3n/game-of-thrones-winds-of-winter-neural-network

• Translate text so that it doesn't look machine (DeepL as if the translated text was written by a human being!)
o You can localize every game and make multi-languages with it.
 https://www.deepl.com/translator
• "The first translator with the wow effect. The start-up DeepL brings an interpreter who "clearly beats Google". Tages-Anzeiger, Switzerland

• create new artist pictures, newer art style
o It may be possible to create new paintings or pictures on packaging or in games, perhaps even textures, without expensive graphic designers. It would at least be a help.
o The AI then generates hundreds, you only need to select one.
 https://www.spektrum.de/news/kreativitaet-aus-der-maschine/1557286
• Even music and songs as well as pictures in a new silent

• Photorealistic portrait photos of people who do not exist
o With it you can create the portrait images of NPC and are always unique and look photorealistic and that also in thousands of images.
o By proper training the pictures can also look like comic figures
 https://www.golem.de/news/nvidia-neuronales-netzwerk-generiert-fotorealistische-bilder-1711-130913.html

• Photorealistic images of animals and other objects
o The same as with the photorealistic portrait photos of people
 https://jaai.de/kuenstliche-intelligenz-generiert-hochaufloesende-bilder-von-menschen-die-nicht-existieren-1311/

• Natural, real-sounding voice output from people who don't exist.
o You can set everything to music in the right language without any real people, which is especially good in the RPG, because there is a lot of text in these games.
 https://www.golem.de/news/tacotron-2-googles-sprachsynthese-erreicht-fast-menschliche-qualitaet-1712-131817.html
 https://google.github.io/tacotron/publications/tacotron2/
• It's really amazing!

• Recognition of text contents, orders (email) recognize from the text and text context.
o I'm sure you can include that in games somehow, that you can write and the game understands the content in the game or the NPC.
 https://azure.microsoft.com/de-de/services/cognitive-services/text-analytics/

• Recognition of image content, image recognition
o https://azure.microsoft.com/de-de/services/cognitive-services/computer-vision/

• face recognition (?)

• Autonomous driving (?)
o Not needed in game development, or perhaps in games, driving virtual cars?

maouw

I'm interested to see if AoC will use Deep Learning technologies too.


If they do, I dunno if they'd tell us tho hahaha

perseus01

Just count to 5 and move on....(kidding)
Great ideas. Not sure how it would ever play out in this new style of marketing, however the future does hold its possibilities and savoring nectar.

polarbear

I've worked on a few Neral Networks and the problem is that they are currently super finicky and supper expensive on computing. I don't think we will see much of that stuff running client-side any time soon though with Nvidias new A100 cards they might be able to do some of those things on the back end. One use of Dnn's I think would work is making quest prompts and then using a model like GPT in the back end to regenerate the text for the quests every day or two.

maouw

It has great potential for Boss AI - because DotA2 pros said it felt like they were playing against something scary that would punish their smallest miss-step.

That level of difficulty is too high for the average player, but could be an option for the top 1%. Most people only managed to beat the AI by finding cheese methods that the AI didn't understand.

Huge problem with this idea: where do you get training data from?

Noaani

polarbear wrote: »

I've worked on a few Neral Networks and the problem is that they are currently super finicky and supper expensive on computing. I don't think we will see much of that stuff running client-side any time soon though with Nvidias new A100 cards they might be able to do some of those things on the back end. One use of Dnn's I think would work is making quest prompts and then using a model like GPT in the back end to regenerate the text for the quests every day or two.

This is basically what I was going g to say.

Some of the above may be viable, but most of it is 10+ years away from being something companies would consider using in an actual product. If all the information on something is in news articles or research papers, it isn't something that can be used commercially.

Yet.

polarbear

@Noaani I totally agree with your point, the development cycle is too unpredictable for a profit-seeking company to attempt implementing a DNN. Realistically, they could do it... but it would simply take too long and cost too much money to be worth it.

Though if you're looking for a game based on a neural network, check out AI Dungeon 2. It's a DnD inspired text game built on the GPT-2 language model.