【Board Game Miscellaneous】Can the DM in the game be replaced by machine or AI intelligence?

2020-02-05

That's a good question! The answer is possible. In fact, the technology to make a DND game host is complete. For the sake of discussion, we can call it an AI-driven game. Of course, any idea that looks good must have numerous pits behind it. To replace the host of a game with a machine or artificial intelligence requires a huge investment, a long research process, and a lot of work from the game industry. In other words, money is needed, a lot of money.

For gaming companies, even if they are very concerned about AI technology, the investment in this area is huge, and the R & D cycle is very long. It may take more than ten years to make playable games, and the market prospect is not obvious. Therefore, in the foreseeable future, it will be difficult for us to see a truly playable AI host. However, I find it interesting and interesting to discuss how to make a playable AI host. Now let's think about what kind of technology we need to achieve this level, what kind of technology is already available.

At present, there are two core tasks for game hosts.

-Design a plot text

-When the player starts the action, throw the dice to determine the result of the action, and then tell the player what will happen.

Currently, designing a plot text can take the GPT-2 model.

With 1.5 billion parameters, GPT-2 is currently the most powerful text generation system. It can find the embedding vector corresponding to a word from the embedding matrix. Before processing each word (passing it into the neural network), it incorporates the model's understanding of the related words used to explain the context of a word.

For this AI text adventure game, players can choose the theme freely after entering the game. When the player enters a plot, it will feedback a plot in time.

Since a sufficient amount of text was entered, a self-attention model was used, and if you tell the NPC in AI Dungeon, they are fictional. AI Dungeon will even tell you that NPCs have a frustrated expression.

However, AI Dungeon is an unqualified design and it has several problems.

-Lack of proper plot structure

-There is no critical game experience of running a team: throwing dice to determine.

-There is no numerical system.

——No art design.

-Once the number of characters increases, the computer is prone to chaos.

Then, the player will easily lose the interest of exploration, because no matter what he does, the player can succeed, and he cannot find the cause of the failure, and there is no numerical upgrade experience.

A suitable plot structure includes: cause, action, conflict, development, climax, ending, and cause of the next story. Therefore, an AI capable of acting as a host must first create appropriate motivations for the player's actions, and then when the player starts the action, it must set enough obstacles for the player's actions, and such obstacles should follow the plot The development of the game is gradually strengthened and has a turning point. In the end, let the player complete the goal, resolve the contradiction, and set the cause of the next story.

For example: In AI Dungeon, killing a dragon is the ultimate goal, it only makes the whole plot line tend to kill a dragon. However, for a running game, it should actually provide various secondary goals in addition to the primary goal. For example, what it takes to kill a dragon. It should also have the ability to determine goal completion and come up with new keywords to continue the game. When the dragon is half-blooded, a new twist must emerge. This may be a madness. This dragon has become more angry and more powerful. When the player is beaten and urinating, and inquiring about the information, they may find that to kill a mad dragon, A sedative potion for a mad scientist who needs the brain of a zombie as a raw material.
After the player killed the dragon, they received the reward and won the victory. At this time, the AI must understand that the goal is over. It must quickly point out that in the dragon's treasure, you found a treasure map. Or leave the player in a keyword-triggered state. Once the player decides to continue an action, the AI can come up with a whole new series of stories.

The design of dice judgment is the top priority. For a running game, the conflict setting depends on the judgment of the dice. A complete dice decision can be divided into at least five steps. 
1. The player says an action.
2. The AI must calculate the resources consumed by the player based on this action to determine whether the action can be performed.
3. Evaluate the difficulty of the action.
4. Throw the dice, plus the character's correction result based on the type of action.
5. Give different plot expressions according to the dice results.

All the technologies to replace AI with human moderators are complete. Unfortunately, huge investment is still needed to complete such a powerful function. This process requires continuous function iteration, and various methods are used to bypass the difficulties . The host of the game can theoretically be replaced by AI. But now is the world of cyberpunk, low-life high-tech is the theme. Technology can do it, it does not mean that humans can do it, because human beings always need to measure the ratio of input and output. AI is not a machine that relies on preset scripts and random number generators. It has its own logic and ideas, but the limitations of the current technology make it more like an animal than a living human. Even so, we are standing at a turning point in history, and intelligent technologies represented by artificial intelligence are increasingly entering our lives. We do not know how many years this process will take, nor do we know where it will go in the future.