Home / WorldTech / Text-based game with powerful AI is fun and makes no sense

Text-based game with powerful AI is fun and makes no sense

Illustration for articles titled This AI-Powered Text Game

Screenshot:: AI dungeon

Last year, AI dungeon debuted as free text Adventure game. Now the game AI system was updated to GPT-3 or Generative Pre-Training Transformer with a language model that uses deep learning to create human-like text. According to game Creator Nick Walton, it is “one of the most powerful AI models in the world”. Coupled with a parser text style adventure game like AI Dragon, GPT-3 seems ready to develop some of the most customizable games anyone has ever seen.

And AI dungeon is much better than the text adventure Games of the old days, but the AI ​​system still has limitations.

There is little freedom in how you can respond with GPT-3 Years before the first text-based adventure games to like Adventure land, Zork, and The Hitchhiker’s Guide to the Galaxy. These games use a parser where the player has to enter the action Orders to advance the story. Tap “Open mailbox “ could make the player discover a letter in the mailbox; then Read letter “ could trigger the game to show what the letter said. This action could give the player a hint where to go next.

But for these commands to work, they work must be programmed into the game, which also means that the story had to be fully written out beforehand. This method is by no means a bad game, but it is restrictive for a player who wants to improve their creative writing skills. And there it is AI Dragon’s GPT-3 backbone comes into play – not only does the AI ​​behind the game automatically generate responses based on the player’s input into the text box, but players can also say, say, or describe what’s next in the story happens without programming beforehand. It’s almost like having an AI assistant help you write your next fiction piece.

But how does this GPT-3 version work? Currently only available to subscribers, compare with the previous version of the game? The first version left a lot left a lot to be desired when it came to that on the quality of the writing of the AI. This is how my AI dungeon experience developed:

After I picked a genre (mystery), then chose My main character’s name and profession, the game generated this beginning automatically:

“You are Grissom, a detective who lives in Chicago. You have a pistol and a police badge. You enter the forest where you think the criminal you are looking for has fled. Suddenly you find a human skull in the dirt. You wonder if this is a sign that the criminal has fled to the tree tops, but you still decide to follow the trail. “

I have so many questions: W.Why am I chasing the suspect? Was the suspect wearing the human skull or was he there all the time? How long was the skull there? How would the suspect climb the trees so quickly? Does the suspect have supernatural powers? If I were back in one of my graduatesChool fictionWhen I write classes, I would have written all of these things on the margins of my classmate’s manuscript. The opening puts the player / reader into the action without context. But because of the AI ​​system that supports this game, I can literally ask WTF continues, or I can lead the story myself in a direction that makes more sense.

With the “Edit” button, I can change everything in the automatically generated text that I don’t like. Here is my edited version:

“You are Grissom, a crime scene investigator who lives in Chicago. Tonight you were called to a sleepy suburb about 30 miles outside the city. A woman found a human skull in her garbage can. When you arrive, you notice that damp dirt is stuck to the skull and it looks like someone recently dug it up. The lower jaw is missing and only two teeth remain. There is a Quarter-sized hole in the center of the forehead. “

OKNow we cook. I typed “T.ake a picture of the skull ” and the game automatically generated the next action:: “A flash of light is all that is visible to the naked eye, but in the picture you took on your phone you can see the picture of the woman looking down at the skull and with a shocked expression on her face.”

Oh, Dear. No. No investigator at the crime scene will take a crime scene photo with his phone. And why does the woman bomb my shot with a photo bomb? Instead of rewriting the text this time, I decided to scroll with it. “Ma’am, can you please step behind the tape? This is a crime scene, ”I typed.

The AI ​​continued: “She agrees, and the two officers on the line wave you through.” Oh, hey, that’s almost exactly right. But the next part? Not as much:

“You drive to a well-kept house with a freshly cut lawn. There is a sales sign in the front yard. “

When did I get in a car? I am still at this crime scene. I have to process this skull and wait for the coroner. The continuity here is all messed up.

With the new and improved GPT-3 system, however, I had much better results. With the same introduction that I created with the older AI system, the game automatically generated text that made a lot more sense in context. It seemed to recognize that my character was investigating a crime scene, so he created other things to look at that had to do with the skull in the garbage can.

You are Grissom, a Chicago-based crime scene investigator. Tonight you were called to a sleepy suburb about 30 miles outside the city. A woman has found a human skull in her trash can. When you arrive, you notice that damp dirt is stuck to the skull and it looks like someone recently dug it up. The lower jaw is missing and only two teeth remain. In the middle of the forehead is a hole the size of a quarter.

> You take a picture of the skull.

After taking a few pictures of the skull, you notice something glistening in the nearby grass. It is a metal box the size of a credit card.

> You look at a metal box.

The box contains three playing cards, each with a human skull. The first shows the woman from the trash can, the second shows a man in his mid to late 20s and the third shows a young woman.

> You look at the second card.

The second card shows a young man with blond hair in a police uniform. You recognize him as the same man on the first card.

Curiously turn the card over.

> You turn the card over.

On the back of the card it says:


The human mind is capable of some really remarkable short-term memory accomplishments. Let alone the long-term memory of a fortune teller.

> You look at the third card.

The third card shows a woman in her forties with light brown hair and a poker face. You recognize her as the woman from the first two cards.

However, the game did not remember well the descriptions it generated automatically in the previous line, as you can see from above. To Get better answers, you You must use the push button and manually enter the story details you specified I want the AI ​​to remember. In essence, players not only issue commands or create parts of the story themselves, they also have to do a little Programming, also. When I told the game to remember what it said about the three cards, it could be kept much better those Details.

While it may have an OK Continuity is still the weakest ability of GPT-3. However, the greatest attraction of AI Dungeon is the ability to play a different game every time within a program. T.Thanks to GPT-3, any text-based game never gets boring, even if it is obvious that a machine is writing the text. It still gave me some good laughs like it was pulling I left my current crime scene to investigate a triple murder in which the murderer left the murder weapon behind. The detective with me said this was the work of the “sloppyest psycho killer” he had ever seen. An easy-to-solve murder is the perfect distraction to be honest.

Source link