The player who posted “nice job” in the game is actually a robot?

For a long time, we will encounter robot opponents (AI players) who are mentally handicapped or powerful in the game. Sometimes they have good marksmanship, and sometimes they move stupidly. But distinguishing them is actually very simple. No matter how we admire or spray them, they will not respond and hide their merit and fame. However, this situation may change. Soon, high-level robot players who kill indiscriminately in the game are likely to play "nicely" while in the game to engage in teammate mentality and increase (not) the game experience.

▲ The AI ​​of the King of Glory challenges Jue Enlightenment. Picture from: inkl

The “Think, fight, feel: how video game artificial intelligence is evolving” in the “Guardian” indicates that in the future, we may not be able to distinguish whether the players in the game are real people or robots. They may encourage you when you click to surrender. You: Hold on, we can come back.

But this is far from enough.

In May of this year, Sony CEO Kenichiro Yoshida issued an interesting statement: Their artificial intelligence research department Sony AI will work with PlayStation developers to create robotic AI players. This is nothing new. The key is that he mentioned that “by using reinforcement learning, we are developing game AI that can become players’ opponents or companions.” Reinforcement learning is a field of machine learning in which artificial intelligence can effectively teach itself through trial and error. How to give feedback.

▲ Picture from: Karina Nooka

In short, these robot players will imitate human players. To some extent, they (they) think.

This is just the latest example of the continuous iteration and development of artificial intelligence in the development of video games. As open world games have become more complex and grandiose, there are now hundreds of characters and multiple intertwined stories. Developers must build systems that can produce intelligent, reactive, and creative roles to respond to emergencies.

The game developer Monolith created the acclaimed Nemesis AI system, which allows enemies in the game to remember their battles with the player, and then creates blood and hatred throughout the adventure. That is to say, the game characters are no longer forgetful, but will Hold grudges. The most recent game "Watch Dogs" generates life stories, interpersonal relationships and daily life for players and every London citizen who interacts with them. Therefore, if a player saves a character's life one day, then their best companion will most likely join the player's camp the next day. Yes, the game character will not only hold grudges, but also repay gratitude.

▲ "Watch Dogs". Picture from: epicgames

How to automatically and randomly generate interactions and stories between game characters and players is actually not a difficult task. The previous text adventure game AI Dungeon used OpenAI's natural language modeler GPT-3 to create a new narrative experience. In other words, computers can now write stories by themselves.

However, there is a diversity problem in the field of artificial intelligence, mainly because the proportion of women and people of color is not high. A study published by New York University in 2019 found that 80% of artificial intelligence professors who spoke at major events were men, only 15% of artificial intelligence researchers at Facebook were women, and only 10% at Google. The statistics for people of color in the technology industry are even worse: only 2.5% of Google’s employees are black; only 4% of Facebook’s employees are black. The risk of this homogeneous work culture is that gender and race biases may uncontrollably affect AI algorithms, and the resulting results will replicate deep-rooted imbalances and biases.

In the past five years, there have been many examples of this, for example, some facial recognition systems discriminate against people of color; or artificial intelligence recruitment tools that prefer male applicants.

In fact, video game development also has a similar homogeneity problem. If we see artificial intelligence generating diverse characters and stories in the future, should developers also consider diversifying the teams behind them? Will the stories and plots written by the computer always be between two American men? Sorry, we prefer to watch "American Beauty" rather than "American Man".

Although the status quo is not optimistic, the game industry has realized this problem and started to solve it.

Uma Jayaram is the general manager of SEED, an innovation and applied research team at Electronic Arts. As a female executive, she naturally supports diversity. In addition, as an influential technology entrepreneur, she has worked in the fields of cloud computing, virtual reality, large-scale data, and artificial intelligence. Her identity, influence and job attributes make her seek to form her own global team. In the end, her team members are located in Sweden, the United Kingdom, Canada and the United States, and have different genders, races and cultures. she says:

A diversified team is able to integrate multiple concepts, and then more benchmark products (games) are more likely. This kind of benchmarking game can also allow us to have a broad vision, know others, and strengthen our empathy. Video games are to some extent an extension of our physical world, a place where people spend their time and experience happiness. This experience can be traced back to the collective consciousness of self and community. Therefore, in the face of this excellent opportunity, we have two ways to introduce diversity: to allow the teams that design and build these worlds to achieve diversity, and to allow the worlds that are being created and the residents who live in them to achieve diversity.

▲Uma Jayaram. Picture from: businesswire

Electronic Arts is currently researching and developing a system that can replicate facial expressions, skin types, and body movements from videos and photos through machine learning, eliminating the need to bring actors into motion capture studios.

In theory, this should expand the range of genders and races that can be produced in the game. Jayaram stated that Electronic Arts is committed to using diverse data in its research and development projects. The company is also considering using user-generated content in games and allowing players to capture their portraits and expressions on their smartphones or webcams and upload them to the game to create unique avatars.

We are not unfamiliar with this kind of small function. The key is whether we are willing to upload our facial data.

▲American Electronic Arts (EA). Picture from: businesswire

The importance of diversified data is emphasized because it is actually correcting a misunderstanding of artificial intelligence: it is objective to some extent because it is the result of calculation.

In fact, the AI ​​algorithm relies on data. If the data comes from a single demographic data, it will reflect the bias and blind spots of that group.

"We are used to thinking of artificial intelligence as a physics engine or multiplayer game code, sometimes as something that happens behind the scenes," said Michael Cook, an artificial intelligence researcher and game developer. "But today's artificial intelligence is part of creative work. It controls the behavior and relationship of AI people in "The Sims"; it generates culture and religion in games like "Cad Cave" and "Ultima Ratio Regum". It will also be part of the political statement in the "Watch Dogs" game. The responsibility of artificial intelligence engineers to players is as great as that of screenwriters and designers. They can create a happy experience, and they can also cause the tears of men and women, even if they do not No plots are written. For example, the recent text adventure game AI Dungeon has produced stories that may cause harm to the player without any warning.

▲Game "Ultima Ratio Regum". Picture from: Youtube@Blind

At Microsoft, the company’s Cambridge artificial intelligence research team is conducting several studies on machine learning and games, including the Paidia project, which is studying the application of reinforcement learning in game artificial intelligence agents. For example, AI agents can interact with Human players collaborate.

Microsoft also sees the potential of player modeling-artificial intelligence systems learn how to act and react by observing the behavior of human players in the game world. As long as you have a broad player base, this can be a way to increase the diversity of data input into the AI ​​learning system. Katja Hofmann, Microsoft's principal researcher in Cambridge, said:

The next step will be training to provide more diverse or humane roles. The scenario where robot players learn from human players is one of the most challenging but also one of the most exciting directions.

▲ Katja Hofmann. Picture from: slideslive

Similar to the opinions of the previous game industry insiders, Katja Hofmann also emphasized: artificial intelligence technology will not automatically bring a diversified gaming experience. Developers and creators need to choose how to use artificial intelligence technology. These choices determine whether the resulting roles and experiences reflect diversity.

Whether it is a robot opponent in StarCraft, or a Go AI such as AlphaGo or Tencent Juewu, the game industry tends to design artificial intelligence as a confrontation system. The role of artificial intelligence is to create stronger opponents or teammates. But if we look farther, we do see other options. Coder and New York University professor Mitu Khandaker and technical expert Latoya Peterson established her Glow Up Games studio to create social narrative games for different audiences. The team is currently developing Insecure: The Come Up Game, a mobile game based on popular HBO episodes that allows players to explore the relationship between characters.

▲ Picture from: twitter@insecuregame

The development studio Tru Luv previously created the popular Self-Care application, and now they are developing artificial intelligence technology that reflects the company's own diverse, progressive, and supportive studio culture. The founder of this studio, Brie Code, said his company currently has one-third of blacks, indigenous and colored people, two-thirds of women, 100% of the decision-making team, and one-third of the BIPOC (black, Indigenous peoples and people of color) and two-thirds of women, they work with consultants and partners from emerging development communities such as Pakistan, Tunisia and Morocco.

Like Khandaker, Code believes that a diversified workforce can not only eliminate the prejudices that exist in traditional games, but also develop new interactive experiences. "For many years, the gaming industry has focused on a small part of human psychology," she said. "It is very good at creating experiences that help people feel a sense of accomplishment or dominance. Game AI created by a diversified workforce will bring life to NPCs and experiences that represent the breadth and depth of human experience. We will see more nonsense. Zero-sum experience, more compassion, more emotional resonance, more insight, more transcendence. We will see a brand new game format, which will more mobilize the player’s sense of creativity, love and happiness , Not just a sense of victory and dominance.”

▲ Picture from: syncedreview

Now that game consoles and high-end PCs have more powerful computing capabilities, the complexity of artificial intelligence systems will increase accordingly. Developers will explore elements such as natural language processing, player modeling, and machine learning to develop imaginative and reactive artificial intelligence characters. The key issue is the issue of diversity mentioned above. Mike Cook said that because of the problem of the composition of people in the AI ​​research field, we missed a lot of ideas and opinions.

Diversity is not just about avoiding mistakes or harm-it's about new ideas, different ways of thinking, and new voices. The diversification of game AI means that smart people can put their ideas into practice, which means that AI will not only appear in games in new ways. It may also mean that new game types are coming.

▲ Diversified games represent "The Sims". Picture from: origin

So, going back to the title, it’s not really important that AI communicates with players like real people in the game. What’s important is what kind of identity AI communicates with players…

#Welcome to follow Aifaner's official WeChat account: Aifaner (WeChat ID: ifanr), more exciting content will be provided to you as soon as possible.

Ai Faner | Original link · View comments · Sina Weibo