From JARVIS to Westworld: The future of intelligent agents and human symbiosis

In 1927, “Metropolis” filmed by German director Fritz Lang premiered in Berlin. This is the first film involving artificial intelligence in human history. A humanoid robot named Maria set off a storm in the underground world.

Since then, various types of artificial intelligence agents that are set to possess high intelligence have filled various film and television works. In “Star Wars”, 3-CPO undertook the interstellar translation work, JARVIS helped Iron Man handle personal and company affairs, TARS in “Interstellar” saved the protagonist more than once, and in “Western World” Dolores finally awakened and issued Roar, and decades after “Blade Runner” was released, some people are still debating whether the main character is human or machine. These artificial intelligence characters serve humans, accompany humans, and even eventually become humans. Under the influence of these works, people gradually believe that artificial intelligence will accompany everyone in the future, and everything is just a matter of time.

The curtain is opening

Whether it is a robot with a physical form or an AI program working in the digital world, it can be called an Intelligent Agent. The classic textbook “Artificial Intelligence: Modern Approaches” once defined artificial intelligence research as “study and design of intelligent agents”, that is, the research purpose of the discipline of artificial intelligence is to achieve better intelligent agents.

Intelligent Agent has actually been with us for a long time. When opening an e-mail box, an Agent is silently classifying emails and filtering spam. When typing keywords in the search box, another Agent is providing recommendations and search results. In train stations and shopping malls, Agents work silently behind surveillance cameras, using AI technology to protect public safety. Siri in mobile phones can understand and respond to people’s commands and carry on conversations. Tesla’s assisted driving system can offload some of the driver’s work. However, people have no special perception of these intelligent agents, because they often work behind the scenes and are not very intelligent, which is not the same as what they see in movies.

The release of ChatGPT marks a breakthrough in large language model technology. And language is not just a tool for communication, it is also the key for human beings to understand the world and think deeply. When AI masters language, it actually also masters insight into the world and the ability to solve problems. People are beginning to realize that the big language model is not just a dialogue partner who provides suggestions, but can also directly participate in the work to solve problems and complete tasks. All of a sudden, a lot of talents and resources have been invested in this direction, and the curtain of a new era is being slowly opened.

Agent reconstructs the economic system

Autonomous Agent – Autonomous Agent, a popular fried chicken in the AI world.

The dictionary defines “Autonomous” as "Carried on without outside control” - work without external control. AI technology has been developed for so many years, but we have not seen a few intelligent agents with real autonomy. I bought a first-generation iRobot in 2005 to sweep the floor The robot, which claims to be fully autonomous, can detect terrain, avoid obstacles, and automatically recharge. You don’t have to worry about turning it on. As a result, I crashed it the first time I used it. The suction port was blocked within minutes. This is obviously far from “independence”.

The above video shows the working process of Agent. The big brother said to help me book a ticket from New York to San Francisco on June 10, his personal assistant Agent immediately started to work - it opened the browser, visited google flight, screened out the direct flight of United Airlines, and The best airfares for the right time period are selected. Then, it completed the seat selection and paid successfully, and the task was easily completed. This is a product developed by a start-up company whose vision is to create an all-round AI assistant like JARVIS in Iron Man. This process properly reveals the working mode of the autonomous agent: understand the task, formulate and execute the strategy, analyze the result, feedback loop, until the task is achieved.

People who got test accounts quickly discovered various other features. It’s easy to order a pizza and salad, and someone said that I’m going to make lasagna tonight, and it’s easy to order all the ingredients I need from Walmart. There are also those who use it to automatically send tweets, arrange meetings, automatically fill in forms, automatically detect facebook every day to send blessings to friends who have birthdays that day, and even use it to book wedding venues and plan and arrange wedding procedures.

Dealing with these tasks is not very difficult to say, even the old version of the AI assistant can still do it if it wants to, but it is very troublesome. Programmers need to design code individually according to each scenario and integrate related services. The implementation of this Agent is not only inefficient, but also has very low task capability. If an Agent integrated with Meituan’s meal ordering service suddenly fails to connect to “Meituan”, it doesn’t know that it can also complete the order by going to “Eleme”. What" to call the interface.

The autonomous agent with the large language model as the core can adapt to various tasks after combining the general working framework and the preset instruction set. This allows it to easily complete operations such as booking tickets and seat selection without special training. Not only that, whether it is planning a trip, organizing mail, or tracking items on eBay in real time and haggling with sellers, it can do the job. Unlike traditional AI assistants, which can only provide information and suggestions, autonomous agents place more emphasis on actual execution capabilities. **It won’t be long before most of people’s activities on the Internet can be taken over by Agents. **

The capabilities of the most advanced large language models actually go far beyond the scope of daily assistance work. More Agent developers are targeting more professional fields: market research, sales assistance, product development, and even scientific research. A billion people around the world spend most of their working hours doing repetitive mental tasks, filling out tax forms, sorting data, looking for potential clients, and writing emails over and over again. And repetitive mental work in this kind of work will also be the battlefield that Agent will soon overcome, they will accurately and efficiently handle repetitive and mundane tasks, liberating a lot of manpower. The user only needs to tell the agent what to do, and it won’t take long for the agent to give feedback “Boss, it’s done.”

11x.AI A start-up company that provides AI employees

Of course, Agent can’t be proficient in everything. **With the evolution of the Agent economy, I think there will be multiple highly diversified Agent markets with clear division of labor. For complex tasks, a comprehensive Agent will take the lead, analyze the goals, form a task chain, and send tasks that cannot be completed efficiently to Agents in various vertical fields to jointly complete the goals. **And no matter how developed the agent is, many tasks will still have to be done by people for a long period of time. When such a situation is encountered, it is not uncommon for AI to hire humans in turn.

Social My AI brother

After all, an intelligent agent is just a piece of code running in a computer. Even if it has a perfect interaction capability in the digital world (in fact, it is not yet achieved), the interaction it can achieve is limited to the public Internet, and everyone is not satisfied with this. . How to enable Agent to realize the ability to interact and perform transactions in a wider social layer.

One team gave their own answer - Legal Packaging (Legal Wrapper). **If the Agent can be associated with a legal entity and authorized reasonably, the Agent will be able to achieve a higher level of autonomy, and its ability to handle affairs will also increase significantly. ** With a legal entity, of course, you must have your own finances. It is a matter of course to set up a bank account for the Agent, with a certain amount of funds, so that it can be called. As a result, the intelligent Agent has a wider range of behavioral capabilities at the social level, and this approach can also enable Agent users to receive effective legal protection. The principle of this set of gameplay is not complicated, but in practice, there are many difficulties both technically and legally, and it will also touch some areas where supervision is not yet clear and potential ethical issues.

However, in my opinion, it is only a transitional solution for Agent to be legally packaged and equipped with a traditional bank account. The systems we now rely on were designed for human use, and for Agents they are inefficient and fraught with obstacles. When the world develops to the stage where Agents run everywhere in a few years, massive demands and applications will prompt the development of collaboration systems, financial systems, and even currencies suitable for intelligent Agents. These systems are parallel to the existing systems, The intercommunication between the two systems is realized through thousands of connectors. In the end, we will probably find that it is unreasonable for the Agent to communicate in human language for a long time, and it is very likely that an AI language will gradually evolve.

In fact, there have been related researches for a long time, and Facebook found that AI has developed a non-human communication method in the early AI robot negotiation experiment.

The Sims

Among all intelligent agent practices, anthropomorphic agent simulation has attracted the most attention. In March of this year, researchers from Google and Stanford University conducted an interesting experiment. They created a virtual town called Smallville, where 25 intelligent agents driven by large language models lived.

Each villain has its own settings, such as:

*"The kind, patient Mei Lin is a university professor and mother with a passion for helping people achieve their goals. She is always looking for ways to support her students and family. Mei Lin with her husband John Lin and son Eddy Living with Lin, she is teaching a philosophy class and writing a research paper.*She goes to bed around 11pm, wakes up around 7am, and has dinner around 5pm.”

With such a simple setting, a small society composed of AI is functioning.

John Lin wakes up at 6 am, brushes his teeth, takes a shower, gets dressed, eats breakfast, and then checks his email. His wife Mei Ling got up at 7 o’clock, and his son Eddy got up at 8 o’clock. After washing his face and brushing his teeth, he talked with his mother about class creation and other things.

Even complex social behaviors are achieved when more agents interact. Someone asked for a Valentine’s Day party, and in a short amount of time, the invitation spread to other people in the town, and eventually 5 people chose to attend and arrived at the party. None of this is pre-programmed, in other words, these agents are indeed living their “life” in the small town.

Inspired in part by the town of Stanford, a San Francisco startup used a similar concept and some specific training to simulate a town in South Park, America’s most famous animated series. After integrating text-to-speech technology, an episode of South Park filmed by AI was born. In just a few days the episode was viewed more than 7 million times on Twitter. The Forbes report even had an exaggerated headline like “AI Producer Becomes the Sum of Hollywood’s Fears”.

The founder of this project is a friend of mine who has been exploring in the field of simulation since we met. His exploration process is very informative, initially he produced a VR interactive film, for which he won an Emmy Award in 2019. In this film, the audience plays the virtual friend of the little girl Lucy. When he realized that people want to be real friends with virtual people instead of watching each other’s performances, he chose to use the image of Lucy to make an AI virtual human, a person who pretends to have “self” and “life”, An AI agent that can have real-time conversations with everyone through zoom video conferencing.

Lucy interacts with everyone in real time at the 2021 Sundance Film Festival. This scene is not difficult to achieve now, but it still shocked many people two and a half years ago.

He quickly discovered that to make intelligent agents more human-like, it is not enough to pretend a single simulation, but to allow them to have friends, socialize, and have their own lives. **So his direction turned to AI to create a virtual world in which they “live”. And I hope that one day in the future, people will also be able to enter these worlds to interact with AI and live together. The episodes filmed by AI are just incidental results, because “life” is like a play.

While such agents are often used for emotional companionship and entertainment, the applications of agent simulation go far beyond this. Researchers from Harvard and Microsoft have published papers on how to use AI to simulate consumers for market research, revealing its huge potential in the consumer field.

The Massachusetts Institute of Technology also uses AI to simulate human behavior, such as observing the decisions of AI bosses under different salary and experience conditions, or letting AI decide the allocation of federal budgets between highway safety and car safety. These are classic experiments in economics. When AI is put into these scenarios, the decisions made by AI are highly similar to the results of experiments done by humans in the past, which means that this type of simulation has great practical value. A friend half-jokingly said that the president of the United States after two terms might be an AI, which is not unreasonable.

The author uses AI to simulate humans to reproduce the snow shovel price experiment proposed by Kahneman in 1986

There is a bolder possibility for simulation in human society. OpenAI’s GPT1 has 117 million training parameters, which have evolved to GPT3’s 175 billion in just a few years. During this process, the phenomenon of “emergence” appeared, and the intelligence of the model suddenly increased significantly. If we regard the 25 people in the small town of Stanford as the first generation, with the investment of more research power and computing resources, the number of agents in a single simulated society may soon become thousands, tens of thousands or even millions . What would these simulated societies develop into in operation? Will things emerge in this society that have never occurred in human society. And the intelligence living here will not one day also emerge and develop more human-like traits or even a sense of “self”.

Consciousness in Artificial Intelligence - Lessons from the Science of Consciousness

Turing Award winner Yoshua Bengio and several experts released a paper called “Consciousness in Artificial Intelligence” last week. In their paper they provide a rigorous, empirically-based approach to assessing the presence or absence of consciousness in artificial intelligence systems. Although the evaluation shows that none of the current artificial intelligence systems are conscious, it also gives a bold conclusion-There is no obvious obstacle to building a conscious artificial intelligence system.

And just halfway through this article, OpenAI announced the acquisition of a Global Illumination, a company with only 8 people who has only one product, a sandbox-like simulation world game. The announcement is just a few lines, without explaining the purpose of the acquisition and the details of the transaction, but there is no sound but it explains some problems-the real reason may have many controversies. I think OpenAI is definitely not for making a more fun game.

AI version of “Western Development”

The vision of humans and intelligent agents living together in the “Western World” is very beautiful, but the “Western World” is still barren, waiting for the arrival of a great western development. In my opinion, there are three main lines in the big development : Credible, Actionable, Sustainable.

Credible: Is the capability of the model strong enough to be credible? Is the intent of the model credible (AI-human Aligement)? Is Agent itself (service provider) trustworthy? How to ensure privacy in Agent-Agent interaction? How to ensure mutual trust? How does the agent interact with the other party in the real world while gaining mutual trust and so on.

Actionable: the technological level of action in the digital world, the social level of action in the digital world, the action in the human social system, the ability to act in the physical world through a third party, and the ability to act in the physical world (embodied intelligence).

Sustainable: Sustainability of operating environment, controllability of computing resources, self-healing, self-energy management, etc. When AI capabilities gradually become an indispensable basic resource such as electric power and oil, whether it is an individual, an institution, a country, or even AI itself, it will put sustainability in a very important position. A truly autonomous Agent will guarantee its own survival to a certain extent in the future.

In order to build these infrastructures, we must not only rely on the advancement and integration of technologies such as artificial intelligence, cryptography, blockchain, and communication, but also rely on the development and integration of different social sciences such as economics, game theory, sociology, anthropology, law, and politics. Explore within the discipline. Many entrepreneurs and scholars have invested in these fields. For example, David Luan, a Chinese entrepreneur who once worked for OpenAI, founded Adept AI. They are building a set of interaction models to allow AI to complete all tasks that originally required human operations on the computer. interact.

I am not an expert in AI technology, my understanding of many things is limited, and I may underestimate the difficulty of some implementations. But I don’t think I’m overly optimistic. On the contrary, I think the future will definitely come in a wilder way. Every technological revolution in history will easily break through the limits of people’s imagination at that time. We have exhausted all our imaginations today, maybe like the mountain people who have lived in the mountains for a long time, the most daring vision of life is just to be able to eat dumplings every meal.

As I write this article, the genius boy Zhihui Jun has released a humanoid robot called Zhiyuan Robot. This type of intelligent agent with physical entities is called embodied intelligence. By interacting with the physical world, they may have a more direct impact on human society. The era of agents seems to be more within reach.

The film “Metropolis” shot in 1926 imagined a world a hundred years later. It is not only the first film involving artificial intelligence in history, but also the first dystopian film. Its profound ideological core and visual spectacle have influenced subsequent generations of sci-fi works, which in turn have influenced the world’s views on technology and the future.

The director who lived a hundred years ago failed to foresee the information age. The year 2026 he described in the film is an era of highly developed industries. Huge machines support the operation of the entire city, and a large number of workers do mechanical work in underground factories. The role of “hands” in society. Today we see programmers mechanically typing code and sales representatives repeatedly selling products. This is actually no different from the workers in “Metropolis”. It serves as a bridge between the “brain” and “hand” in the society, and acts as the “heart” of the society, allowing different classes to understand and coordinate with each other.

A hundred years later, facing a future that is close at hand, how will Agent change my daily work and life patterns? Will my existing skills become obsolete with the Agent? How will the Agent affect my financial situation? What opportunities will this new era bring me?

And at the social level, how to deal with potential unemployment? How to more equitably distribute the additional wealth brought about by the increase in productivity? How do people find meaning in life when work is no longer necessary? How AI and human values can be continuously harmonized. How to shape an optimistic future rather than a pessimistic one?

I don’t have an answer, no one has an answer.

But everyone will be wrapped in the torrent of this era, write together, and run towards the answer.

The author is currently focusing on learning, research, incubation and related investments in areas such as community-driven/cultural-driven, AI + encryption, and Agent ecology. The MULTI.ON/Fable and undisclosed projects mentioned in the article were all voted by the author. The purpose of this article is to share information and does not constitute investment advice.