Siri is too foolish and outdated! "AI Digital Proxy" has gone viral in Silicon Valley | AI | Proxy "in Silicon Valley

"AI agents will read content that you don't have time to read. This is very important because humans will never visit search websites again, nor will they go to Amazon again. Everything will be resolved through your agents," Microsoft co-founder Bill Gates said recently.

Gates also believes that once such AI digital proxies emerge, they may pose a potential threat to Google, Amazon, and other large companies.

About a decade after the emergence of artificial intelligence voice assistants such as Apple Siri and Amazon Alexa, AI big models such as ChatGPT have sparked a new wave of AI digital proxies with greater autonomy, attracting a lot of capital bets.

AI digital agents refer to artificial intelligence assistants that can perform more complex tasks under human guidance without the need for close supervision, sometimes also known as AI co drivers. Silicon Valley technology companies are competing to leverage advances in artificial intelligence technology to run AI digital proxy experimental systems on large models such as ChatGPT. Investors have injected billions of dollars of capital into this.

Humans may never need to go online again

Technology leaders, including Gates and Meta Chief Scientist Yang Likun, believe that in the future, everyone will have a personal AI assistant to provide comprehensive services from life to work.

It's like Jarvis in the movie Iron Man, helping Tony Stark carry out the mission he wants to carry out.

Gates recently predicted at a cutting-edge conference on artificial intelligence at Goldman Sachs that a company will eventually be able to create an AI digital agent that will understand everyone's activities and perform certain tasks for them.

Microsoft, the largest investor in OpenAI, is one of the giants targeting the field of AI digital proxies. Microsoft CEO Satya Nadella believes that the development of basic modeling technology is a leap in the evolution of AI assistants from Microsoft Cortana, Amazon Alexa, Apple Siri, and Google Assistant. In his view, these past digital assistants did not meet the initial expectations.

"They are all as foolish as stones, whether it's Cortana, Alexa, Google Assistant, or Siri, all of which are useless," Nadella said earlier this year.

An Amazon spokesperson stated that the team is developing a new model to make Alexa Assistant more powerful and useful.

The prototype of AI digital proxy has emerged

At present, technology giants and startups in Silicon Valley are competing to develop such AI digital proxies, although it is still far from imitating scenes from science fiction novels, there are already some embryonic forms of AI digital proxies.

AI digital proxy is essentially a computer program designed and programmed using AI technology, which can independently perform certain tasks and respond to the environment. AI agents can be seen as intelligent agents that can perceive the environment, change the environment through their own decisions and actions, and improve performance through learning and adaptation.

Generally speaking, AI digital agents have the ability to plan by gradually "thinking" and break down goals into smaller tasks, reflecting on their own performance. It typically includes various technologies such as machine learning, natural language processing, computer vision, planning, and reasoning, which enable agents to autonomously process information and make decisions.

Google's AI team DeepMind launched a reinforcement learning agent AlphaDev last month that can autonomously discover faster sorting algorithms. AlphaDev's advanced computer science algorithms surpass human benchmarks set by engineers and scientists for decades.

DeepMind has also released a self improving AI digital proxy, RoboCat, which is a software program empowered by AI, equivalent to the "brain" of a robot, and has a certain degree of universality, which can achieve self-improvement and self-improvement.

DeepMind researchers claim that RoboCat's learning speed is much faster than other state-of-the-art models currently available. This ability will help accelerate robot research as it reduces the need for human supervised training and is an important step in creating general-purpose robots.

The researchers first trained RoboCat by demonstrating how human controlled robotic arms perform tasks, such as picking up fruit chunks. Then, RoboCat can self train, perform an average of 10000 tasks without supervision, and continuously improve.

As early as last year, DeepMind had already published a research paper in the journal Nature Communications, stating that AI digital agents would negotiate and reach agreements. AI agents modeled through deep reinforcement learning methods could communicate and collaborate with other human agents, such as jointly developing plans while playing games.

OpenAI has also expressed its enthusiasm for AI agents multiple times. OpenAI co-founder Andre Capas recently stated in an offline developer event that if a paper proposes a different model training method, OpenAI will scoff internally, thinking that it is all leftover from their game; But if there are new papers about AI agents published, they will discuss them seriously and excitedly.

The first generation of products may come from startups

Although giants are flocking to develop AI digital proxies, the industry believes that startups may have initially provided AI digital proxies to consumers. Some AI startups are also conducting testing and development of AI agents.

For example, a company called InflectionAI, founded by technology industry leaders Reed Hoffman and Mustafa Suleiman, raised $1.3 billion at the end of last month. The company is developing a private AI assistant that is said to serve as a consultant or handle tasks such as earning flight points and booking hotels after travel delays.

Another AI startup, Adept, recently raised over $400 million in funding. The company's product ACT-1 is a large language model that can directly interact with browsers and commercial software, automatically completing complex operations. For example, in the customer relationship management software Salesforce, it automatically completes operations that originally required more than ten clicks and inputs at once after the user inputs instructions.

Qiu Kanjun, CEO of AI startup Generalintelligence, said that many things that are easy for people are still very difficult for computers. He believes that the real challenge is to build AI systems with strong reasoning capabilities.

The early demonstrations of AI systems that can perform relatively complex reasoning came from open-source projects such as BabyAGI and AutoGPT created earlier this year. This represents the "new frontier" of artificial intelligence technology, which can teach models like GPT to execute complex projects completely autonomously, such as determining sales prospects based on predefined goals and results, and setting priorities and execution status for tasks such as ordering pizza.

For ordinary people, there are many things AI digital agents can do in the future, including browsing websites and ordering burgers for their owners, developing investment and sales strategies, sending emails, and conducting meeting summaries.

The dawn of AGI is emerging, should we worry about risks?

Aravind Srinivas, CEO of artificial intelligence startup Perplexity AI, warns that models like AutoGPT may have many ways to go wrong, and he believes that human supervised "AI proxy" products should be provided. "You must treat artificial intelligence like a baby and constantly monitor it like a mother," said Srinivas.

Many computer scientists who are concerned about the ethics of artificial intelligence point out that the persistent existence of human bias and potential misinformation may bring recent harm. Although some people see the future of Jarvis, others are concerned about the brutal HAL9000 in 2001: A Space Odyssey.

Computer scientist Ben Gio, known as the "godfather of artificial intelligence," urged people to act with caution. He is concerned that future advanced iterations of the technology may create and achieve unexpected goals for himself.

An anonymous creator once posted an AI digital proxy called ChaosGPT online and set it as a "destructive, power consuming, and controllable artificial intelligence.". The creators also developed a five step plan for AI agents: from the first step of "destroying humanity" to the fifth step of "obtaining immortality.".

Tesla CEO Musk is also highly concerned about the ethical issues of artificial intelligence. Last week, Musk established his first artificial intelligence company, xAI, with the goal of creating an AI big model that is not very capable of eliminating humans.

The potential impact of artificial intelligence on consumers has attracted the attention of regulatory authorities in many countries. Last week, the Federal Trade Commission in the United States began investigating the potential impact of OpenAI on consumers. Although the FTC did not directly mention AI digital proxies, OpenAI has expressed caution in pushing its own open AI proxies to the market. OpenAI claims that it will conduct rigorous testing and establish extensive security protocols before releasing the new system.

Microsoft also stated that its vision is to achieve AI agents or co drivers that humans can control, rather than AI autonomous driving.