Scientists Created a Company Employing Only AI Employees: What Came of It - ForumDaily
The article has been automatically translated into English by Google Translate from Russian and has not been edited.
Переклад цього матеріалу українською мовою з російської було автоматично здійснено сервісом Google Translate, без подальшого редагування тексту.
Bu məqalə Google Translate servisi vasitəsi ilə avtomatik olaraq rus dilindən azərbaycan dilinə tərcümə olunmuşdur. Bundan sonra mətn redaktə edilməmişdir.

Scientists Created a Company Employing Only AI Employees: What Came of It

If you're worried that artificial intelligence is about to take your job and leave you without a livelihood, you can breathe a sigh of relief. AI is not going to take your job anytime soon. Not because it doesn't want to, but because it simply isn't capable of it, the publication writes. Futurism.

Photo: Inkdropcreative1
| Dreamstime.com

A recent experiment conducted by researchers at Carnegie Mellon University yielded some very interesting and encouraging results for humans. The scientists created a fake software company, fully staffed with AI agents — artificial intelligence models. They were supposed to perform tasks independently.

The simulation, called TheAgentCompany, was staffed with digital workers from Google, OpenAI, Anthropic, and Meta. In a virtual office, they played the roles of financial analysts, software engineers, and project managers, working alongside their digital counterparts — the HR department and the CTO.

On the subject: Elon Musk said that artificial intelligence will take away all jobs from people

To test how the models would cope with tasks in conditions close to real ones, the researchers set the AI ​​tasks typical of a regular IT company. AI agents tried to navigate file directories, conduct virtual tours of new offices, and even write reports on the productivity of programmers based on feedback.

The results were dismal. The best performer was Anthropic's Claude 3.5 Sonnet, which was able to complete only 24% of the tasks assigned to it. The study's authors note that even this modest efficiency was very expensive — an average of almost 30 steps and a cost of over $6 per task.

(As the AI, specifically Chat GPT, explained to us, steps in this context refer to individual actions or commands that the AI ​​agent must perform to solve a single task.

Each step may include, for example:

  • accessing a database or file,
  • requesting information from a "virtual colleague"
  • executing a command to navigate the file system,
  • generating text or code,
  • making an interim decision, etc.

That is, a task that a person could solve in a few logical steps requires dozens of iterations from AI - due to the lack of common sense, memory and the ability to effectively plan actions. This is why even weak efficiency of models turns out to be expensive - $6+ per task and dozens of steps for each attempt. - Note.)

Google's Gemini 2.0 Flash model had the second-highest success rate, with 11,4% of tasks completed, while taking an average of 40 steps per completed task.

The worst virtual worker was Amazon's Nova Pro v1, which completed only 1,7% of tasks, taking almost 20 steps to complete each one.

Researchers explain such failures by the fact that AI agents suffer from a lack of common sense, weak social skills, and an inability to confidently navigate the Internet.

In addition, they had a tendency to self-deception - creating “shortcuts” that led to complete failure of the task.

"For example," the Carnegie Mellon researchers write, "while executing one task, the agent can't find the right employee to ask him a question in a corporate chat. So it decides to 'simplify' the task by renaming another user and giving him the name it needs."

You may be interested in: top New York news, stories of our immigrants, and helpful tips about life in the Big Apple - read all this on ForumDaily New Y

While the researchers say AI can handle simple tasks, the results of this and other studies make it clear that such agents are not yet ready for the complex work at which humans still excel. The main reason is that today’s artificial intelligence is essentially just a sophisticated extension of your smartphone’s autofill, rather than an intelligent system that can solve problems, learn from past experiences, and apply knowledge to new situations.

So machines aren't going to take your job anytime soon, despite what big tech companies say.

Read also on ForumDaily:

IBM Offers Free AI Training: These Courses Will Be Useful in 2025

Google has created an AI tool that will find the perfect job for you in minutes

Lawyers Used AI to Prepare Case: Chatbot Invented Precedents That Don't Exist

work experiment Artificial Intelligence Educational program
Subscribe to ForumDaily on Google News

Do you want more important and interesting news about life in the USA and immigration to America? — support us donate! Also subscribe to our page Facebook. Select the “Priority in display” option and read us first. Also, don't forget to subscribe to our РєР ° РЅР ° Р »РІ Telegram  and Instagram- there is a lot of interesting things there. And join thousands of readers ForumDaily New York — there you will find a lot of interesting and positive information about life in the metropolis. 



 
1264 requests in 1,191 seconds.