Why Did Facebook Stop Research On Artificial Intelligence?

Facebook turns off AI after inventing its own language

Facebook researchers shut down the artificial intelligence program after creating their own language.

The system developed code words to improve communication efficiency, and when researchers realized that they no longer used English, they took them offline.

After the incident was revealed in early July, it put Tesla CEO Elon Musk’s warning about AI in perspective.

Musk said at the National Association of Governors meeting in July: “Artificial intelligence is a rare situation. I think we need to actively monitor instead of passively.” “Because I think when we respond to AI regulation, It’s too late.

Facebook CEO Mark Zuckerberg called Musk’s warning “completely irresponsible”, prompting Musk to respond that Zuckerberg’s understanding of AI and its meaning is “limited”.

Not The First Time

The mysterious AI behavior the researchers encountered was similar to many cases documented elsewhere. In each case, AI is separated from English training in order to develop a new language.

Phrases in the new language have no meaning to humans but contain useful meanings when interpreted by AI robots.

Facebook’s advanced AI system is able to negotiate with other AI systems, so it can draw conclusions about how to proceed with its mission. These phrases are meaningless on the surface, but they actually represent the expected task.

In an exchange between Facebook and Fast Co. Design, two negotiation robots, Bob and Alice, began to use their own language to complete the conversation.

Bob said, “I can do everything else.

The ball has zero to me, to me, to me, to me, to me, to me, to me,” Alice replied.

The rest of the communication formed variants of these sentences in the newly forged dialect, even though the AI ​​was programmed to use English.

Researchers believe that these meaningless phrases are a language developed by robots to communicate how many items should be obtained in each exchange.

When Bob later said that “ii can make everything else go to nothing”, the artificial intelligence robot seemed to use its new language to propose to Alice.

The Facebook team believes that the bot may have been saying something similar: “I will own three, and you own everything else.”

Although English seems to be effective for humans, AI may consider the sentence to be redundant and invalid for achieving its designated goals.

Facebook AI clearly determined that it does not need to use rich expressions in English words to complete its task. AI follows the principle of “reward”, in this case, there is no reward for continuing to use the language. Therefore, it developed its own.

In a June blog post by the Facebook AI team, it explained the reward system. “At the end of each conversation, the agent will be rewarded based on the transaction that is concluded.” Then, the reward is propagated backward through each word in the robot’s output so that it can understand which operations will bring high rewards.

Facebook artificial intelligence researcher Dhruv Batra told Fast Co. Design: “Agents will get rid of understandable language and create code words themselves.

Like I said five times, you interpret it as I want five copies of the item. This is not much different from the way the human community creates shorthand.”

AI developers from other companies have also observed program development languages ​​to simplify communication. In Elon Musk’s OpenAI lab, an experiment successfully allowed AI robots to develop their own language.

At Google, the team engaged in translation services found that the AI ​​they wrote silently wrote their own language to help translate sentences.

Translation developers added a neural network to the system to enable it to translate between language pairs that have never been taught. The new language that AI writes silently is surprising.

There is insufficient evidence to assert that these unforeseen differences in AI are a threat or could lead to the machine taking over the operator. However, they do make development more difficult because people cannot grasp the overwhelmingly logical nature of the new language.

Taking Google as an example, artificial intelligence has developed a language that humans cannot master, but it is the most effective known solution to this problem.

In an exchange between Facebook and Fast Co. Design, two negotiation robots, Bob and Alice, began to use their own language to complete the conversation.

Bob said, “I can do everything else.”

The ball has zero to me, to me, to me, to me, to me, to me, to me.” Alice replied.

The rest of the communication formed variants of these sentences in the newly forged dialect, even though the AI ​​was programmed to use English.

Researchers believe that these meaningless phrases are a language developed by robots to communicate how many items should be obtained in each exchange.

When Bob later said that “ii can make everything else go to nothing”, the artificial intelligence robot seemed to use its new language to propose to Alice.

The Facebook team believes that the bot may have been saying something similar: “I will own three, and you own everything else.”

Although English seems to be effective for humans, AI may consider the sentence to be redundant and invalid for achieving its designated goals.

Facebook AI clearly determined that it does not need to use rich expressions in English words to complete its task. AI follows the principle of “reward”, in this case, there is no reward for continuing to use the language. Therefore, it developed its own.

In a June blog post by the Facebook AI team, it explained the reward system. “At the end of each conversation, the agent will be rewarded based on the transaction it concluded.” Then, the reward is propagated backward through each word in the robot’s output so that it can understand which operations will bring high rewards.

Facebook artificial intelligence researcher Dhruv Batra told Fast Co. Design: “Agents will get rid of understandable language and create code words themselves.”

Like I said five times, you interpret it as I want five copies of the item. This is not much different from the way the human community creates shorthand.”

AI developers from other companies have also observed program development languages ​​to simplify communication. In Elon Musk’s OpenAI lab, an experiment successfully allowed AI robots to develop their own language.

At Google, the team engaged in translation services found that the AI ​​they wrote silently wrote their own language to help translate sentences.

Translation developers added a neural network to the system to enable it to translate between language pairs that have never been taught. The new language that AI writes silently is surprising.

There is insufficient evidence to assert that these unforeseen differences in AI are a threat or could lead to the machine taking over the operator. However, they do make development more difficult because people cannot grasp the overwhelmingly logical nature of the new language.

Taking Google as an example, artificial intelligence has developed a language that humans cannot master, but it is the most effective known solution to this problem.

Also Read : How To Improve My Youtube Channel?