Google company introduces human-smart artificial intelligence

In a statement released on May 18, experts at DeepMind affirmed that the most difficult tasks to create an AI like a human have been solved.

The London-based AI company wants to build an “Artificial General Intelligence” (AGI) that can handle tasks like a human without any formal training.

Artificial intelligence synthetic

Nando de Freitas, a researcher at DeepMind and professor of machine learning at the University of Oxford, says the AI ​​company has found a way to solve the toughest challenges in the race to achieve AGI.

AGI is a type of artificial intelligence that is capable of understanding or handling any task like a human without any formal training. According to De Freitas, the task of scientists now is to scale AI programs, which increase the data processing capacity and computing power to create a complete AGI.

Earlier this week, DeepMind revealed a new AI program called “Gato” that can complete 604 different tasks.

Gato uses a single neural network – a computer system with interconnected nodes that behave like the neurons in the human brain. DeepMind says the AI ​​program can talk, recognize images, stack blocks with a robotic arm or even play simple games.

There are many controversies

In response to a comment posted on The Next Web stating that the current human race will never achieve AGI, Mr. De Freitas has bluntly asserted that creating AGI is only a matter of time and scale.

“We’ve done that, and the task is now to make these models bigger, safer, compute more efficient, faster…”, Mr. De Freitas shared on Twitter.

However, he admits that humanity is still far from creating an AI that can pass the Turing test – a test of the ability to exhibit intelligent behavior of an equivalent or indistinguishable machine. with humans.

After DeepMind’s announcement about Gato, The Next Web countered that AGI is nothing more than virtual assistants like Amazon’s Alexa and Apple’s Siri, now available on the Internet. market and in people’s homes.

“Gato’s multi-tasking is like a console that can store 600 games, not a game you can play 600 different ways. It’s not synthetic artificial intelligence, it’s a bunch of AI that’s pre-trained and wrapped up.” Tristan Greene, writer of The Next Web comment.

According to some other reviewers, Gato can handle hundreds of different tasks, but this ability can affect the quality of each task. Tiernan Ray, columnist editor technology of ZDNet argues that Gato doesn’t really do many simple tasks well.

“On the one hand, Gato can do better than a dedicated machine learning program at manipulating the robotic arm that stacks blocks. On the other hand, it proved to be quite awkward with simple tasks like recognizing images or talking.”Ray commented.

For example, when conversing with humans, Gato mistook Marseille for the capital of France.

‘Its ability to talk to humans is only average, sometimes it even makes contradictory and meaningless statements,’ The editor continued.

In addition, the image recognition ability of this AI program is not really good. Gato wrote a caption for the photo that reads a man holding bread is “man holding up banana for photo”.

Is it a danger?

Contrary to the limitations of the current level of technology, some scientists see AGI as a future threat that could accidentally or intentionally wipe out humanity.

Dr Stuart Armstrong at Oxford University’s Institute for the Future of Humanity once suggested that one day, AGI will realize human inferiority and wipe us out.

He believes that machines will work at unimaginable speeds and eliminate humans to control the economy and financial markets, transportation, healthcare and more.

The doctor warns that a simple instruction to AGI to “stop human suffering” could be interpreted by supercomputers as “kill all humans”, because our language is so easy to understand. wrong.

Before his death, Professor Stephen Hawking told BBC:”The development of full-fledged artificial intelligence could mark the end of humanity.”

In a 2016 paper, DeepMind researchers acknowledged the need for a “self-destruct button” to prevent artificial intelligence from attempting to “topple humanity.”

DeepMind – founded in London in 2010 before being acquired by Google in 2014, known for creating an AI program that beat professional Go player Lee Sedol in a five-game match in 2014. 2016.