Artificial Intelligence (AI) is the ability of machines to simulate intelligence and mimic human cognitive skills. That is, to collect and adapt external data, and from that data learn to make decisions and draw conclusions as a human could. The concept of artificial intelligence is inextricably linked to several terms:
Big data: AI needs enormous amounts of data to learn and evolve. Machine learning, or machine learning, is a subfield of AI that trains machines to perform tasks independently based on the data they receive, rather than simply acting on instructions.
Deep learning, or deep learning - a section of machine learning that uses neural networks to train machines to perceive large amounts of raw data (text, audio, video) and take it into account.
Natural language processing, or natural language processing, is a branch of AI that analyzes text, audio, and video data to synthesize and mimic live human communication.
A neural network is a computational system that learns by analyzing examples and gradually improves its abilities. A neural network acts on the principle of the central nervous system.
Types of Artificial Intelligence
The terms above fall into the Narrow AI or narrow/weak artificial intelligence category. This is a category of AI where the technology bypasses humans in a specific area or task. The voice assistants Alexa, Google Assistant, and Siri are the most obvious examples of narrow AI.
This category includes uncrewed cars, retail bots, facial recognition tools, spam filters, and even Google's search bot.
Narrow artificial intelligence is only effective in its domain. It is relatively inflexible: it will take time and serious resources to learn if the work environment changes. Also, this AI has no prospect of complete autonomy from humans, so you don't have to worry that your Roomba will get fed up one day and run you over in your sleep.
But General AI, or full/strong artificial intelligence, is precisely the kind of machine consciousness that Asimov wrote about, Cameron filmed, and Stephen Hawking and Elon Musk warned about.
Full AI allows the machine to apply its knowledge and skills to various domains. Its architecture and abilities are more in line with those of the human mind, and the AI can learn and perform tasks at its discretion.
Complete AI is still a theoretical concept. Even only matching the power and number of simultaneous processes with the human brain requires a computing capability of more than one exaflops, which is the estimated power of the human brain.
When will a computer be powerful enough
The machine managed to cross the one exaflops boundary for the first time in June 2020, when Japan's Fugaku supercomputer reached 1.42 exaflops. However, Fugaku's peak capability, while its average level is 442 petaflops.
A flop is a unit of computer performance equal to the number of moving-point operations per second.
Experts note that China is secretly working on two exascale computer sites, but they could not confirm this information because their data was not provided for analysis.
The ambition to create its exascale supercomputer Dojo also spoke about Elon Musk. After two years of announcements, Tesla opened a white paper on October 26, 2021. It turned out that the company had only developed its chip, but the supercomputer had not yet been physically assembled. Even after production, it will not be available to anyone outside of Tesla for a certain amount of time.
On January 24, 2022, Meta joined the chase: it unveiled the AI Research SuperCluster (RSC) supercomputer, which should reach five exaflops of performance. Development of the RSC should be completed in mid-2022.