What Is Agi (Artificial General Intelligence)? Is It The End Of Deep Learning?

Sun, 18 Apr, 2021

Fetured image

What Is Agi (Artificial General Intelligence)? Is It The End Of Deep Learning?

What is AGI? Is it the end of Deep Learning?

Artificial General Intelligence (AGI) is the gin 'ability of an intelligent agent to understand or study any task that a human being can do.

This is the primary goal of some artificial intelligence research and is a common theme in science fiction and future studies.

AGI is also known as robust AI, full AI, or general intelligent operation. Some educational sources refer to the term "powerful AI" as computer programs that can be used to describe emotion, self-awareness, and consciousness.

Today's AI is thought to be decades away from AGI. In contrast to strong AI, weak AI (also known as narrow AI) is not intended to demonstrate human-like intellectual abilities and personality.

Instead, the proposed AI is limited to the use of software to learn or complete pre-learned problem-solving or logical tasks. Various criteria for intelligence have been suggested (usually the Turing test) but to this day, there is no definition that satisfies everyone.

However, there is a widespread understanding among artificial intelligence researchers that intelligence is needed to do the following:
1. Use reason, strategy, solve puzzles, give judgments in uncertainty
2. Common sense refers to knowledge including knowledge, planning and
practice
3. Communicate in natural language and connect all these skills with common
goals.


Other important skills are the ability to understand and work (e.g. moving and manipulating objects) in a world where intelligent behavior is to be observed. It includes the ability to recognize and respond to danger.

Many interdisciplinary approaches to intelligence (e.g., cognitive science, computational intelligence, and decision-making) emphasize the need to consider additional features such as the ability to form mental images, ideas, and autonomy.

These capabilities are present in most computer-based systems (e.g. computational creativity, automation logic, decision support systems, robots, evolutionary calculations, and intelligent agents), but not yet at the human level.

 

Problems that AGI needs to solve:

An AI is officially called "AI-complete" or "AI-hard" to solve the most difficult problems of computers, suggesting that solving them is equivalent to powerful AI beyond the general taste of human intelligence or the purpose-specific capabilities of powerful AI. Algorithm.


AI-complete problems include simple computer vision, natural language comprehension and dealing with unexpected situations when solving any problem in the real world.


Current computer technology alone does not solve AI-complete problems and requires human computation. Targeting captcha’s, this existence serves to verify human existence; For computer security to counter brute-force attacks.

Advanced Artificial General Intelligence Research:


Mark Gubroad used the term "artificial intelligence" in 1997 to discuss the implications of fully automated military production and operations. The term was reintroduced and popularized in 2002 by Shane Legg and Ben-Gortas. The research objective is very old.

 for example, the Doug Lennets Psych Project (launched in 1984) and Alan Newell's Throat Project. Pee Wang and Ben Goertzel described the 2006 AGI research activities as "publications and preliminary results". The first summer school at AGI was held in 2009 in Xiamen, China. Xiamen University Artificial Brain Laboratory and Open Cog.

The first university course was given by Todor Arnadov in 2010 and 2011 at Plovdiv University in Bulgaria. Presented a course at MIT 2018 AGI hosted by Lex Friedman and featuring several guest speakers. However, so far, most AI researchers have not focused much on AGI, with some saying that intelligence is too complex and will not be a complete replica in the near future.

However, very few computer scientists are active in AGI research, and many in this group contribute to AGI conferences. Research is very diverse and often takes precedence over nature. In his book introduction, Gortzell states that the time required to build a truly simple AGI varies from 10 years to a century, but the consensus in the AGI research community seems to be the timeline discussed. Unity is near (i.e between 2015 and 2045) and Ray Kurzweil is trustworthy.

However, many mainstream AI researchers have commented on whether this progress can be made faster. The 2012 meta-analysis of 95 such comments found a bias in anticipating that AGI would begin in 16-26 years, along with modern and historical predictions.

It was later discovered that the dataset listed some professionals as non-experts. Organizations that clearly follow AGI include the Swiss AI Lab IDSI, Nicene, Vicarius, Maluba, the Open Co Foundation, Adaptive AI, Lida, Numenta, and affiliate Redwood Neuroscience. In addition, organizations such as the

Machine Intelligence Research Institute and Open AI were established to influence the development trajectory of AGII. Finally, projects such as The Human Brain Project aims to create functional simulations of the human brain. AGI's 2017 survey categorized forty-five well-known "active R&D projects" that are explicitly and implicitly (through published research) research AGII, the largest three Deep Minds, the Human Brain Project, and Open AI.

In 2017 [citation needed] Ben Gortzal founded the AI Platform Singularity Net with the goal of facilitating democratic and decentralized control when it comes to AGI. In 2017, researchers Feng Liu, Yong Shi, and Ying Liu conducted an intelligence test on publicly available malicious AI such as Google A or Apple's Siri. At most, this AI reached an IQ value of approximately 47,.

which is equivalent to a six-year-old child in first grade. The average adult is about 100. Similar tests were performed in 2014, and the IQ score reached 27. In 2019 video game programmer and aerospace engineer John Cormack announced plans to do research on AGI.


In 2020, Open AI developed GPT-3, a language model that can perform a variety of tasks without special training. Although Gary Grossman argues in the Venture beat article that GPT-3 is not an example of AGI, some believe that much progress has been made in classifying it as a narrow AI system.