What is General Artificial Intelligence (AGI) in One Article
Background
From ancient mythology to modern science fiction, humans have dreamed of creating artificial intelligence for thousands of years. But the effort to synthesize intelligence didn't really begin until the late 1950s, when a dozen scientists gathered at Dartmouth College in New Hampshire for a two-month workshop to create machines capable of "using language, forming abstractions and concepts, and solving a variety of problems. Now reserved for humans and to improve themselves."
The workshop marked the official beginning of the history of artificial intelligence. But the two-month effort - and the many others that followed - only proved that human intelligence is very complex, and the complexity becomes even more apparent when you try to replicate it.
That's why, despite six years of research and development, we still don't have an AI that can match the cognitive abilities of a human child, let alone one that can think like an adult. What we do have, however, is a field of science that is divided into two distinct categories: narrow artificial intelligence (ANI), which we have today, and general artificial intelligence (AGI), which is what we hope to achieve.
What is generalized AI?
If your computer or robot can think and solve problems like a human, that's generalized artificial intelligence (AGI) . It's not just intelligence that performs specific tasks, but has a broad range of understanding and learning capabilities that allow it to deal with a wide range of complex situations just like a human.
AGI (also known as Strong AI or Deep AI, what is Strong AI? move to Layering Artificial Intelligence in What is Artificial Intelligence) is an AI framework based on the theory of mind. Essentially, the theory of AI at the level of mind involves training machines to learn human behavior and understand fundamental aspects of consciousness. With such a strong AI foundation, AGI can plan, learn cognitive abilities, make judgments, deal with uncertainty, and integrate prior knowledge or improve accuracy in decision-making. AGI helps machines perform innovative, imaginative, and creative tasks.
What can AGI do?
Most of today's AI is "narrow", meaning that it can only work in specific areas. For example, a speech recognition system may be very good at understanding what you say, but it won't understand what you're talking about. AGI, on the other hand, is able to understand the meaning behind language and use that information to make decisions, just like a human would.
Understanding the world
AGI is able to understand the world around us in the same way that you know why the sky is blue, and there are a lot of human knowledge points involved, such as the laws of physics, logical inferences, and objective laws.
Learning new things
It is able to learn new skills, like learning a new language or playing a new game. It is able to learn to accumulate its knowledge gradually like a newborn baby, able to learn and update itself.
Solving Problems
AGI is able to solve problems that require creativity and common sense, such as fixing broken appliances.
Adapting to New Environments
If AGI goes to a completely new place, such as Mars, it can quickly adapt and solve problems.
Challenges for AGI
Sensing the world
AGI needs to be able to perceive the world like a human, which includes seeing, hearing, and understanding everything around us. Although deep learning systems show great promise in the field of computer vision, AI systems lack the ability to perceive with human-like senses.
Thinking like a human
AGI needs to be able to perform complex thinking, such as solving problems that require logic and creativity. While AI-based machines are already capable of composing and drawing, demonstrating human-level self-optimizing creativity will require further development of AI.
Natural communication
AGI should be able to communicate naturally in human language, understanding our emotions and humor. For AI robots to operate in the world, human interaction is inevitable. Therefore, these robots will need to understand humans, facial expressions, and changes in tone. Considering the perceptual challenges discussed above, AI systems that can empathize with emotional connections seem like a distant reality.
Self-improvement
AGI needs to be able to learn and improve itself, just as a child grows.
Pathways to AGI
There are significant challenges to realizing robust AI. For example, Fujitsu has built one of the fastest supercomputers called the K computer. Although this computer broke the barrier of 1,000,000,000,000,000 floating-point operations per second, it took more than 40 minutes to simulate one second of neural activity, thus blurring the vision of strong AI. However, the future of AI is bright, as it can have a large-scale impact on society by utilizing its ability to handle complex situations such as economic crises.
Various methods have been tried and tested to gain human-like intelligence. Listed below are some of the core methods to realize AGI.
1. Symbolic methods
Symbolic approaches involve the use of logical networks (i.e., if-then statements) and symbols to learn and develop a comprehensive knowledge base. This knowledge base is further expanded through the manipulation of these symbols that represent essential aspects of the material world. This approach mimics a higher level of human brain thinking.
Theoretically, the symbolic approach can accomplish higher levels of logic and thinking, but in reality, it lacks the ability to learn lower level tasks such as perception. An apt example of the symbolic approach is the CYC project started by Douglas Lenat of Cycorp in 1980 to advance work in the field of artificial intelligence.CYC has a large knowledge base, a logical system, and a powerful representational language.CYC has a large knowledge base, a logical system, and a powerful representational language.CYC has a large knowledge base, a logical system, and a powerful representational language.
2. Connectionist approach
The connectionist approach is a sub-symbolic approach that utilizes architectures similar to the human brain (e.g., neural networks) to create general-purpose intelligence. The approach expects higher-level intelligence to emerge from lower-level subsymbolic systems (e.g., neural networks), which has not yet happened. Deep learning systems and convolutional neural networks (e.g., DeepMind's AlphaGo) are good examples of the connectionist approach.
3. Hybrid approaches
Hybrid approaches are a mix of connectionist and symbolic systems. The leading architectures in the AGI competition tend to use hybrid methods, e.g., the CogPrime architecture. It represents symbolic and subsymbolic knowledge through a single knowledge representation (called AtomSpace). The famous social humanoid robot Sophia was created by Hanson Robotics and OpenCog with the help of the neural architecture CogPrime.
4. Holistic Organismic Architecture
Experts believe that a truly generalized AI system should have a physical body and learn from physical interactions. While there is no such system yet, the closest thing is Sophia - a humanoid robot that can mimic human gestures and facial expressions and indulge in conversations on predetermined topics.
Future Trends in AGI
The development of artificial intelligence (AI) came to the forefront during the New Crown Pneumonia epidemic, with humanoid intelligence evolving at a faster rate than ever before. While a complete AGI system is not currently realistic, the latest trends in AI could advance AGI and significantly accelerate its development. Here are the top 10 AI trends that could push AGI forward.
NLP Development
Natural Language Processing (NLP) is an AI technology that understands human language and greatly reduces the need to interact with screens. AI-enabled devices can convert human language into computer code used to run applications.
Recently, OpenAI released GPT-3, the most advanced version of NLP to date.GPT-3 uses over 175 billion parameters to process language. In addition, OpenAI is working on GPT-4, which is expected to process about 100 trillion parameters for comprehensive language processing. As AI advances, it is certainly possible to develop machines that can interact with humans in a way that is as good as the real thing.
Virtual Worlds
Virtual worlds are booming as companies and individuals explore immersive technologies to work and interact in this virtual world. According to DappRadar, Inc. in November 2021, users spent about $106 million on virtual property in virtual worlds, mainly digital land, luxury yachts and other assets.
Considering this trend, AI and ML are expected to drive the growth of virtual worlds by building virtual worlds through virtual AI chatbots to give users a real feel of the world.
High level of automation
From Robotic Process Automation (RPA) to Intelligent Business Process Management, many industries are utilizing AI and machine learning technologies to automate multiple processes. Hyper-automation adds an extra layer of advanced automation capabilities as it expands an organization's automation outlook. The hyper-automation market reached $600 billion between April 2021 and 2022, according to a Gartner report.
More governance
Algorithmic biases can appear suddenly due to lack of model governance. In this regard, AI experts should pay more attention and ensure that AI/machine learning models do not create biases or make wrong decisions. Most recently, in October 2021, Twitter admitted that its algorithms produced biases in favor of right-wing politicians and news media.
In a similar incident in 2015, Amazon realized that its algorithm for hiring employees was biased against women. This was because the algorithm scrutinized resumes from the past decade, and since most of them were male, it was trained to be biased against women.
Examples such as these will pave the way for an increase in positions such as Chief Artificial Intelligence Officer and Chief Artificial Intelligence Compliance Officer. This is expected to increase in the near future with the rapid adoption of AI/Machine Learning.
The rise of low or no-code AI
Today, the demand for skilled AI engineers is high. Organizations are always on the lookout for engineers who can develop AI algorithms and tools for their business operations. Low-code and no-code AI solutions can solve this problem by providing intuitive interfaces that help create complex systems.
Often, low-code solutions offer drag-and-drop options that simplify the application building process. Additionally, NLP and language modeling technologies can be used to provide voice-based instructions to complete complex tasks.
Increasing labor force
Concerns about AI replacing human jobs have been around for a long time. In fact, organizations seem to be using AI/ML models to collect and analyze data and gain insights that help in making business decisions. In this case, organizations have to make their employees and AI machines work together.
Several departments including sales, marketing and customer service are already using AI/ML systems to aid their operations. However, this has not reduced the potential reliance on humans. In fact, it has only increased the efficiency of these departments. This trend is only going to rise from here on out.
Conversational AI Chatbots
Conversational chatbots refer to AI-enabled virtual assistants. They perform natural conversations and certain rule-based actions, such as responding to queries or resetting passwords. These chatbots have replaced customer service, thus significantly reducing the operational costs of businesses. As the field of natural language processing continues to evolve, conversational AI chatbots could revolutionize the field of artificial intelligence in the future.
More focus on AI ethics
Recently, there has been a significant increase in AI use cases across verticals. Despite the benefits of AI technology, the potential risks of AI cannot be ignored. As a result, the focus on AI ethics will rise in the coming years, as things could be reversed if these technologies are not used for good.
Recruitment process based on artificial intelligence
As the epidemic has already affected the hiring process, organizations are now expected to use more AI/machine learning based systems as the virtual world replaces the traditional physical world. Additionally, employers are expected to use AI-powered tools to manage the hiring process as language modeling technology advances and conversational AI chatbots become more sophisticated.
Quantum of AI
While considerable progress has been made in the field of AI over the past few years, quantum AI can push the boundaries of AI even further as quantum computing accelerates machine learning algorithms and delivers results in a much shorter period of time. Quantum AI can remove the barriers to AGI as it can help create powerful databases in the shortest possible time by analyzing large amounts of data from books, articles, blog posts, and other similar sources.
Summary
General Artificial Intelligence (AGI) is a very cutting-edge field of technology that aims to create intelligent systems that can think and solve problems like humans. While it's still in its early stages, as technology advances, we may see AGI working in the future in a variety of fields, from helping us with household chores to solving complex scientific problems. For those without a technical background, understanding the basic concepts of AGI and the changes it could bring can help us better prepare for a future full of possibilities.
