Last Updated on August 15, 2022
In the 21stcentury, automation is extremely beneficial to the global economy. The future of our modern society is highly dependent on all the innovations around automation and other technology.
Ever since we began to flirt with the idea of automating mundane tasks, the world has been faced with an unprecedented level of digitization. And this has led to Artificial Intelligence (AI).
We encounter plenty examples of AI in our day-to-day life, whether it is in search engines, language recognition programs, virtual assistants, video games, smart devices, GPS, etc… Our lives are filled with intelligent agents that do dynamic tasks.
Artificial intelligence is very good at what it was programmed for. It is very good at analyzing data and seeking patterns from it. And they can learn to some extent from the data feed. As of today, AI has wrapped its digital tentacles around many industries.
For instance, the stock market is being shaped by artificial intelligence. Many trading firms are relying on this technology to analyze millions of data points and execute trades of their own in real-time. Thus mitigating risks and creating opportunities for higher returns.
But AI has some limitations. For example, the AI that is used for playing chess on a computer or in a search engine is only going to be concerned with making winning chess moves or providing the best result on the web.
For the stock market, the intelligent agent is only good for analyzing stock data, as it is programmed to do so. These expert systems are useful for some specific kind of purposes and you cannot scale them.
Artificial intelligence is about putting commands in a box through data entry or voice. Basically, you get the output of only what you put in the algorithm. But now another revolution is taking place to go further than what AI can do.
A paradigm shift is taking place in the field of artificial intelligence. Today, the action is around machine learning. Algorithms that learn from raw perceptual data. Basically, the same thing that human does. The result is an intelligent agent that is not limited to one domain.
The same system can learn to do multiple tasks. Sophisticated computer algorithms can create their own conclusions without human input. These algorithms can recognize patterns in behavior and create their own logic or take decisions.
Artificial General Intelligence often called strong AI has the ability to adapt, plan and learn. It has the potential to perform any intellectual task at or above human-level reasoning. So what exactly is artificial general intelligence?
Contents
What is Artificial General Intelligence?
In terms of definitions, artificial general intelligence (AGI) is a type of AI that mimics human cognitive ability and is capable of doing many human-level tasks. Its capabilities are based on a theory called “theory of mind AI” that trains machines to understand and mimic human thought processes, emotions and beliefs.
Many individual problems require general intelligence. The idea behind artificial general intelligence is to create a machine that can reproduce human reasoning while incorporating the computational advantages of a computer.
General AI, also called strong artificial intelligence (strong AI) aims at creating a program that is better at solving particular tasks than a human. However, it is still a long way from becoming a fully-fledged AI.
In the 1960s, AI researchers attempted to encode knowledge with expert systems. These systems were only effective in narrow domains and required a large effort from subject matter experts and programmers.
The most famous milestone in AI research was achieved by Alan Turing in the 1950s with the Turing Test. This AI program is capable of fooling humans by interpreting millions of texts. Currently, the reigning AI attorney is ELBOT.
Moreover, IBM Deep Blue algorithm has beaten the world chess champion. For instance, a machine capable of reading and translating two languages would be able to understand the arguments of an author and solve multiple problems at once.
General intelligence is defined as the ability to do many different things, from understanding language to making things. The term is also used to describe a computer system that is capable of understating situations and context.
For instance, an AGI system can enjoy jokes and understand the context in which they are made. However, this level of intelligence in machines is more advanced than just recognizing jokes, making them, or generating them.
Ray Kurzweil a director of engineering at Google, claims that we will have advanced AI that passes the Turing Test in 2029. Kurzweil also believes that it will perform the same number of calculations per second as humans.
The law of accelerating returns means that technological development will continue at such a rapid pace that we’ll have the equivalent of 20,000 years of human progress in the 21st century. This belief is based on rapid changes in computer processing power and brain-mapping technologies.
Why do we need artificial general intelligence?
We are living in an era where our organizations need to adapt to new levels of complexity. With today’s increasingly complex systems, current organizational structures and rigid operating models aren’t sufficient.
In today’s current complex systems, flow-to-the-work models are needed, which allow people to shift and move seamlessly from one task to another. The majority of AI systems are specialized and perform specific tasks.
For example, a system that learns to name people in photos cannot distinguish between a dog and an elephant. These systems also lack the commonsense knowledge needed to engage in conversations or answer questions. Therefore, they’re not representative of a general AI system.
The goal of artificial general intelligence is to produce machines that learn as humans do. Most AI systems are related to narrow domains with specific capabilities and applications. They have very limited adaptive learning abilities.
In other words, we need to develop an intelligent agent capable of reasoning about various alternatives and scenarios. If these systems can’t learn to reason about different scenarios, we’ll unable to fully use them in real life.
An application of general AI is computer-generated intelligence. The development of AGI has led to the creation of supercomputers. IBM Watson is an example of a supercomputer.
These machines combine AI with computing power to simulate the human brain and the Big Bang theory. In addition, the expert system mimics human judgment. They can predict the molecular structure of molecules in a human’s blood and even prescribe medicine based on the data of a patient.
A General AI system is the technology that trains machines without human guidance. By incorporating machine learning algorithms, we can make a computer that learns from experience and memory.
For example, a polyp detection algorithm can be trained without a human. However, it is not possible to create rules for every aspect of intelligence. With AGI, unsupervised learning algorithms can perform tasks that are more complicated than supervised learning systems.
These machines will learn in a wider range of situations. As long as we use them responsibly, they will contribute to human happiness and welfare also. Once we can make strong AI, we may be able to create a superior robot that can perform many tasks.
Artificial general intelligence technology has endless applications. From self-driving cars to video games, advanced AI can do many tasks. The capabilities of this technology have even advanced to the point that it can play chess without a human player.
When you think of AGI, you may think of robots. For instance, Dr. Igor Aleksander from Imperial College, London, UK argued that the principles of a conscious machine already exist.
Igor Aleksander said that if there are principles for a conscious machine, it would take 40 years to teach it language. While the exact role of consciousness in a strong AI is debatable, many researchers consider research into consciousness to be vital for AGI development.
Nowadays, AI research focused on narrow tasks, the human mind on the other hand is more versatile and general. Only when an AI program can match the level of human comprehension and compete with us, it can be considered strong.
Uses of artificial general intelligence
Natural language processing
Natural language processing (NLP) is an AI technology that comprehends human natural language and speech. This AI-enabled program can turn human languages into computer languages so that the machine executes the order.
OpenAI’s GPT-3 is the most advanced NLP version to date. GPT-3 analyzes over 175 billion parameters to synthesize languages. OpenAI is also working on GPT-4 which is estimated to handle around 100 trillion parameters for language processing.
The global market size of natural language processing was estimated at USD 13.5 billion in 2021. A CAGR increase of 27.2% is expected during 2022-2030. The market value is projected to reach USD 91 billion by 2030. These numbers show the growing importance of NLP.
Metaverse
The Metaverse is also considered to be a disruptive technology in the 21st century. It is an immersive world where people interact using their digital avatars. It is an internet where instead of looking at it on a screen, you are inside it.
And just like the internet generates a lot of data and uses some kind of AI to process them to deliver a better experience, the metaverse is also heavily reliant on artificial intelligence. Machine learning and AI are very important to the metaverse and they can be a major boost for AGI because of the sheer amount of data it will generate.
Robotic Process Automation
Many industries have already leveraged AI, ML, and IoT technologies to automate their processes and operation, but administrative processes have remained pretty much the same. Hence robotic process automation (RPA) can leverage and brings more intelligence to business process management.
Robotic process automation is an AI-based software that is used to process administrative tasks. RPA is set to add another layer of advanced capabilities to administrative processes to increase the efficiency of bureaucracy.
Low-code and no-code AI
Demand for skilled AI engineers is high but supply is low. Organizations are continuously looking for developers to write AI algorithms to meet their business operations. Low-code and no-code AI can address this issue.
It is a technology that enables non-AI experts to implement and test AI solutions without the need of an AI engineer. It is a code-free solution that helps build AI and ML solutions faster with fewer efforts without having to write a single line of code.
Chatbots
An AI chatbot is like a virtual assistant that can carry out natural conversations and perform certain operations, such as responding to queries. These chatbots are entering a lot of companies hence replacing customer support agents.
With NLP and conversational AI entering different industries as well as our life, it will possibly revolutionize the field of AGI. The artificial agent will be interacting more and more with humans in their natural environment.
Quantum AI
Considerable progress has been made in the field of AI, but quantum AI could further push AI boundaries. Quantum computing could speed up AI progresses to achieve greater results in a shorter time. Quantum AI could reduce AGI barriers by creating a strong knowledge base system.
Autonomous Vehicles
Autonomous vehicles have been touted as the next big thing as big industry players such as Tesla, Uber, and Waymo are working on the technology. But for a vehicle to be fully autonomous a lot of real-time data need to be processed.
An autonomous vehicle will require general AI to deal with all the possible road scenarios that could happen during a journey. General AI will give the vehicle human-level intelligence while navigating.
Advancement in AGI technology
Strong AI is about training machines to understand humans better. The machine uses the theory of mind framework not just to replicate or simulate human behavior but also to differentiate needs, emotions, beliefs, and thought processes.
Researchers and scientists are working to find ways to make machines conscious by implementing a full set of cognitive abilities. Big tech companies like Google, Microsoft, IBM, and Facebook and research laboratories such as Elon Musk’s OpenAI are all working on AGI.
OpenAI’s program called DALL-E 2 can seemingly create amazing images from any text description in natural language. Another OpenAI project called GPT-3 can talk about just about anything.
In May 2022, DeepMind, a subsidiary of Google released a program called Gato which is a generalist agent that can process every task a company presents to it. Gato is a trained model that is capable of chatting, image captioning, playing, and even controlling a real robotic arm.
Gato can decide based on context and act accordingly. Another Google program called Pathways Language Models or PaLM is an AI capable of handling millions of different tasks. PaLM can also learn and reason.
PaLM can outperform humans in language and reasoning. Moreover, DeepMind has introduced Flamingo which is a visual language model. The program can be used for image and video understanding tasks. For now, the NVIDIA DGX A100 supercomputer is considered to be the most advanced AI system in the world.
In 2022, another google program called LaMDA make the headline. An engineer at the company claimed that LaMDA is sentient. LaMDA which stands for Language Model for Dialogue Application is a conversational neural language model that can engage in a free-flowing conversation about seamlessly endless topics.
These programs are becoming very good at language understanding, natural language writing, creating and describing images and videos as well as performing specific tasks.
However, none of these companies are focusing on developing a basic, underlying AI technology that can completely mimic the contextual understanding of human intelligence.
At the very moment, AI can only give the appearance of intelligence but not reproduce human common sense. Even if the AI we have now show impressive capabilities, they still follow predetermined scripts and variables.
Is it possible to create AGI?
Achieving Artificial General Intelligence is not that easy. For a machine to become a robot and a robot to reason like a human is not simple. Even defining what a robot is isn’t easy because there may be general agreement that it’s a machine.
The definitions vary from expert to expert. Most people agree that for a machine to qualify as a robot it must show some form of intelligence and have the capacity to perform tasks commonly done by humans or animals.
In addition to this, the robots would require to have some human or animal physical features like feet, arms, eyes, or ears. Moreover, the biggest challenge facing the future of AI is the ability to train itself without any human input.
An AI system learns by analyzing data to seek patterns and deliver relevant information and predictions. But they understand little about why things happen and lack human common sense.
Humans develop common sense from an early age. Kids don’t analyze data to acquire skills instead they observe their environment and learn from the consequences.
For instance, a child playing with legos understands that each module of the game exists in a 3-dimensional world and has physical properties such as weight, shape, and color. The child also knows that if the modules are stacked too high they might fall.
And over time as the child grow and its capabilities are enhanced at the same time. The child grows into a fully functioning, generally intelligent adult. This type of ability is lacking in today’s AI and is still impossible to implement in them.
No matter how sophisticated the AI is, the latter remains completely unaware of its existence in its environment. Artificial intelligence also has no understanding that present actions have an impact on the future.
To achieve AGI, researchers should shift their attention to replicating the contextual understanding of humans. Experts say that machines need to start learning more as kids do.
A subdomain of AI is already showing these capabilities. Machine Learning (ML) as its name suggests can learn from data. Machine learning is an algorithm that can process massive amounts of data sets.
And while doing so ML can learn from that data and improve itself. After each process, the algorithms become more efficient, accurate, and sophisticated. However, it is unrealistic to achieve a human-level AI only with training data.
An AI system will need to experience the physical environment to be able to understand real-world concepts. In addition to machine learning, the intelligent agent needs mobile sensory.
Just like a human has senses and learns to use them from an early age. The intelligent agent needs to be equipped with sensory pods to learn first-hand from the physical world, moving objects, performing tasks, acting, and more importantly learning from the consequences of those actions.
With sensory devices, the artificial entity can learn and understand how things work and operate. Which would have been impossible only with text-based and image-based systems. Here’s what AI needs to become AGI:
- Sensory capabilities
- Motor skill and agility
- Natural language processing
- Problem
- Navigation
- creativity
- Social and emotional engagement
Artificial intelligence pioneers conjectured that every aspect of human learning and intelligence could be described and emulated, thereby allowing machines to solve human problems and improve themselves.
But a more pressing issue is that it is difficult to quantify how much data a brain is needed to appropriately understand something. Furthermore, humans interpret everything in the present context and the logic applied has already been experienced and learned.
While the world is certainly on the road to artificial general intelligence, many researchers claim that we’ll never have the opportunity to duplicate human intelligence at a scale acceptable to us.
As we develop these programs, we must consider the common qualities shared by real-world general intelligence. While the AGI community intuitively believes that real-world general intelligence shares some characteristics, there is less consensus about what these characteristics are.
A common problem in this context is the need for an environment that scales up to the real world. Without a realistic environment, the machine could not develop enough intelligence to be useful.
A better simulation environment is necessary to improve strong AI performance. MMORPGs are a prime example of such an environment. Using these environments as a foundation for AI can help solve many problems that are common to humans.
Challenges of artificial general intelligence
The future of AGI will largely depend on the advancement of machine learning and further development in artificial intelligence. The data explosion generated from the highly digitalize world will create fertile ground for machine algorithms and robotic approaches.
Achieving a human-like level of reasoning may become a reality soon. However, achieving this may not be that easy. One challenge in creating a machine that can perform human-level intelligence is common sense knowledge.
Human intelligence is not just limited to doing complex tasks. Humans also have common sense knowledge. Moreover, humans are capable of multitasking. For instance, a human can be a doctor, but he can also be good at writing, singing, playing chess, knowing several languages, riding a bike, playing violin and list goes on.
Further development is needed in the field of AGI. While machine translation services are becoming increasingly accurate through deep learning, they still lack context over multiple sentences, which is something toddlers are naturally good at.
For instance, if a general intelligence is to solve problems just like humans would do, for example, in machine translation, the system will need to understand language both in reading and writing (context).
Then the machine will need to understand the user’s argument (reason) and know what is being talked about (knowledge). The machine will need to reproduce the user’s original intent (social intelligence). All of these should be done simultaneously for the machine to reach human-level intelligence.
There are other aspects besides intelligence that make a human, human:
- Consciousness – which is having subjective experience and thought.
- Self-awareness – which is to be aware of oneself (both as a separate individual or one’s own thoughts).
- Sentience – which is the ability of perceptions, feelings, and emotions.
- Sapience – which is the development of wisdom.
More importantly, all these traits have moral dimensions. Moreover, if a machine develops all these capabilities, the machine may be eligible to have rights. As such, ethical agents with existing legal and social frameworks must be implemented in the program.
Moreover, narrow AI solutions are often mistaken for human-level intelligence. The key to general AI is that it should be able to function across multiple domains. A general AI system should be able to take data from many fields, make decisions and take actions to improve its state.
Unfortunately, the reality of this type of intelligence is far from perfect, but it is possible to make it more efficient and believable. The key is to get data from multiple sources and process it into creative solutions.
But this is just the first step in a long road toward AGI. Further R&D is needed in the field of augmented intelligence, machine learning, and artificial intelligence. While the role of consciousness in strong AI is controversial, many researchers see research on the topic as essential.
Furthermore, there are no concrete tests for consciousness and the role of sentience in these systems is still unclear. Some people think that a fully intelligent machine with neural correlates of consciousness may have sentience.
Alternatively, sentience may be a natural attribute of a fully intelligent machine. The possibilities are endless. However, other kinds of issues may also arise and it may even affect mental health.
Moreover, you all know that AI can be used for unethical purposes, such as spreading misinformation and DeepFakes. Deepfakes are the new type of artificial intelligence system can create even worst fabricated news.
While DeepFakes can be detected by close scrutiny, their potential to propagate misinformation has only become worse with time. An AGI could lead to even worst devastating consequences.
Another set of studies predicts a mixed bag of effects. While some experts predict that AGI will lead to a new era of technological advancement, there are also inevitable consequences. Many experts predict that half of the world’s population will be unemployed by 2050 with another human-like intelligence on earth.
In the 2030s, unemployment will start to rise dramatically, causing social and political gridlock. During this time, the global order will fall apart into nation-states, mega-corporations, local militias, and organized crime.
In the long term, AI will exacerbate existing inequalities. Autonomous weapons and broader military use are also two prime examples of unethical applications. The use of AGI can spark a world war. Some believe it could result in billions of deaths by the end of the century.
Recently, Google received backlash after the release of Project Maven, an AI software that analyzed drone footage. There is also concern about AI machines could ally themselves with human resistance or supporters, causing a major global conflict.
As strong AI becomes more advanced, it is important to be cautious about what society might do if it encounters such a problem. If there is no regulation in place, the future may be far different.
This doesn’t mean that AGI will turn the military into Terminator-style killer robots but it does mean that it will continue to develop its capabilities and its applications in the military. Experts argue that a machine with human traits may be a threat to dignity or even humanity.
The challenges of AI technology go beyond ethical concerns. In many cases, AI will collect data that are not wholesome or appropriate. Consequently, there are legal and ethical concerns with this technology.
A societally appropriate approach to AGI development is necessary to avoid any pitfalls as well as improve moral status in the system. This is especially important if AGI is used to improve day to day life of people and society.
Final word
Artificial general intelligence is already being developed in various fields. Some of these machines will be useful for many purposes and may even take over the world from us so that humans can concentrate on more important tasks.
But it is also crucial to remember that AGI is not a magical machine. Some people believe that this type of AI will take over the world and destroy everything. People are notorious for doing stupid things and machines may be no different.
While it is still difficult to predict how artificial general intelligence will change the future of humanity, it is certainly worth investigating. However, one important thing to remember is that there is still a long way to go before we can create a fully intelligent machine.
0 Comments