Artificial general intelligence is an AI agent that is fully capable of the advanced problem solving, critical thinking, and cognitive function of a human being. This is currently a hypothetical branch of artificial intelligence that researchers and businesses are actively working to develop.
Artificial intelligence (AI) has seen rapid expansion in recent years. It has also been a hot topic for most businesses that are trying to navigate the modern world. Professionals at all levels, from executives to entry level workers, have demonstrated excitement (and more than a little concern) about the future of AI. And with the sudden boom of available generative AI, many initially wondered if AI would soon replace human employees in many fields.
So far, AI has not made humanity obsolete. The business world has instead begun to tap into the capabilities of AI to improve efficiency and optimize workflows. This may be in part due to the fact that the AI currently available is what researchers consider weak AI. Strong AI, also known as artificial general intelligence (AGI) that has human-like capabilities could have a more drastic effect on the world—provided this currently-hypothetical dream ever becomes a reality.
AI (or weak artificial intelligence) is technology that has the ability to perform specialized tasks, often better than humans could. The AI most people are currently familiar with is weak AI, and despite its limited capabilities it has been a highly successful AI development. Weak AI includes two types of machines: reactive machines and limited memory machines. Reactive machines respond to immediate requests but cannot store any data or learn from past experiences. Limited memory machines do store information and learn as they encounter new data.
Many people don’t realize how much weak AI is a part of their daily lives. Some examples of AI include:
- Chat GPT, a generative language model
- Email spam filters
- Music application shuffle, like Spotify
- GPS navigation, like Google Maps
- Autocorrect features on SMS messaging'Smart assistants like Siri or Alexa
Strong AI, or artificial general intelligence, is fully capable of human cognitive thinking, including the ability to problem solve, think critically, and learn. AGI can learn information and then apply it in a new scenario, as well as adapt to changing environments. Right now, it does not exist, but humans have long imagined how it might look—from the quirky or competent androids of Star Wars and Star Trek to the malevolent AGIs represented in 2001: A Space Odessey and the Terminator franchise.
Weak AI can currently only perform specific tasks, but they can do each task very well and beyond human capabilities. Typically, the use of AI reduces human error and enhances efficiency. Strong AI will likely offer similar benefits. It will be able to perform tasks without human error and at an incomprehensible speed.
The main difference between these two is that strong AI will have the ability to genuinely learn and think like a real human being is able to. Weak AI is only able to work exactly as its programming instructs it. When it is able to produce its own ideas and improve on its own, then it will become strong AI
AGI may still be a long way off, but researchers aren’t letting that stop them from looking beyond it to the next major milestone. Scientists predict that after developing true artificial general intelligence, the next step will be to create artificial superintelligence. Just like strong AI, it will be capable of everything human cognitive functions can do. However, superintelligence will be completely self-aware and surpass human intelligence. It would be able to perform beyond what a human could. Currently, superintelligence is hypothetical, just like AGI.
Before current AI can become AGI, it will need to master certain skills:
AGI will need to develop the ability to see and to judge spatial characteristics. Currently, AI cannot fully identify colors. For example, AI is often unable to tell if a stop sign is red if there are stickers on it. A human and an AGI would not get stuck on the multitude of colors and be able to identify the object as a stop sign. AI is also unable to extract three-dimensional information from a static image. AGI will be able to look at an image and know that something is a sphere, even if the image is only two dimensional.
Humans can also perceive what direction sound comes from and understand background conversations. AI cannot do that, but AGI would be able to judge spatial distances and pick up on quieter conversations happening behind other louder talking.
Human hands can perform a wide variety of tasks without any effort. For something to become true AGI, it would need to develop similar fine motor skills. That would allow AGI to solve puzzles and maneuver objects.
Even young children can read multiple sentences and comprehend them beyond the current capabilities of AI. For AGI to develop, it will need to be able to read and watch all types of communication and understand it. That includes understanding the language itself as well as the meaning behind it. This skill will provide an essential foundation for AGI to learn everything it needs to know to perform more advanced tasks.
For AGI to develop, it will need to be able to identify and solve a problem. Right now, no known systems possess the commonsense skills necessary to solve a problem effectively without prompting. AGI in the future may be able to diagnose problems and address them.
There are AI models that can navigate and project three dimensional spaces. However, true AGI will have these abilities without any human intervention. Currently, AI models can do a lot of these but not without human guidance. AGI will be able to do simultaneous mapping and localization all on its own.
AI as it exists right now isn’t able to truly create anything. It can follow prompts and string words together, but it cannot make something unique. AGI will be able to truly create. In fact, experts predict that AGI will need to be able to rewrite its own code and find novel improvements for that to continue excelling.
Ideally, AGI will be something humans want to interact with. For that to be possible, AGI will need to learn to recognize human emotions from body language and facial expressions and determine how to interact with them from there. AI is beginning to be able to identify emotions with facial scanning, but it is severely limited and often inaccurate. After all, humans can struggle with understanding emotions as well.
When AGI is able to understand emotions, it can then interact with humans in a way that will feel natural. It will not be like the generated responses from a personal assistant like Siri and Alexa. It will be capable of unique conversations and be able to determine what to say based on the emotions of the people around it.
Researchers are constantly working to expand the field of artificial intelligence and create AGI. These are some of the ways people are approaching creating artificial general intelligence:
Some people believe that being able to understand and use symbolic thought is the crux of what human intelligence is. These researchers are trying to create a way of teaching technology how to think like that. They believe that if they achieve this, AGI could exist.
This research area focuses on the human brain being a complex web of neurons firing electrical signals. They are trying to recreate that kind of system in the hopes that it will create AGI.
Some individuals consider human intelligence to be a hybrid system with many different pieces working together to create something greater than the sum of its parts. Researchers are, in a wide variety of ways, trying to mimic this to create human-like intelligence.
These researchers believe that if they can mathematically solve the theory of general intelligence, they can then create it. They’re working through purely theoretical models with the hopes of transferring it to the real world.
Some scientists believe that creating human intelligence is only possible when there is a physical body as well. They are working to integrate AI with physical representations of the human body to find breakthroughs to create AGI.
AGI is a distant goal for most researchers. Still, they know that developing certain technological capabilities will drive AGI research. These are some of the principal areas being continually improved:
Deep learning is an AI discipline that works on training neural networks to understand complex relationships between data. Researchers will build complicated webs and train AI to understand text, audio, images, video, and other information types.
Generative AI is a subset of deep learning where AI can produce content from learned knowledge. These models train using massive amounts of data to learn how to create content that resembles a human creation.
Natural language processing (NLP) is a branch that helps AI models learn to understand and generate human language. This is how AI tools like chatbots work.
Computer vision is AI’s ability to extract, analyze, and comprehend spatial information from visual data. It is being developed more to create self-driving cars that can notice obstacles and move the vehicle out of the way.
This field is working to create mechanics that can perform physical tasks. Creating this would allow AGI to physically manifest more fully and perform more services and tasks.
AGI does not exist yet, and there are currently some hurdles that make this field of research extremely challenging:
Right now, many AI systems cannot communicate with each other. There are conflicting interests that lead to a lack of data sharing between researchers and models. These gaps prevent the growth of the universality of AI.
Understanding how the human mind works is difficult enough for humans. Without full comprehension of what it means to be intelligent, it can be difficult to create technology with human-like capabilities. To make this happen, researchers will need to understand how intelligence works for humans first.
AI systems work in isolated and standalone environments. There aren’t currently protocols in place to regulate and enable sharing and collaboration. That does not align with the complex network of a social human environment that an AGI model would need to develop.
The ultimate business strategy is to see a return on investment in resources put toward developing and utilizing AI. However, AI returns are hard to measure because so much of the development happens in little pieces or stages, rather than a final product. That can make research difficult to align with strategy.
Too often, organizations lack plans or policies for AI and how it will be implemented in their business operations. The executive teams also rarely have a deep understanding of how the AI systems work and need to hire costly AI experts. That makes implementation costly and adds a roadblock to developing a complex AGI system.
Discussing AI systems that can think and act like human beings typically generates a lot of fear. People are worried about AGI taking over the world or that they will lose all sense of privacy. Most of these worries are emotions-based rather than rational. Still, there are some very real risks that come with creating an artificial general intelligence system. These are some of those fears:
People using AGI for nefarious purposes
Bias influencing the way AGI is trained, leading to a biased model
Lack of security for data and personal information
How to legislate this type of technology wisely
AGI will be designed to perform anything a human could, and that comes with many advantages:
Many day-to-day tasks are monotonous and get in the way of higher levels of productivity. AGI could perform more of these tasks and do so much quicker than humans. That would eliminate the need for everyone to perform these simple tasks. For example, fully functioning self-driving cars could remove the need for humans to drive. Instead, they can be transported from one place to another. AGI might also be able to stock shelves or even perform household chores.
AGI will be able to work long hours without breaks and with the same high output. That will allow it to complete tasks without losing concentration or getting distracted or tired. That might mean AGI could perform many tasks completely on its own, removing the need for human agents. In other instances, it may take more of a support role, such as helping surgeons during long-duration procedures.
Some jobs are highly dangerous to humans, and strong AI could be a way to reduce the need for human involvement in those environments. For example, mining can be harmful and underwater welding is incredibly dangerous. AGI robots who are as dexterous as humans could perform any necessary tasks in these places without risking human safety.
Some people hope AGI robots could be a solution to interstellar exploration. These machines would need fewer resources to travel and could provide much more in-depth research data to scientists than current technology can.
Weak AI is already able to detect disasters in certain contexts to help prevention. Strong AI will only improve upon that and provide invaluable help in the face of uncertain events. An AGI model may, for example, be able to predict a disaster and outline ideal evacuation routes.
AGI would raise some ethical concerns. The primary worry would be regulation of such powerful technology. Who would decide how AGI is used? Who is responsible when something goes wrong? These are crucial questions to answer when creating AGI, but it can be incredibly challenging to do so. Currently, there is nobody responsible for AI and creating ethical policies.
Another ethical obstacle is how to fairly train AGI technology. Experts predict that to create something like strong AI, the model will need to be trained on large amounts of data. That could open the possibility of human bias entering the AGI. If an AGI is trained only on certain data, it could perform with that bias later.
Knowing when someone has created AGI can be difficult to tell because intelligence cannot be measured. Researchers have proposed tests to determine if something is able to be considered true strong AI:
The Turing test is the original method of testing for AI from Alan Turing, a British researcher in the 1950s. He determined that true AGI will be able to hold a conversation with a human without being revealed as a machine. The human will believe they are conversing with another human because the machine can imitate humans so well.
The test involves three parties: a human guesser who must determine which of the other two is the human and which is the machine. If the interrogator fails to identify the machine, it would be considered AGI.
The co-founder of Apple, Steve Wozniak, proposed that it would be highly likely that a machine contained human levels of intelligence if it could follow the entire process of making coffee. The machine would be able to search for ingredients, locate them, gather them in one place, and perform the task.
In 2012, a researcher named Ben Goertzel proposed that a machine would have human intelligence if it could get admitted to college, take courses, and pass enough tests to get a degree like a human can. An AI robot from China has passed two math tests for admission, but the rest of the steps have yet to be completed by any weak AI.
Nils J. Nilsson proposed that an AI will have reached AGI levels when it can perform jobs at the same level as humans. Essentially, the AGI would be employable equally to human workers.
AGI may be something of the future, but there are still many advancements that researchers are making right now. These are some of the latest trends in general AI:
Natural language processing has grown significantly in the last few years with Open-AI’s ChatGPT. GPT-4 can handle 100 trillion parameters for comprehensive language processing. That indicates a very real possibility for the development of AI that can interact and engage with humans well.
The drive to increase the metaverse holds potential for creating a space for AGI development. AI can help build out the metaverse, and chatbots can help users feel at home in the virtual world.
Automation is already a reality in most businesses and industries. Hyperautomation is taking that to the next level by scaling automation prospects for organizations. That is an area where AI has played a significant role.
Experts are predicting more people will pay attention to bias that can come from AI. That will, in turn, result in more people who regulate and govern AI usage and training. More businesses are expected to hire AI officers and chief AI compliance experts.
Low-code or no-code systems provide a user-friendly interface that someone without coding knowledge can use and experiment with. This development could increase who is able to work on AI and increase the odds that breakthroughs will be made in the creation of AGI.
While people are afraid of AI replacing their jobs, the current trend is to implement AI into the workplace but keep it dependent on humans. This trend is only expected to become more entrenched in how people and AI work together, laying the foundation for AGI.
Chatbots are virtual assistants that can carry out certain tasks. For example, a chatbot can help users reset their passwords without requiring the help of a customer service representative. These agents have replaced business’s dependence on human employees and operational costs. AGI would only continue to increase this.
As people are more aware of biases and errors in AI, there has also been an increase in the discussion of AI ethics. A focus on the ethics of using AI will only continue to be a constant in conversations about AI and AGI.
Companies are already beginning to incorporate AI into hiring. While there are some bias issues, these practices can save HR teams significant amounts of time and companies substantial amounts of money. An example of this is using technology to analyze resumes and select a pool of candidates to interview.
AI in its limited form is already becoming an important part of how people live and how businesses operate. It can take care of monotonous tasks and speed up processes. That’s why the Now Platform® at ServiceNow includes generative AI, machine learning frameworks, natural language understanding, search and automation, and analytics and process mining.
The AI in this platform is designed to improve productivity and efficiency to elevate work experiences. Learn more about AI from ServiceNow to see how generative AI can be the next evolution of your business.
Quantum computing could speed up algorithms and allow models to digest the high volumes of data necessary to create AGI. There’s continuing research in how quantum computing could amplify AI.