Artificial General Intelligence vs. AI

The definition of artificial intelligence (AI) might vary depending on who you ask. For example, if you consult the charter of OpenAI, artificial general intelligence (AGI) involves “highly autonomous systems that outperform humans at most economically valuable work” [1]. Another definition, from a Microsoft research team, calls AGI “systems that demonstrate broad capabilities of intelligence, including reasoning, planning, and the ability to learn from experience, and with these capabilities at or above human-level” [2]. 

Both definitions define intelligence—as economically valuable work or as reasoning, planning, and the ability to learn from experience—and each offers a measurement to compare against—human intelligence. But what is artificial intelligence capable of, and how do scientists know when we’ve reached artificial general intelligence (AGI)? 

In this article, we will closely examine a comparison of artificial general intelligence versus AI. You will learn about the implications of AGI for humanity, the technologies helping scientists progress toward AGI, and some of the challenges we need to overcome to achieve AGI. 

Read more: What Is Artificial Intelligence? Definition, Uses, and Types

Artificial general intelligence vs. AI 

Artificial general intelligence is a theoretical field of computer science that aims to create a computer capable of complex reasoning and problem-solving similar to or better than human intelligence. In the last few years, generative AI like ChatGPT entered the mainstream and inspired a lot of opinions about our proximity—as a society—to achieving true artificial general intelligence. Although we have much to be optimistic about regarding artificial intelligence, our current capabilities fall far short of true artificial superintelligence. 

The artificial intelligence employed by companies in almost every industry today is narrow—or weak—artificial intelligence. This technology may have advanced capabilities, but it remains focused on a specific, pre-identified task or goal. For example, a robot playing chess may create a challenging opponent, but without much training, it would make a terrible solitaire player.

Strong AI, on the other hand, would have self-awareness. It would be able to plan for its future and set its own goals. It could learn how to play solitaire the way anyone else learns, modeling human intelligence but potentially surpassing it. Strong AI, or artificial superintelligence, represents a pure form of artificial general intelligence that currently only exists in science fiction. 

Even advanced AI like the kind found in a Tesla can make mistakes when presented with something unique that a human could immediately understand. For example, a Tesla on autopilot cannot determine the appropriate response when confronted with a pedestrian carrying a stop sign. It understands both a pedestrian and a stop sign from its training but doesn’t understand that the two together mean it should still cause the car to stop. 

Read more: 4 Types of AI: Getting to Know Artificial Intelligence

Artificial general intelligence vs. generative AI

Despite the incredible capabilities of generative artificial intelligence, researchers and theorists are still debating whether it marks the beginning of our understanding of artificial general intelligence. Generative AI, like ChatGPT, is a form of artificial intelligence that uses deep learning to generate text or images that appear similar to, but not the same as, the vast amount of training data scientists provide. 

ChatGPT can do incredibly complex tasks in what seems like the blink of an eye. A team of Microsoft researchers found that the performance of ChatGPT-4—the newest version of the software—“could reasonably be viewed as an early (yet still incomplete) version of an artificial general intelligence (AGI) system” [2]. OpenAI tested ChatGPT-4 by simulating a bar exam. ChatGPT-4 passed with a score that would place it in the top 10 percent of test-takers [3]. 

Certainly, ChatGPT can outperform humans at many tasks, such as coding or generating informative text. Yet the technology still doesn’t qualify as artificial general intelligence, not because it isn’t impressive but because of how it does what it does. ChatGPT is a predictive language model built on a massive amount of training data. When it generates an answer, it’s merely predicting the correct information to generate based on what it has learned in the past. For all its remarkable ability, ChatGPT still parrots the bias of its training data and can be prone to misinformation—if given incorrect information to learn from. ChatGPT can’t plan for the future, can’t develop its own set of goals, and doesn’t exist in the real world with manual dexterity.

How close are we to developing artificial general intelligence?

One of the factors that make it difficult to predict how far away our technology is from developing true artificial general intelligence is the disagreement about what precisely AGI consists of and how to measure it.

In OpenAI’s definition of artificial general intelligence, AGI can perform most work that benefits the economy better than humans. But most of the work that humans do across all industries involves manual dexterity, such as preparing food or working in the construction industry. While these industries can both use artificial intelligence to operate more efficiently, we don’t have AGI advanced enough to completely replace a chef or a carpenter. 

The challenges of creating true artificial general intelligence pose problems for which we don’t yet have solutions. Some experts argue we are beginning to cross into AGI with our current capability, while others believe we are decades away from such a discovery if such a discovery is possible. 

Challenges of creating artificial general intelligence 

Scientists have several problems to overcome before we have the sort of artificial intelligence technology we imagine in science fiction films and books. Here are a few examples of what our current AI systems need to develop to achieve artificial general intelligence: 

• Sensory perception: Although sensor technology is allowing artificial intelligence more capability than ever before, particularly in home automation, our technology still has a long way to go before artificial intelligence has the capacity for sensory perception that humans do. For example, if you are talking on the phone with someone, you can create a mental image of their environment based on the background noise you hear. This is beyond the capacity of our current AI systems. 

• Manual dexterity: Many common jobs require manual dexterity, and it can be difficult to program a robot hand to operate the same way a human hand operates. Would you be comfortable allowing a pair of robot arms to wash your hair, give you a physical examination, or pat you down at the airport? We may need to improve our technology before we are comfortable allowing artificial intelligence to provide services that come close to our bodies.

• Social and emotional understanding: Humans can read subtle emotional and social context clues that our current artificial intelligence isn’t capable of understanding or considering emotions when generating a response. Until artificial intelligence can humanly understand the social context of our lives, it will miss out on much-needed context about human experience. 

What are the implications of AGI for humanity?

Artificial general intelligence can potentially change or eliminate many types of work that humans do, including computer programming, creating tasks, and medical research, to name a few examples. Artificial general intelligence will give us tools to increase our creativity, change education entirely, and offer personalized health care to each individual, regardless of economic status. 

Advanced general intelligence also comes with potential negative implications. For one thing, AGI could fuel an arms race worldwide where companies and governments scramble to hold the most advanced artificial intelligence technology. Many analysts say that artificial intelligence is progressing at an alarming speed already. If international pressure from our rivals spurs even faster development, the consequences of the systems could be unexpected and dangerous. For example, AGI could lead to major job displacement, accelerate economic inequality, or use methods and solutions that don't consider human safety, morality, and values. 

Outside of geopolitical peril, artificial intelligence could allow humans to use their natural brains less, which could lead to reduced cognitive abilities. We will need to use caution and regulation to ensure that scientists develop AGI responsibly. 

Learn more with Coursera.

To take the next step and learn more about artificial intelligence, enroll today in Introduction to Artificial Intelligence (AI) offered by IBM on Coursera. This course takes approximately eight hours to complete and will help you learn skills in machine learning, deep learning, and artificial intelligence. Upon completion, gain a shareable Professional Certificate to include in your resume, CV, or LinkedIn profile.