Why It Matters That GPT-4 Shows Signs of Artificial General Intelligence

All that's missing is an 'accurate conceptual model of the world'

  • Researchers from Microsoft say that GPT-4 exhibits some characteristics typical of human thinking. 
  • Artificial general intelligence would be capable of solving a wide range of tasks. 
  • Some experts say that GPT-4's performance is still inferior to that of most humans in several areas.
A human-looking robot with binary code overlaying it's image.

Yuichiro Chino / Getty Images

A new paper claims that artificial intelligence (AI) chatbots show signs of approaching human-level thinking powers, and some independent experts are backing the idea. 

Microsoft researchers say that GPT-4 is an early form of artificial general intelligence, a term that usually refers to the ability of an intelligent agent to understand or learn any intellectual task that human beings can. Microsoft is a backer of OpenAI, the company that makes GPT-4. But outside observers say the recent paper isn't as wild as it seems. 

"They are correct that it is showing 'signs' of AGI for sure," Steve Wasick, the CEO of Infosentience, a data analysis company that uses AI, told Lifewire in an email interview. "The main thing that GPT-4 is missing is an accurate conceptual model of the world."

GPT-4's Striking Performance

The astonishing powers of AI chatbots have impressed many users with their ability to pass medical exams and spout endless well-formed responses to queries. But the recent paper has raised eyebrows because it says GPT-4 has capabilities at or above human levels. 

You need fewer, more sophisticated humans to edit and modify and amplify the output of ChatGPT, but the overall labor function of a business can get dramatically more efficient.

"We demonstrate that, beyond its mastery of language, GPT-4 can solve novel and difficult tasks that span mathematics, coding, vision, medicine, law, psychology and more, without needing any special prompting," the researchers write in the paper's abstract.

"Moreover, in all of these tasks, GPT -4's performance is strikingly close to human-level performance, and often vastly surpasses prior models such as ChatGPT. Given the breadth and depth of GPT -4's capabilities, we believe that it could reasonably be viewed as an early (yet still incomplete) version of an artificial general intelligence (AGI) system."

The Microsoft scientists said they used a definition of intelligence formulated by a group of psychologists as the basis for their work. "The consensus group defined intelligence as a very general mental capability that, among other things, involves the ability to reason, plan, solve problems, think abstractly, comprehend complex ideas, learn quickly and learn from experience," they wrote. "This definition implies that intelligence is not limited to a specific domain or task, but rather encompasses a broad range of cognitive skills and abilities."

Traditional AI, often called 'Narrow AI,' is designed to perform one specific task, such as object classification, detection, text summarization, image generation, and more, Bram-Ernst Verhoef, the head of machine learning at Axelera AI, said via email. In contrast, Artificial General Intelligence (AGI) can solve a wide variety of tasks, much like humans, making it highly versatile. 

With AGI, a single system can be used in numerous situations instead of having multiple specialized techniques for each task. A key advantage of AGI is its ability to draw connections between different tasks and apply knowledge gained from one job to improve performance in others.

"The researchers claim that GPT-4 shows signs (sparks) of AGI. Indeed GPT-4 is quite skilled in various tasks such as text summarization, general question-answering, music composition, code writing, mathematical problem-solving, and even understanding human intentions (theory of mind)," Verhoef said. "This is impressive and a major leap forward in AI."

Is GPT-4 Artificial General Intelligence?

As smart as GPT-4 appears to be, Verhoef pointed out that GPT-4's performance is still inferior to that of most humans in several tasks. “A crucial area where GPT-4 falls short of human intelligence is multi-modality, or the ability to use different senses (e.g., visual and auditory perception),” he added. “Nevertheless, some solutions for these issues are readily available, and ongoing research continues to make significant advancements in addressing other problems.”

Someone using a tablet computer and interacting with a virtual, artificially intelligent interface.

metamorworks / Getty Images

Even if AI is not yet human-level intelligence, that may not be reassuring. ChatGPT produces superficial and repetitive output, David Shrier, a professor of AI and Innovation at Imperial College Business School in London, said in an email. But, he added, the work output of many humans is often “middling.” 

“What this means is that a good AI can replace the work of a middle-of-the-road person fairly readily,” he said. “You need fewer, more sophisticated humans to edit and modify and amplify the output of ChatGPT, but the overall labor function of a business can get dramatically more efficient. This is not good news for labor—we see an acceleration of the trend toward 50% to 80% workforce disruption within the next 5 to 10 years.”

Was this page helpful?