Source: Quantum

In recent years, artificial intelligence technology has made amazing progress and has shown strong potential in many fields. However, the real goal of artificial intelligence research, general artificial intelligence (i.e., intelligent systems with human-like cognitive abilities), is still a controversial topic. As the public has high expectations for the development of artificial intelligence, researchers have also questioned the current research direction.

So, are we moving in the right direction? What is the current situation of artificial intelligence research? According to Gizmodo, the recent research results of a team of hundreds of artificial intelligence researchers show that humans are currently taking the wrong direction in the pursuit of general artificial intelligence. What is going on?

The results of a team of hundreds of AI researchers show that the field is currently pursuing universal AI in the wrong way.

This insight was revealed at the 2025 "Artificial Intelligence Research Future" presidential panel meeting organized by the American Association for the Promotion of Artificial Intelligence. The lengthy report was co-written by 24 AI researchers whose expertise covers multiple aspects, including the status quo of AI infrastructure and the social impact of AI.

Each section of the report contains a main conclusion and has a “community opinion” section that surveys respondents’ perceptions of the section.

The gap between AI cognition and reality

The “Artificial Intelligence Cognition and Reality” section was hosted by MIT computer scientist Rodney Brooks, and cited the Gartner Hype Cycle model, which describes the five stages of technology hype. The report noted that in November 2024, Gartner estimated that “the hype of generative artificial intelligence has just reached its peak and is in a downward phase.”

In the “Community Opinion” survey, 79% of respondents believed that the public’s perception of artificial intelligence capabilities did not match the reality of AI research and development. And 90% said this mismatch hindered AI research—74% of whom believed that “the direction of AI research is driven by hype.”

"I quote Gartner's hype cycle model because it has been widely used over the years and is applicable to hype in multiple fields and the disappointment that comes with it," Brooks said in an email. "So the existence of this model and its accuracy in multiple fields tells us that the hype in the current development of AI should be taken with caution."

"I think the public has too much discussion of AI in the public's belief in the accuracy of hype," he added.

Existing methods are not enough to achieve universal artificial intelligence

General artificial intelligence (AGI) refers to machines with human-level intelligence—that is, they can understand it like humans.Impersonational intelligence that information and learn from it. General artificial intelligence is the "holy grail" in this field, with its impact on automation and efficiency across multiple industries and disciplines. For example, any trivial tasks you don’t want to spend time doing, from planning a trip to filing taxes, general AI can help ease the burden. In addition, general artificial intelligence may also promote advancement in areas such as transportation, education and technology.

However, the survey results show that the vast majority of researchers believe that the current approach is not sufficient to achieve universal artificial intelligence. Of the 475 respondents, 76% believed that scaling up existing AI methods alone was not enough to generate universal AI.

“In general, these responses demonstrate a cautious and forward approach: AI researchers prioritize security, ethical governance, benefit sharing and progressive innovation, advocating collaborative and responsible development rather than competing to develop universal AI,” the report reads.

Although the hype distorts the true state of AI research and the current approach has not put researchers on the best path to universal AI, the technology has made rapid progress.

Advances and Challenges in Artificial Intelligence Research

"If it were five years ago, we would have had almost no such discussion—Artificial intelligence was used primarily for applications with high fault tolerance (such as product recommendations), or applications where knowledge areas are strictly limited (such as classification of scientific images)," Henry Coots, a computer scientist at the University of Virginia and head of the report's "Factuality and Credibility" section, explained in an email. “However, in a relatively sudden time on a historical scale, general AI began to play a role and entered the public eye through chatbots like ChatGPT.”

However, the factual problems of AI are still “far from being solved.” The report notes that in a 2024 benchmark, even the most advanced large language models had a correct answer rate of only about 50%. However, new training methods can improve the robustness of these models, and new methods of organizing artificial intelligence can also help further improve their performance.

"I believe that the next phase in improving AI credibility will be to replace a single AI agent with a team of collaborative AI agents, allowing them to constantly fact-check each other to ensure each other's honesty," Coots added. “Most public and the scientific community, including the group of AI researchers, underestimate the quality of today’s most advanced AI systems – people’s perception of AI is often one or two years behind technology development.”

The future direction of AI

Different AI application scenarios are in different hype stages, but amid the hustle and bustle of the field of AI – whether from private enterprises, officials, or even members of our own family – this report provides a sober reminder that AI researchers are taking a thoughtful look at their field. From the way artificial intelligence systems are built to how they are applied,There is room for improvement and innovation.

Since we will not return to the era without artificial intelligence, the only direction is to move forward.