According to a panel of hundreds of artificial intelligence researchers, the field is looking for artificial general intelligence in the wrong way.
This vision was revealed in the Association for the Advance of Artificial Intelligence (AAAI) in the 2025 presidential panel on the future of AI investigation. He Long report They were gathered by 24 researchers from AI whose experience covers from the state of AI infrastructure to the social aspects of artificial intelligence.
The report included a main conclusion for each section, as well as an opinion section of the community where the respondents were asked their own thoughts about the section.
The section on “perception of the versus reality”, chaired by the MIT computer scientist, Rodney Brooks, referred to the characterization of the Gartner Bombo cycle, a common five -stage cycle for technological exaggeration. In November 2024, Gartner “estimated that the exaggeration for the generative AI had just passed its peak and was on the end,” the report said. 79% of respondents in the community’s opinion section declared that current public perceptions of AI capacities do not coincide with the reality of the research and development of AI, and 90% say that the mismatch is hindering the investigation of AI: 74% of that number say that “the directions of the research of AI are driven by exaggeration.”
Artificial General Intelligence (AGI) It refers to intelligence at the human level: the hypothetical intelligence of a machine that interprets information and learns from it as a human being. AGI is a holy grail of the field, with implications for automation and efficiency in innumerable fields and disciplines. Consider any servile task that does not want to spend much time, from planning a trip to presenting your taxes. AGI could be implemented to relieve the burden of memory tasks, but also catalyze progress in other fields, from transport to education and technology.
The surprising majority, 76% of the 475 respondents, said that simply expanding the current AI approaches will not be enough to produce AGI.
“In general, the responses indicate a cautious approach but forward: the researchers of AI prioritize security, ethical governance, the exchange of benefits and gradual innovation, advocating for collaborative and responsible development instead of a race towards AGI,” the report wrote.
Although the exaggeration distorts the state of the investigation, and the current approaches for AI do not put researchers on the most optimal path towards AGI, technology has come out and limits.
“Five years ago, we could hardly have had this conversation: the AI was limited to applications in which a high percentage of errors could be tolerated, such as the recommendation of the product, or when the domain of knowledge was strictly circumscribed, such as the classification of scientific images,” said Henry Kautz, computer scientist at the University of Virginia and president of the section of the report on the report on the report on the report on the report on the report on the report on the report on the report on the report on the report on the report on the report on the report on the report on the report on the report on the report on the report on the report on the report on the report on the report on the report on the report on the report on the report on the report of the to an email to an email. “Then, suddenly in historical terms, the general Ia began to work and reached public attention through chatbots such as Chatgpt.”
The factual of AI is “far from being resolved,” reads the report, and the best LLMS only answered approximately half of a set of questions correctly in a reference test of 2024. But the new training methods can improve the robustness of those models, and the new ways of organizing AI can further improve their performance.
“I think that the next stage to improve reliability will be the replacement of individual AI agents with cooperating equipment of agents that continuously verify others and try to stay honest,” Kautz added. “The majority of the general public, as well as the scientific community, including the community of AI researchers, underestimates the quality of the best AI systems today; the perception of AI is left behind about one year or two behind technology.”
Ai does not go anywhere; After all, the Gartner Bombo cycle does not end with “fading in oblivion”, but the “productivity plateau”. Different cases of use of AI have different levels of exaggeration, but with all the cry on AI, from the private sector, of government officials, devils, of our own families, the report is a refreshing reminder that Ia researchers are thinking very critically about the state of their field. From the way in which AI systems are built to the ways in which they are implemented in the world, there is space for innovation and improvement. As we are not returning to a moment without ia, the only direction is ahead.