According to a team of hundreds of AI researchers, the field is currently pursuing artificial general intelligence in the wrong way.
This insight is revealed in the 2025 presidential group of AI research futures in the Association for the Advancement of Artificial Intelligence (AAAI). 24 AI researchers have put together a lengthy report, with expertise ranging from AI infrastructure to the social aspects of artificial intelligence.
The report includes the main gains for each section, as well as a community opinion section in which respondents were asked about their thoughts about the section.
The section “AI Perception and Reality” hosted by MIT computer scientist Rodney Brooks mentions the characterization of the Gartner hype cycle, a five-phase cycle common to tech hype. The report noted that in November 2024, Gartner “estimated that the hype that generates AI has just passed its peak and is on a downward trend.” 79% of respondents in the Community Opinion section said that current public perceptions of AI capabilities are inconsistent with the reality of AI R&D, with 90% of whom said mismatch is hindering AI research, with 74% of whom “indications of AI research are driven by hype.”
Artificial General Intelligence (AGI) refers to human-level intelligence: the hypothetical intelligence of a machine, a machine that interprets information and learns from it like humans. Agi is the holy grail of the field, having an impact on the automation and efficiency of countless fields and disciplines. Consider any trivial tasks you don’t want to spend a lot of time doing, from planning your trip to submitting taxes. AGI can be deployed to relieve the burden of rote memorization, but it can also promote advancement in other areas, from transportation to education and technology.
A surprising majority (76% of 475 respondents) suggests that simply expanding current AI will not be enough to generate AGI.
“Overall, the response demonstrates a cautious and forward-looking approach: AI researchers prioritize safety, ethical governance, welfare sharing and progressive innovation, advocating for collaborative and responsible development rather than a competition for AGI,” the report reads.
Although the hype distorts the state of research and the current AI approach has not taken researchers to the best AGI path, the technology has made a leap forward.
“Five years ago, it was nearly impossible for us to have conversations – AI was limited to applications that could tolerate most of the wrong ones, such as product advice, or strictly limit the scope of knowledge areas such as classification of scientific images.” “Then, from a historic point of view, the AI general started working and attracted public attention through chatbots such as Chatgpt.”
The report reads that AI facts are “stay away from solving” and that the best LLMs answer only half of a set of questions correctly in the 2024 benchmark. However, new training methods can improve the robustness of these models, while new methods of organizing AI can further improve their performance.
“I believe the next phase of increasing credibility is the agents that will replace individual AI agents, with teams of partnering agents constantly checking others and trying to keep each other honest,” Kautz added. “Most of the public, as well as the scientific community (including the community of AI researchers) are within the quality of today's best AI systems; perception of AI lags behind the technology about a year or two.”
AI won't go anywhere; after all, the Gartner hype cycle won't end with “fading into forgetting”, but rather a “plateau of productivity.” Different areas of AI use cases have varying degrees of hype, but with all the calls from the private sector, from government officials to our own families, the report is refreshing, reminding people that AI researchers are thinking critically about the situation in their field. There is room for innovation and improvement from the way AI systems are built to how they are deployed in the world. Since we don't have time to go back for a while without AI, the only direction is to move forward.
