Artificial Intelligence Research: A New Direction

The field of artificial intelligence is currently pursuing artificial general intelligence (AGI) in the wrong way, according to a panel of hundreds of AI researchers. This insight was revealed at the Association for the Advancement of Artificial Intelligence (AAAI)’s 2025 Presidential Panel on the Future of AI Research. A comprehensive report, compiled by 24 AI researchers with diverse expertise, provides a detailed analysis of the current state of AI research.

The Report’s Findings

The report highlights a significant mismatch between public perceptions of AI capabilities and the reality of AI research and development. According to the report, 79% of respondents believe that current public perceptions do not match the reality of AI research, with 90% stating that this mismatch is hindering AI research. Furthermore, 74% of respondents believe that the directions of AI research are driven by hype. The report also notes that the Gartner Hype Cycle, a five-stage cycle common for technology hype, has characterized the current state of AI hype, with Generative AI having just passed its peak and being on the downswing.

The Pursuit of Artificial General Intelligence

Artificial general intelligence (AGI) refers to human-level intelligence, where a machine can interpret information and learn from it as a human being would. AGI is a long-sought goal in the field, with implications for automation and efficiency across various fields and disciplines. However, a surprising majority (76% of 475 respondents) believe that simply scaling up current approaches to AI will not be sufficient to yield AGI. Instead, researchers prioritize safety, ethical governance, benefit-sharing, and gradual innovation, advocating for collaborative and responsible development rather than a race toward AGI.

Progress and Challenges

Despite the hype surrounding AI, the technology has made significant progress in recent years. According to Henry Kautz, a computer scientist at the University of Virginia, "Five years ago, we could hardly have been having this conversation – AI was limited to applications where a high percentage of errors could be tolerated." However, AI factuality is still "far from solved," with the best language models only answering about half of a set of questions correctly in a 2024 benchmark test. New training methods and ways of organizing AI can improve the robustness of these models, and new approaches, such as cooperating teams of agents, can further enhance their performance.

The Future of AI Research

The report is a timely reminder that AI researchers are thinking critically about the state of their field. With the Gartner Hype Cycle predicting a "plateau of productivity" rather than "fade into oblivion," AI is here to stay. As Kautz notes, "Most 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 lags about a year or two behind the technology." As we move forward, there is room for innovation and improvement in AI research, from the way AI systems are built to the ways they are deployed in the world.