In a groundbreaking survey, nearly 2,800 AI researchers have shared their insights on the trajectory of AI progress, its potential milestones, and the risks and societal impacts associated with advanced AI systems. The survey, conducted by AI Impacts, captures the diverse views of experts who have published in top AI venues, offering a detailed glimpse into the expectations and concerns of the field’s thought leaders.
Key Findings: Rapid Progress and Uncertain Outcomes
The survey’s findings underscore a mixed outlook for AI’s future. The researchers estimate a 50% chance that AI will achieve significant milestones by 2028, including autonomously creating complex websites, producing music indistinguishable from popular artists, and fine-tuning large language models without human intervention. Looking further ahead, a majority predict that unaided machines could outperform humans in all tasks by 2047, a timeline that has advanced by 13 years compared to a similar survey conducted just one year prior.
However, when it comes to the full automation of all human jobs, the projections are more cautious, with estimates placing a 50% chance of achieving this by 2116. This slower pace reflects uncertainties around integrating AI into occupations requiring complex physical tasks or nuanced human judgment.
The Spectrum of Optimism and Pessimism
While 68.3% of respondents are optimistic about the long-term benefits of AI, almost half of these optimists still assign at least a 5% probability to catastrophic outcomes, including human extinction. This anxiety is not unfounded; between 38% and 51% of researchers believe there is at least a 10% chance that advanced AI could lead to extinction-level scenarios. Concerns range from the spread of misinformation and authoritarian use of AI to exacerbated social inequality and potential loss of control over AI systems.
Forecasts on Key Milestones: What AI Might Achieve Next
Researchers provided predictions for 39 specific AI milestones, with most expected to be feasible within the next decade. These include economically significant tasks such as autonomous website creation and creative achievements like composing original music comparable to that of major artists. In robotics, expectations for AI-driven advancements remain high, with tasks like folding laundry or navigating complex environments forecasted to become increasingly manageable for AI systems.
Despite these optimistic timelines, a few tasks, such as installing electrical wiring in homes or solving unsolved problems in mathematics, are expected to take longer. Predictions for these milestones extend well beyond the next decade, reflecting the challenges that physical tasks and complex scientific reasoning still pose for AI.
High-Level Machine Intelligence vs. Full Automation of Labor
The concept of High-Level Machine Intelligence (HLMI)—where AI can outperform humans in every task—garnered varied timelines among respondents. The 2023 survey results indicate a 50% likelihood of HLMI being achieved by 2047. Interestingly, the timeline for full automation of all occupations (FAOL) lags significantly behind, with a 50% chance projected by 2116. This gap suggests that while AI might soon excel in individual tasks, the broader challenge of automating entire jobs remains formidable.
Potential for an Intelligence Explosion
The survey also explored the potential for an “intelligence explosion,” where AI systems, once able to self-improve, could trigger a rapid acceleration of technological progress. A significant number of researchers (around 60%) believe there’s at least an even chance that HLMI could lead to such an outcome within 30 years of its achievement. However, opinions diverge on whether this would manifest within a brief window of just a few years, reflecting ongoing debates about the pace and manageability of AI advancements.
Societal Impacts: Concerns Over Misinformation, Inequality, and Control
The survey highlights deep concerns about how AI might impact society. Scenarios such as AI-enabled misinformation, population control by authoritarian regimes, and worsening economic inequality were identified as top worries. Over 70% of respondents felt that substantial or extreme concern is warranted for these potential futures.
Despite recognizing the transformative potential of AI, researchers remain vigilant about the risks. Most believe that research into AI safety and alignment should be prioritized more than it currently is, emphasizing the importance of developing AI systems that can be trusted to act in humanity’s best interests.
Conclusion: A Complex, Uncertain Future for AI
The future of AI is marked by rapid progress, but also profound uncertainty. As AI continues to evolve, the insights of researchers in the field provide critical guidance. Their forecasts suggest that while AI holds the promise of revolutionizing many aspects of society, careful consideration and proactive measures are essential to navigate the potential dangers that accompany this transformative technology.
The full report is available from AI Impacts, providing a comprehensive overview of expert opinions on AI progress and the road ahead.