As per a survey conducted among artificial intelligence (AI) experts, there is a prevailing belief that advancements in AI are rapidly progressing and hold considerable potential benefits for humanity. However, there exists a significant level of concern regarding the potential negative implications associated with the pursuit of highly advanced AI technologies.
The survey, which is yet to undergo peer review, engaged 2,778 AI researchers in discussions regarding various aspects of AI, including the impact of high-level machine intelligence (HLMI) on humanity. Participants were asked to assess the likelihood of scenarios such as “human extinction or severe disempowerment of the human species,” with the mean forecast indicating a 5% probability. Interestingly, when considering the same scenario within a 100-year timeframe, the forecast remained consistent.
The researchers noted that between 41.2% and 51.4% of respondents expressed a belief in a greater than 10% chance of such catastrophic outcomes. This sentiment aligns closely with the 38% of participants who envisioned at least a 10% possibility of extremely adverse consequences, including human extinction, when questioned about the impact of High-Level Machine Intelligence on humans.
Comparisons were drawn to a similar survey conducted in 2022, revealing a general consensus among participants that advancements in AI, such as job automation and the creation of bestselling novels, would materialize sooner than previously anticipated. Projections for AI achieving the feat of producing a bestseller novel were notably advanced, with expectations set before 2030, potentially influenced by the excitement surrounding recent developments in chatbot technology. Similar adjustments were observed in predictions related to language-based tasks, albeit with tasks like autonomous vehicle operation and participation in competitive marathons expected to occur later, likely in the 2030s.
The researchers highlighted a shift towards earlier expectations among participants, with a considerable number predicting accelerated progress towards various AI-related milestones. Notably, forecasts for achieving human-level performance were estimated to occur 13 to 48 years sooner on average compared to the previous survey. Additionally, 21 out of 32 short-term milestones are now anticipated to be reached sooner than expected.
In conclusion, the survey reflected diverse perspectives on the cultural implications of advanced AI technologies, with individuals weighing both positive and negative outcomes. While concerns regarding extinction-level risks serve as a stark warning of the potential hazards associated with AI development, optimistic scenarios underscore AI’s capacity to revolutionize work and lifestyle dynamics.
