A Survey of 2,778 Researchers Highlights Fragmentation and Accelerated Progress in AI

By The Engineer

January 5, 2024

The "2023 Expert Survey on Progress in AI," which polled 2,778 researchers, has revealed a deeply fragmented scientific community with no clear consensus on the risks and opportunities of AI. However, one thing is clear: AI development is moving faster than previously anticipated.

Key Takeaways:

• Accelerated Development: The pace of AI progress will continue to accelerate.
• Diverse Opinions: There's no consensus on whether AI development should be slowed or sped up.
• Early Milestones: Several significant milestones are expected to be reached by 2028, much earlier than previously thought.

Accelerated AI Development

The survey found that the majority of researchers believe AI development will continue to accelerate. Notably, there's a 50% probability that AI systems will achieve several key milestones by 2028:

• Autonomous Web Development: Creating a payment processing website from scratch.
• Indistinguishable Music Creation: Producing a song that can't be differentiated from one by a well-known musician.
• Comprehensive Language Modeling: Downloading and refining a comprehensive language model.

These milestones are expected to occur much earlier than previously estimated. For instance, the survey predicts that an AI will write a fictional New York Times bestseller around 2030, which was previously estimated for 2038.

Diverse Opinions on Development Pace

When it comes to the pace of AI development, opinions are divided:

• Support for Faster Development: About 15.6% of respondents support a much faster development pace.
• Support for Slower Development: Only 4.9% advocate for a much slower pace.
• Neutral Stance: A significant 27% believe the current pace is appropriate.

This fragmentation highlights the lack of consensus on how to manage AI's rapid advancement. The "much faster" group is three times larger than the "much slower" group, indicating a strong push for accelerating progress despite concerns about risks.

High-Level Machine Intelligence (HLMI) and Full Automation of Labor (FAOL)

The survey also delved into more ambitious milestones:

• High-Level Machine Intelligence (HLMI): The probability that machines will outperform humans in all possible tasks without external help is estimated at 10% by 2027 and 50% by 2047.
• Full Automation of Labor (FAOL): Similar to HLMI, the forecast for FAOL points to a much earlier occurrence than previously expected.

These estimates are 13 years ahead of predictions from a similar survey conducted just one year ago. This shift underscores the rapid pace of technological progress and the growing belief in AI's potential to achieve these milestones sooner rather than later.

Existential Fears and Risk Mitigation

Despite the fragmented views, existential fears about AI remain a significant concern. However, these fears are becoming more moderate:

• Moderate Concerns: While many researchers acknowledge the risks associated with advanced AI, there is a growing belief that these risks can be managed through careful development and regulation.

This moderation in existential fears could be attributed to increased efforts in risk mitigation and ethical considerations within the AI community. However, the lack of consensus on how to address these risks remains a challenge.

Conclusion

The "2023 Expert Survey on Progress in AI" paints a picture of a rapidly evolving field with significant fragmentation among researchers. While there is no clear consensus on the pace and direction of AI development, the overall trend points to accelerated progress and earlier achievement of key milestones. As the community continues to grapple with these challenges, it's crucial to balance innovation with responsible governance.