Most AI experts say chasing AGI with more compute is a losing strategy

Most AI experts say chasing AGI with more compute is a losing strategy

  • 23.03.2025 16:47
  • techspot.com
  • Keywords: AGI, AI

Most AI researchers doubt that increasing computing power alone will achieve artificial general intelligence (AGI). A survey of 475 experts found 76% believe scaling current models is unlikely to succeed. While tech companies continue investing heavily in AI infrastructure, many researchers argue the returns are diminishing and alternative approaches are needed.

Alphabet ReportsMSFTsentiment_dissatisfiedAMZNsentiment_dissatisfied
open_in_newOriginal article

Estimated market influence

Microsoft

Microsoft

Negativesentiment_dissatisfied
Analyst rating: Strong buy

Microsoft is investing in nuclear power deals to fuel their data centers, which may have negative environmental impacts.

Google

Negativesentiment_dissatisfied
Analyst rating: N/A

Google's approach of scaling up AI without understanding the fundamentals could lead to diminishing returns and potential environmental issues.

Amazon

Amazon

Negativesentiment_dissatisfied
Analyst rating: Strong buy

Amazon is also investing in nuclear power deals, which may have negative environmental impacts on their data centers.

Context

Analysis of Business Insights and Market Implications

Key Findings from the Survey

  • 76% of AI researchers believe adding more compute power and data to current models is "unlikely" or "very unlikely" to achieve AGI.
  • Diminishing returns on investments in scaling: Despite massive spending, performance improvements have plateaued.

Investment Trends

  • $56 billion: Venture capital funding for generative AI in 2023 (TechCrunch report).
  • $626 billion: Semiconductor industry revenue in 2024.
  • Massive energy demands: Companies like Microsoft, Google, and Amazon are entering nuclear power deals to fuel data centers.

Performance Metrics

  • Plateaued improvements: OpenAI's latest models show only marginal gains over predecessors.
  • Energy costs: Scaling up models increases power requirements significantly.

Alternative Approaches

  • "Test-time compute": OpenAI experiments show performance boosts without scaling, but experts warn it’s unlikely to be a "silver bullet."
  • Efficiency focus: Researchers explore cheaper, more efficient methods to reduce reliance on brute-force computing.

Expert Opinions and Market Shifts

  • Skepticism among leaders: Stuart Russell (UC Berkeley) criticizes the lack of theoretical understanding in scaling efforts.
  • Shift in priorities: 77% of researchers prioritize risk-benefit profiles over AGI development, signaling a move toward ethical considerations.

Competitive Dynamics

  • Tech optimism vs. researcher skepticism: Google’s Sundar Pichai remains bullish on scaling, while the broader AI community questions its efficacy.
  • Regulatory concerns: 82% believe AGI should be publicly owned if developed by private entities to mitigate global risks.

Long-Term Effects and Market Implications

  • Diminished ROI for scaling: Companies may need to pivot toward more efficient technologies or alternative strategies.
  • Potential regulatory shifts: Public ownership of AGI could lead to new regulations on AI development and deployment.
  • Energy cost optimization: Focus on reducing power consumption could reshape the semiconductor and data center industries.

Strategic Considerations

  • Balancing growth and ethics: Businesses must weigh technical advancements against ethical and safety concerns.
  • Investment in theoretical research: Companies may need to allocate more resources to understanding AI fundamentals rather than just scaling.