Most AI Researchers: The Technology Industry Is Investing Billions in a Dead End
A recent survey conducted by the Association for the Advancement of Artificial Intelligence (AAAI) among 475 AI researchers reveals a surprising consensus: most experts believe that the technology industry's dominant strategy – investing billions in scaling current AI models and infrastructure – is unlikely to lead to the achievement of artificial general intelligence (AGI).
Survey Findings
A staggering 76% of the researchers surveyed said it was "unlikely" or "very unlikely" that simply scaling existing generative models and data centers would lead to AGI, which is widely considered the ultimate goal of many entities in the field. "Massive investments in scaling that have not been accompanied by comparable efforts to understand what is actually happening have always seemed misplaced to me," said Stuart Russell of the University of California, Berkeley.
Experts such as Stuart Russell note that approximately a year ago, it became clear to many people in the field that the gains from conventional scaling had peaked. Despite massive investments, further increases in model size and computing power are producing diminishing returns. The industry invested more than $56 billion in generative AI funding during 2024, while companies such as Microsoft are planning to spend $80 billion on AI infrastructure in 2025 alone. Yet the latest versions of large language models show significantly smaller improvements than previous iterations.
The Need for Alternative Approaches
Some companies are exploring more efficient methods – for example, OpenAI uses "test-time compute," in which AI spends more time reasoning before making a decision – but these are not considered a cure-all. Fundamental innovations beyond the brute-force scaling method will likely be necessary for a true breakthrough. There is also growing concern about energy consumption and environmental impact, as technology giants sign contracts for dedicated nuclear power plants just to operate their expanding fleets of data centers.
Recent developments confirm this skeptical assessment. Reports suggest that the latest OpenAI GPT model showed little or no improvement over its predecessor. Google CEO Sundar Pichai publicly stated that the easy gains from simple scaling are "over," even as some leaders continue to push for larger models regardless of the results. Microsoft has begun scaling back planned data center expansions amid economic uncertainty and doubts about the continued returns on such investments.
The Necessity of New Approaches
The AAAI Presidential Panel on the Future of AI Research emphasizes that fundamental innovations in basic research will be needed to achieve meaningful progress. The AAAI panel discussion identified several key areas requiring greater attention, including:
- New approaches to learning and reasoning that go beyond the current statistical learning paradigm.
- More efficient and energy-saving AI architectures.
- A deeper understanding of how current models work and how we can improve them.
There is now broad agreement among leading researchers that simply pouring more money into larger hardware and bigger datasets will not deliver AGI or sustainable transformative progress. The industry faces growing pressure – and an opportunity – to pursue new algorithmic innovations rather than relying exclusively on scaling-based approaches. As one researcher cited in the AAAI survey noted: "We need to return to the basics and think through what intelligence really is and how we can model it in ways that are not based solely on accumulating data and computing power."
