Increased Skepticism Among Scientists Regarding the Development of AGI: New Research Findings

A new study reveals that artificial intelligence researchers are largely doubtful that current AI strategies will lead to the development of Artificial General Intelligence (AGI), despite ongoing advancements in the technology.

According to the [AAAI study on the future of AI research](https://aaai.org/wp-content/uploads/2025/03/AAAI-2025-PresPanel-Report-Digital-3.7.25.pdf), over three-quarters of respondents believe that scaling up existing AI systems is unlikely to yield AGI. The findings indicate that 76% of participants rated this possibility as «unlikely» or «highly unlikely.»

The research highlights a strong consensus about the importance of symbolic reasoning, with more than 60% of AI researchers asserting that any system that comes close to human-like cognitive processing must be at least 50% symbolic. However, there remains a lack of formal definitions and agreed-upon testing criteria for AGI in this domain.

Despite these reservations, the majority of researchers advocate for the continuation of AI development. Around 70% oppose halting research until safety mechanisms are in place, while 82% believe that AGI systems developed by private entities should be subject to government oversight.

Researchers also see significant promise in specialized AI systems. They point to [Google DeepMind’s Alphafold](https://the-decoder.com/google-deepmind-releases-alphafold-3-for-scientific-use/) as an example of how AI can enhance scientific progress as an expert system in a particular field, which is one of the key advantages of achieving AGI.

In spite of recent breakthroughs in [testing algorithms](https://the-decoder.com/studie-zeigt-test-time-compute-scaling-ist-der-weg-zu-besseren-ki-systemen/) for [reasoning models](https://the-decoder.com/studie-reinforcement-learning-via-self-play-ist-der-schluessel-zum-reasoning-in-sprachmodellen/), the researchers identified fundamental flaws within current AI architectures. Systems still struggle with long-term planning, cannot learn continuously, and lack the structured episodic memory that humans possess.

Furthermore, researchers indicated notable gaps in causal reasoning and real-world interaction. While advancements in speech and image processing have been impressive, systems still lack a [deeper understanding of physical reality](https://the-decoder.com/ai-video-generators-like-openais-sora-dont-grasp-basic-physics-study-finds/).

These results were compiled by the «AAAI Presidential Panel 2025 on the Future of AI Research,» led by AAAI President Francesca Rossi. The study brought together 24 expert AI researchers between the summer of 2024 and spring of 2025 to explore 17 different AI research topics, including scientific discovery and AGI. An additional survey of 475 participants, primarily from academic backgrounds (67%) and North America (53%), provided further insights into the future direction of the field.

[Source](https://the-decoder.com/most-ai-researchers-are-skeptical-about-language-models-achieving-agi/)