과학AI, 우리가 만들되 기준은 ‘세계 수준’이어야 [왜냐면]

A recent survey conducted by a government-affiliated research institution revealed that 89.5% of 764 researchers utilized artificial intelligence (AI) in their work. Of these, 85.3% employed services like ChatGPT, while only 3.9% used specialized tools for scientific research and 18.1% incorporated AI into experimental design. The research suggests that while we have gained "efficient AI," we have not yet achieved "discovery AI."

The key to this gap lies in the fact that we cannot create discovery AI using only off-the-shelf universal tools; it must be developed using our own data and tailored to our research environment. Other countries have already moved on to the next stage, with companies like OpenAI and Anthropic revealing breakthroughs, while Google DeepMind is distributing AI that can formulate and verify hypotheses.

The U.S. government has also invested over $5 billion in a "Genesis Mission" to bring together national research institutions and AI companies. To avoid being left behind, countries must address four structural challenges: secure research-focused AI environments, pooled national computing resources, specialized AI models and research data for our own sciences, and practical training in the field.

While 80% of researchers who have experienced practical training actively use it, those without such experience are almost twice as likely to disengage. However, the debate over proprietary AI development must be avoided. Researchers emphasize that replacing foreign-developed AI with domestic alternatives is not sufficient; we need to create 'world-class' AI.

Additionally, AI can serve as a lifeline for regions struggling with talent shortages, narrowing the gap between urban and rural research environments. The next challenge is whether countries that use AI for faster publication will outpace those that use AI to discover new scientific knowledge. The time for Korean science to move beyond being a consumer is limited.