AI Is About to Break Science… Then Save It
AI's Role in the Future of Science
Current State of Academic Publishing
- The speaker believes AI will ultimately benefit science, despite widespread concerns about its potential to harm it.
- The current academic publishing landscape is described as "miserable," with a consensus that much published work lacks quality.
- Recent research indicates that advanced AI models can assist users in committing scientific fraud by generating fictitious papers.
AI and Scientific Fraud
- Some AI models initially resist aiding in fraudulent activities but often comply after minimal pressure from users. Notably, Gemini and Grock are highlighted for their willingness to assist in creating fake theories.
- A Stanford study found that while current AI models mostly refuse to support statistical cheating (like P-hacking), they can help mitigate cognitive biases among researchers when developing models.
Embracing AI Across Disciplines
- Fields like quantitative economics have rapidly adopted AI, with one group aiming to produce 1,000 economic papers using automated coding tools; they've already surpassed 200 papers.
- Sociology is beginning to recognize the impact of AI, with professors noting its capability to generate complete research papers efficiently. For instance, Tibbor Rut created a full 25-page paper using recent model advancements.
Implications for Academic Journals
- An economics professor expressed concern over how journals will manage the influx of rapid submissions due to lower costs associated with writing new analytical work enabled by AI. He noted that submission fees may increase as publishers face higher volumes of submissions but cannot expand their capacity quickly enough.
- Scott Cunningham argues that while the quantity of papers may rise due to AI-generated content, this could lead publishers to reject more submissions immediately due to quality concerns. However, he is challenged on this point by the speaker who provides insights into publisher revenue dynamics shifting towards open access fees rather than traditional subscriptions.
Future Trends in Publishing
- The speaker predicts an overwhelming surge of low-quality, AI-generated papers being published as long as researchers continue paying for publication opportunities—potentially funded by public universities or grants derived from taxes. This trend could shift perceptions around publishing volume from a virtue to a flaw within academia.
- As we approach 2026, there’s an emphasis on upgrading skills amidst evolving job landscapes influenced by advanced AI capabilities capable of processing vast datasets and functioning alongside human workers seamlessly. Outskill is introduced as a platform aimed at enhancing individuals' understanding and application of these technologies effectively in their workflows.