AI System Conducts Independent Scientific Research
amsterdam, woensdag, 20 augustus 2025.
A domain-independent, agent-based AI system has independently designed, conducted, and reported three psychological studies. The system collected new online data, developed analysis pipelines, and produced full manuscripts. This marks a step towards more generally applicable AI in science, raising important questions about the attribution of scientific merit.
A Step Towards More Generally Applicable AI in Science
A domain-independent, agent-based AI system has independently designed, conducted, and reported three psychological studies. The system collected new online data, developed analysis pipelines, and produced full manuscripts. This marks a step towards more generally applicable AI in science, raising important questions about the attribution of scientific merit [1].
Autonomous Research Skills
The AI system conducted three psychological studies focusing on visual working memory, mental rotation, and image vividness. The system collected new online data from 288 participants, developed analysis pipelines during sessions lasting more than eight hours, and produced full manuscripts. The results demonstrate that AI scientific discovery pipelines are capable of conducting non-trivial research with theoretical reasoning and methodological rigour comparable to that of experienced researchers, although there are limitations in conceptual nuances and theoretical interpretation [1].
Impact on Scientific Practices
This development marks a significant step towards more generally applicable AI in science. It offers the possibility of testing hypotheses through real experiments, which can accelerate discoveries by exploring areas of scientific space that would otherwise remain unexplored due to human cognitive and resource limitations. However, it also raises important questions about the nature of scientific understanding and the attribution of scientific merit [1].
Application in Other Sectors
While this study focuses on psychology, the implications of agent-based AI are far-reaching. In healthcare, agent-based AI systems can contribute to the development of personalised treatment plans, while they can be used for adaptive portfolio management in the financial sector. In the manufacturing industry, they can ensure optimal process control, and in the transportation sector, they can contribute to the functioning of self-driving vehicles [2][3].
Ethical and Regulatory Issues
With the rise of autonomous AI agents, new issues arise, including cybersecurity and regulation. AI web agents can exhibit human-like behaviour and circumvent traditional regulatory systems, necessitating advanced motive recognition by companies. Legally, AI systems are increasingly being regulated worldwide. The EU AI Act, which came into effect on 1 August 2024, will be enforced from 2 August 2025, with stricter requirements for high-risk and autonomous systems [4].
Economic Impact
According to the Capgemini Research Institute, agent-based AI could generate up to $450 billion in economic value by 2028. However, much needs to be done to unlock this potential. 93% of executives believe that scaling AI agents within 12 months will provide a competitive advantage, but 82% of organisations still lack a mature AI infrastructure and have less than 20% of the required data in place [5].
Trust and Privacy
Trust in autonomous AI agents is declining; the percentage of executives who fully trust autonomous AI agents dropped from 43% to 27% in one year. Privacy is the biggest concern, cited by 51% of executives, but only 34% have taken measures. 61% of employees fear the impact AI agents will have on their future jobs and careers [5].
Future Vision
The future of agent-based AI looks promising, with a focus on collaboration with humans rather than replacement. AI agents can function as intelligent partners, enhancing human capabilities. To realise the full potential of this technology, however, more is needed than just technology; attention must also be paid to responsible development, transparency, and ensuring human-centred innovation [2][3][4].