Several arguments and examples from the sources challenge Vincent Icke's view that AI cannot perform groundbreaking scientific work due to its reliance on existing data and an alleged inability to formulate hypotheses:

• AI can identify patterns and correlations in data AI systems are adept at finding correlations, patterns, and structures within data.
• AlphaFold's success demonstrates AI's problem-solving capabilities: AlphaFold was able to assist in determining the structure of the nuclear pore complex, which governs the molecules and nutrients that go in and out of a cell nucleus. AlphaFold has also been used to design plastic eating enzymes.
• AI can generate novel, testable hypotheses: The AI co-scientist system can generate novel research hypotheses, detailed research overviews, and experimental protocols.
• AI can accelerate scientific discovery: AI systems can act as tools to accelerate scientific discovery. The AI co-scientist is intended to function as a collaborative tool for scientists.
• AI can assist with drug discovery: Isomorphic, a spin-out from DeepMind, aims to revolutionize drug discovery using AI, with the goal of reducing the time to design a drug from years to months or weeks. The AI co-scientist can propose novel repurposing candidates for diseases, such as acute myeloid leukemia, and validate these proposals through experiments.
• Classical systems may be capable of modeling complex systems: Demis Hassabis believes that classical computers, if used correctly, may be capable of modeling complex systems, even quantum systems.
• The scientific method can be applied to AI: Demis Hassabis advocates for using the scientific method to understand AI systems, including neuroscience techniques to model virtual brains.
• AI can improve itself: AI systems can improve themselves through automated feedback and iterative refinement of hypotheses. The AI co-scientist leverages test-time compute scaling to iteratively reason, evolve, and improve outputs.

People who provide these arguments and examples:

• Demis Hassabis: As the CEO of DeepMind, Hassabis highlights the potential of AI in various scientific domains, including protein structure prediction, material design, and mathematics. He emphasizes the importance of interdisciplinary science and the use of AI as a tool to accelerate scientific discovery.
• John Jumper: A director at Google DeepMind, Jumper led the team that built AlphaFold. He notes the impact of AlphaFold on the scientific community, with researchers using it to make discoveries worthy of publication in prestigious journals.
• Juraj Gottweis and Vivek Natarajan: These Google researchers introduced the AI co-scientist, a multi-agent AI system designed to help scientists generate novel hypotheses and research proposals.
• Geoffrey Hinton: A physics Nobel laureate, Hinton suggests AI models are closer to understanding language than many believe, and advocates for more safety research in AI.

Yes, the sources suggest that neural networks can learn the skills to become assistants to scientists and aid in scientific discovery.

Here's how:

• Pattern Recognition and Correlation: Neural networks excel at identifying patterns and correlations in large datasets, which can help scientists in various ways.

• Hypothesis Generation: AI systems like the AI co-scientist are designed to generate novel research hypotheses and experimental protocols. This directly addresses Icke's concern about AI's inability to formulate hypotheses.

• Prediction and Modeling: AI can build models that understand the physics and chemistry of natural phenomena. These models can then be used to efficiently search through combinatorial spaces to find optimal solutions. AlphaFold is an example of AI solving a complex problem and unlocking new avenues of discovery in disease understanding and drug design.

• Interdisciplinary Applications: AI facilitates interdisciplinary work by bringing together experts from different fields. This collaboration can lead to significant advances in areas where multiple disciplines intersect.

• Automation of tasks: AI can automate tasks such as literature review, summarization, and "deep research," allowing scientists to focus on higher-level reasoning and experimentation.

• Accelerating Discovery: AI systems can accelerate scientific discovery by acting as collaborative tools for scientists. For example, Isomorphic aims to revolutionize drug discovery, potentially reducing the time to design a drug from years to months.

• Self-Improvement: AI systems can use automated feedback to iteratively generate, evaluate, and refine hypotheses. The AI co-scientist, for example, leverages test-time compute scaling to improve its outputs.

• Understanding Complex Systems: Demis Hassabis posits that classical computers, when used in the right way, can model complex systems, even quantum systems. He also advocates applying the scientific method, including neuroscience techniques, to better understand AI systems.

The development of AlphaFold showcases AI's ability to make significant contributions to solving complex scientific problems. Similarly, the AI co-scientist is designed to function as a virtual scientific collaborator, assisting scientists in generating novel hypotheses and research proposals.