In Sxsw London, I had the pleasure of seeing Deepmind’s co -founder, Demis Hassabis, talk about the future of artificial intelligence. Among many stimulating points, two comments were left. First, he emphasized the importance of understanding the foundations. Secondly, he defended the scientific method as a guiding principle to make significant progress in AI.
As someone who works at the intersection of open research and AI, I have found myself returning to a specific question: what kind of content really cares about AI? I have previously spoken of a steering wheel effect on the investigationWhere each cycle follows this pattern:
- Unprocessed data are processed by AI to generate initial research results
- Knowledge extraction tools extract these outputs for higher order ideas
- These ideas form a new refining data set
- Ai processes this refined data set, generating more precise analysis
- The cycle continues, with each rotation producing a more valuable knowledge
I interpret “Understand the foundations” as the base layer, unprocessed data. Models can imitate almost any writing style and generate endless text resumes. Not all content is created the same. I have witnessed this first hand as self -denominated “academic” around the world, use generalist repositories to publish content not reviewed by pairs written by LLMS that demonstrates its genius.
AI systems, particularly large language models depend on data to learn. Not only more data, but better data; Data that reveal a new structure in the world. Without novel contributions, the AI models will improve to reformulate the known, but not to understand the unknown. In essence, science is a method to produce high quality structured novelty, a repeatable process to generate new facts, try them against reality and share them with the world. Basic research, often financed for long -term potential instead of short -term applications, is the main engine of this type of content. Alfafold was successful because he was trained in grounded empirical data from Protein Data Bank.
If we want AI to continue advancing in a significant way that you discover new knowledge, we need to prioritize access and support for new scientific research. That means supporting open science. It means investing in infrastructure that guarantees that new data and discoveries are fair (which can be seen, accessible, interoperable and reusable). It means rethinking publication practices to encourage the dissemination of negative results, replication studies and unprocessed data. And means financing basic research worldwide.
China has significantly increased its investment in basic research, with the aim of reducing dependence on foreign technology and achieving self -sufficiency in fundamental sciences. China’s expenditure on basic research approved 6% of its total R&D in 2023 and Continue up. China seems to be an atypical case. Here in the United Kingdom, for example, basic research through UCRI continues, but often faces pressures to demonstrate short -term economic impact. In the United States, NSF and NIH continue to support basic research, but federal R&D budgets have changed towards applied and applied research by the mission. He Inflation reduction law and chips and science act He brought some elevation, but basic science still receives a minority part of total R&D.
The geopolitical landscape is difficult to predict. But the narrative is that all countries want to compete in the AI stage. Radical abundance will only occur in your country if you have any entrance control to the models.
In a separated session of Sxsw London, we also listen to former United Kingdom Tony Blair and The RT Hon Peter Kyle MPSecretary of State for Science, Innovation and Technology of the United Kingdom. Peter Kyle’s plans to integrate Ai in the United Kingdom government They are commendable and seem to advance at a rate by which the United Kingdom government is not famous. However, a comment by Tony Blair highlighted what is at stake if we do not finance a basic investigation: “It is surprising for me that we are not feeding all the NHS data in these AI models.”
Whether it trusts this government with your most sensitive data or not, we will have future governments that may not follow your best interests; In the same way that LLM models ignore copyright in today’s academic publications, they can ignore human ethics when it comes to their medical data. Feeding the NHS to LLM is not the answer. The easiest way to generate new data for AI models in order to advance science and technology is to finance more basic research and insist that the results are opened fairly.
There is much to win. The value of new scientific research has never been greater.
#facts #scientific #research
