👾 AI & Research

Science is currently experiencing a transformative shift: artificial intelligence is penetrating research areas in a depth and breadth that seemed unthinkable just a few years ago. From decoding molecular structures to predicting complex climate patterns, AI models are taking on tasks that were once reserved for human intuition and decades of research. But how did this rapid rise of AI in science come about, and which models are currently shaping this development? In the following, we will discuss the latest developments and provide an outlook on what this may mean for us as a society.

The Rise of AI in Science: Pioneering Models and Their Applications

The integration of AI into scientific research is not a sudden phenomenon, but the result of decades of developments in machine learning and data processing. With the exponential increase of data in almost all disciplines, traditional data analysis reached its limits. This is where AI models offer new possibilities: they can process huge amounts of data in a short time, recognize patterns and make predictions that are almost impossible for the human mind.

AlphaFold from Google DeepMind represents one of the most significant breakthroughs in the field of artificial intelligence. This AI model is able to predict the three-dimensional structure of proteins based on their amino acid sequence alone, with unprecedented accuracy. Since proteins are essential building blocks of life and control almost all biological processes, their spatial structure is crucial to their function. However, determining these structures has been one of the biggest challenges in the life sciences until the development of AlphaFold. Scientists had to rely on experimental methods such as X-ray crystallography, nuclear magnetic resonance spectroscopy (NMR) or cryo-electron microscopy – methods that are not only extremely costly and time-consuming, but often failed due to technical and biological limitations.

The introduction of AlphaFold has fundamentally changed this field of research. While traditional methods often took years, AlphaFold provides its predictions within hours or days – and with an accuracy that is comparable to experimental methods in many cases. This opens up a whole new dimension in protein research: scientists can now study how proteins interact in living organisms, how they cause disease, or how they can be targeted for medical and biotechnological applications much more quickly and efficiently.

Particularly in drug development, AlphaFold has already proven its enormous significance. The precise prediction of protein structures enables researchers to develop targeted drugs that interact with these proteins to combat diseases. This could drastically accelerate the development of drugs for diseases that have been difficult to treat, such as Alzheimer's, certain cancers, or rare genetic disorders. AlphaFold also opens up new perspectives in synthetic biology and biotechnology: the construction of customized proteins for industrial or medical applications will be greatly simplified by AI-based prediction.

The breakthrough AlphaFold has achieved is a striking example of how artificial intelligence is able to solve decades-old scientific problems that were previously considered almost insurmountable. The model has not only revolutionized protein research, but also has a profound impact on the entire life sciences. It is a striking illustration of the fact that AI is much more than a tool for automation – it can enable fundamental scientific insights that would hardly be achievable without it. (I have already written a longer text about AlphaFold 3 on the occasion of the Nobel Prize for Demis Hassabis: https://www.forwardfuture.ai/p/alphafold2-is-finally-getting-the-attention-it-deserves-a-nobel-prize-long-overdue)

GenCast (also from Google DeepMind), on the other hand, is an AI-based weather forecasting model developed by Google DeepMind to significantly improve the accuracy and speed of meteorological forecasts. Unlike traditional numerical weather prediction models, which are based on physical equations and massive computing power, GenCast uses machine learning to predict weather trends. The model was trained on four decades of historical weather data, enabling it to recognize patterns that are often overlooked by traditional models.

A major breakthrough of GenCast lies in its efficiency: while traditional weather models often take several hours to calculate a forecast, GenCast delivers comparable results in just eight minutes. In addition, it achieves a level of accuracy that can match the best numerical weather models and is even superior in some areas, such as predicting extreme weather events. This applies in particular to heat waves, cold spells, strong winds and tropical cyclones, whose more precise forecasting is crucial for disaster protection, energy planning and agriculture. (We have also written a detailed article about GenCast. https://www.forwardfuture.ai/p/gencast-the-best-ai-in-stormy-weather)

Recently, however, OpenAI has also attracted attention with its research model GPT-4b micro, which it announced would be one of the first models to be fully dedicated to research. 

GPT-4b micro is an AI model developed by OpenAI that was specifically designed for longevity research. In collaboration with Retro Biosciences, this model aims to improve efficiency in the production of stem cells. By applying machine learning to biological data, GPT-4b micro can redesign proteins, in particular the so-called Yamanaka factors, to increase their function. These factors play a crucial role in the conversion of human skin cells into pluripotent stem cells, which have the potential to differentiate into various cell types.  The Yamanaka factors are four specific proteins (transcription factors) discovered by Shinya Yamanaka and his team. They are called Oct4, Sox2, Klf4 and c-Myc (often referred to as OSKM factors). These factors have the remarkable ability to reprogram already specialized cells - for example skin cells (fibroblasts) - back into a pluripotent state. These reprogrammed cells are called induced pluripotent stem cells(iPS cells).

GPT-4b micro's ability to optimize such proteins could enable significant advances in regenerative medicine and research into age-related diseases. This approach underscores the growing importance of AI in biomedical research and demonstrates how machine learning can help accelerate scientific discovery. 

In addition to the models mentioned, there are other specialized AI applications that are advancing research. For example, OpenAI has developed a system with the o-series model that answers or even exceeds scientific questions at the doctoral level. With the ability to analyze complex problems and draw logical conclusions, o1/o3 opens up new possibilities in scientific research. 

Another example is the AI model “Evo”, which serves as a creative tool for gene optimization. This model could initiate the next stage of evolution by optimizing genetic sequences and thus enabling new approaches in genetic research.

We could cite countless other models that are already being used in research today and have revolutionized research across the board. However, the examples given here should convey a significant sense of the importance of AI in current research. More important, however, is the question of where this technical revolution is heading and what that means for the future.

Outlook: The Future of AI-Supported Research (Agents)

The future of scientific research will be significantly shaped by autonomous AI agents. While artificial intelligence is used primarily as a tool for data analysis and information gathering today, future AI systems will independently conduct research, formulate hypotheses, carry out experiments and generate new scientific findings. This development could fundamentally change the way knowledge is created.

A key aspect of this revolution is the ability of AI agents to analyze and link scientific literature in real time. Instead of researchers laboriously sifting through thousands of publications, an AI will accomplish this task in seconds, extracting key messages and creating syntheses. What is particularly exciting is that AI agents can recognize patterns across disciplinary boundaries. While a human being often remains within the thought patterns of his or her own field, an AI could create new cross-connections between medicine, physics, sociology and other sciences. For example, an AI agent could discover correlations between genetic factors, environmental conditions and neurological diseases that have been overlooked so far.

But AI agents will not only summarize existing knowledge, they will also formulate new hypotheses and test them experimentally. With the help of large amounts of data and advanced simulation techniques, they could test theories before any physical experiments are carried out. This would mean a huge acceleration, especially in areas such as materials science or drug development. For example, an AI could simulate billions of possible molecule combinations to identify promising drugs for cancer or Alzheimer's before these are tested in real laboratories.

The moment when AI agents not only test theories but actually discover new laws of nature will be particularly revolutionary. In the future, AI agents could identify physical anomalies that lead to the formulation of new theories. Perhaps an AI will succeed in unifying gravity and quantum mechanics in a single unified theory by recognizing mathematical patterns that remain hidden from the human eye. In this context, we have not yet looked at the additional development of AI agents in the embodiment of humanoid robots; in other words, when AI has access to the physical world.

The symbiosis of AI and science is only just beginning. As technology advances and our understanding of the potential of AI grows, we can expect to see its application in research continue to increase. Future developments could include personalized medicine, more precise climate models and deeper insights into complex biological systems.

All of this development could lead to the singularity in the very near future, when all research and development is done by AI itself, and the findings of research flow back as a positive feedback loop into the development of AI researchers. (For more on the concept of singularity, see: https://www.forwardfuture.ai/p/the-concept-of-the-singularity) This so-called intelligence explosion would catapult us into a new era of human intelligence, as Sam Altman calls it, “The intelligence Age”.

—

Get more content from Kim Isenberg—subscribe to FF Daily for free!