The brain is the most outstanding decision-making organ in the human body, not only does it collect various data but also processes these data, guiding us to take further action. Its complexity still conceals many secrets unknown to mankind. It is commonly believed that the task of processing information must rely on a highly complex structure like the brain. However, nature has repeatedly proven that many relatively simple systems also have the capacity to process information.
For example, Eric Winfree, a molecular computing expert at Caltech, has proposed that cells can make decisions about whether to replicate or undergo programmed death through chemical signaling networks. Even phase transition phenomena in physical processes can be considered as water molecules deciding whether to become snowflakes or hailstones. Winfree has long been interested in the information processing capabilities hidden in nature.
In a study published in Nature, Winfree and his team designed a series of bionic DNA chains and combined them to form a system similar to computing systems. This system has parallels with neural network algorithms used in current artificial intelligence and is capable of pattern recognition and information categorization.
Using biomolecules to achieve computer-like circuitry is a hot direction in current scientific research. Researchers tend to choose self-assembling DNA molecules, whose stacking ability comes from their base pairing properties. Specially designed DNA strands, when mixed and cooled in a test tube, can assemble into predictable shapes and mosaic structures, thereby transmitting information.
Researchers hope to explore whether such structures can be used for pattern recognition, for instance, categorizing images by their grayscale values.
The coding method required to represent images in a test tube is to use DNA fragments of different “shapes” to correspond to each image pixel, the brighter the pixel, the more DNA fragments corresponding to it. During the cooling process of the solution, these DNA fragments assemble like puzzle pieces, aggregating into three possible configurations. Which configuration is formed depends on the proportion of different DNA fragments in the mixture.
Another researcher in the field of molecular computing, Konstantin (Gleb) Ovchinnikov from Maynooth University in Ireland, points out that each specific configuration represents a category.
Fascinatingly, this DNA system can not only differentiate 18 different images into three specified categories, but it is also capable of classifying images it has never encountered before, such as distorted versions of the same image.
A physicist from the University of Chicago, Arvind Murugan, notes that, akin to neural networks, this system can recognize the general similarities in pictures without needing to match every detail perfectly. “The goal of this research is not to replace neural networks, but to explore the inherent computational potential of matter itself,” says Murugan. They hope to discover more such computational capabilities in nature, which may be hidden in the things around us, yet to be discovered by us.
In the exploration of new studies in the molecular world, biomedical engineer Rebecca Schulman from Johns Hopkins University expressed great interest in a recent study. She believes this study has revealed an exciting phenomenon, that a large group of molecules can store information in a fuzzy manner through their interactions. She compares this to the ability of large groups of neurons in neural networks to store information, calling this discovery a gateway to a new world for her.
Shulman further expressed her thoughts, likening these discoveries to the first exploration of the unknown wonders of deep-sea ecosystems. Although this exploration has only revealed a glimpse, she believes it to be a call that inspires researchers to turn back and delve deeper into observation and study.
