Each living creature has been able to action from the moment of birth. Many animals demonstrate complex behavior immediately after birth: the spiders weave a web, whales begin to swim. This behavior is associated with the brain that contains trillions of neural connections necessary for controlling complex processes.
However, the genome can contain only a small part of this information, which has long been at a dead end. Researchers from the Kold-Spring Harbor laboratory proposed a solution to this riddle based on artificial intelligence methods.
Scientists suggested that the limited capacity of the genome may not be a drawback, but an advantage. This property may make the body adapt and study quickly, providing intellectual development. This approach forms the basis of the new algorithm of the “genomic narrow place”.
Unlike evolutionary processes, where generations are developing for decades, in artificial intelligence, new models are created instantly. Researchers have developed an algorithm that compresses data as the genome does, packing the information necessary for the formation of functional brain schemes. The results were tested on artificial intelligence networks.
The study is published in the journal Procedings of the National Academy of Sciences . The new algorithm showed high efficiency, solving the problems of image recognition and demonstrating abilities in video games, such as Space Invaders, without prior training.
Despite success, researchers note that the algorithm cannot yet completely compete with the natural capabilities of the brain. The architecture of a person’s bark is capable of accommodating about 280 terabyte of information, while the genome has only an hour of data, which implies 400,000-fold compression.
The new approach opens up the prospects for use in technology, including the possibility of launching large language models on devices with limited resources, such as smartphones. This can accelerate the work of artificial intelligence and make it more compact and effective.