Soon after birth, humans and many animals show abilities that seem to come out of nowhere. The brain, with its trillions of connections, makes these behaviors possible. However, the genome, which is like the recipe book for all life, only has room for a tiny bit of that information. This has puzzled scientists for a long time. How can such limited information produce such complex behaviors?

Researchers at Cold Spring Harbor Laboratory have come up with a new idea. They think that the genome’s limited space might actually be what makes us smart. They suggest that because the genome can’t contain all the details, we’re forced to learn and adapt quickly. This concept is hard to test in real life because evolution takes billions of years, but Artificial Intelligence (AI) offers a faster way to experiment.

In a paper published in PNAS, the researchers describe their new “genomic bottleneck algorithm” for neural networks. This algorithm compacts a lot of information into a smaller space, similar to how our genome might pack the instructions for brain development.

Compressed neural networks

“Our results suggest that compressing a neural circuit through the genomic bottleneck serves as a regularizer, enabling evolution to select simple circuits that can be readily adapted to important real-world tasks,” note the researchers in the paper.

They tested compressed neural networks against other AI systems, and found that their new, untrained algorithm could perform tasks like recognizing images or playing simple video games almost as well as the best trained AI models.

The algorithm’s ability to compress information could be very useful in technology. For instance, it could help run sophisticated AI models on devices with limited memory, like cell phones, by expanding the model layer by layer as needed.

This work suggests that the limitations of our genome might not be a flaw but a feature that encourages adaptability and intelligence. It’s a fascinating hint that our evolutionary journey could inform the future of AI, making it more efficient and perhaps closer to achieving human intelligence.

Let us know your thoughts! Sign up for a Mindplex account now, join our Telegram, or follow us on Twitter.