Reproduction is a biological process unique to only living things.
Every organisms exist as the result of reproduction. It's an important process, and also essential for living things to evolve through natural selection. But humans are trying to debunk that fact.
Two computer scientists, Oscar Chang and Hod Lipson, have successfully created a neural network capable of self-replicating.
"Self-replication is a key aspect of biological life that has been largely overlooked in Artificial Intelligence systems," they argued in a paper titled Neural Network Quine.
Chang as the first author of the paper, explained that the goal was to see if AI could be made to be continually self improving by mimicking the biological self-replication process.
In their work, the scientists compare quines, which is a type of computer program that learns to produce copies of its own source code. In neural network however, it's not the source code that is copied. Instead, it's the weight. This is something that determines the connection between different neurons that are being cloned.
Here they set a "vanilla quine" network, a system that produces and also capable of replicating its own weights as outputs, other than just solving a task.
In a test, the scientists used this system for image classification on the MNIST dataset, where the AI had to identify the correct digits from a set of handwritten numbers.
The network has been designed to have a smaller scale at a maximum of 21,100 parameters, compared to several millions on standard image recognition models.
After feeding the network 60,000 MNIST images for training, and another 10,000 for testing, after 30 training runs, the network had an accuracy of 90 percent.
The "self-replication occupies a significant portion of the neural network’s capacity." What this means, the AI was less capable in doing image recognition if it also has to self-replicate.
"This is an interesting finding: it is more difficult for a network that has increased its specialization at a particular task to self-replicate. This suggests that the two objectives are at odds with each other," Chang said. "It's not entirely clear why this is so. But we note that this is similar to the trade-off made between reproduction and other tasks in nature. For example, our hormones help us to adapt to our environment and in times of food scarcity, our sex drive is down-regulated to prioritize survival over reproduction."
"To our knowledge, we are the first to tackle the problem of building a self-replication mechanism in a neural network. As such, our work should be best viewed as a proof of concept," Chang added.
Here, they hope that the technology can be used for computer security to self-repair damaged systems in both hardware and software.