We all have grown up learning how every living being evolved to survive on Earth by fighting for its existence. The theory ‘survival of the fittest’ has somehow emerged as a proverb, training all of us to adapt external changes to ensure our survival. However, this theory not only applies to living beings, rather on technology as well. For instance, using a VPN was earlier believed to be the sole protection from cyber threats. However, the tech experts are now looking on how does SSL VPN work to ensure better security. Thus, knocking off the older VPN protocols. Likewise, scientists have developed Darwinian concepts to train artificial intelligence. They work out to test the algorithms and let the fittest algorithms survive.
Let’s Recall Darwin’s Theory
Darwin, while reasoning the evolution of living beings, gave one of the most promising theories of all time – the theory of natural selection. It sheds light on the way living beings evolved to adjust with environmental and external changes so as to become ‘fit’ for survival. This change can also be passed to future generations, hence, triggering the process of evolution. The change (i.e., mutation) can either be the development or acceptance of a trait if it’s beneficial, or it can be the abandoning of the trait if it’s disadvantageous.
Training AI Algorithms
Technologists have realized the hidden potential of Darwin’s theory. They now apply the same theory of AI algorithms as well.
The scientists keep working to develop different algorithms with AI features. However, it’s not necessary that all of them work fine. To exhibit near-genuine intelligence, these algorithms should adapt to changes (or mutations) for broad scope application. If an algorithm fails to adapt to changes, it will certainly be abandoned. Whereas, the one bearing most alterations without affecting its expected functionality and application will survive. Such types of changes are specially employed in case of neural networks and delivery routes.
The idea isn’t new though. Scientists are now trying to develop generic algorithms to train and test other algorithms. This will allow working on a single broader algorithm saving the efforts to train multiple algorithms.
For instance, Microsoft’s OpenAI is one such attempt to train other algorithms to perform in unknown situations. This helps in better machine learning processes. Although, such a procedure has its own limitations as well. But the overall concept seems more beneficial than disadvantageous. Perhaps, that’s the reason why tech giants like Google and Microsoft are heavily investing to train algorithms as they strive to pave the way for future AI applications. Not only this will help solve more problems with smartest (or the fittest) algorithms, but will also facilitate in accomplishing tasks that presently require a lot of computing power.