Google Brain has released research about their testing on how machine learning algorithms can be created from scratch and evolve naturally, without human intervention. Using simple maths concepts as building blocks, the AutoML team suggests that the software could potentially be updated to “automatically discover” completely unknown algorithms, somewhat resembling the process of evolution - the code improves every generation. 

The machine will spit out 100 randomly generated algorithms and then work to see which ones perform well against standard artificial intelligence (AI) problems and choose the best for evolution. The automation also frees the AI algorithms from human bias since there is negligible human interaction - the machines are now capable of finding things we’d never think of. 

“Human designed components bias the search results in favour of human-designed algorithms, possibly reducing the innovation potential of AutoML. Innovation is also limited by having fewer options: you cannot discover what you cannot search for,” write the researchers in the paper which was published last month on arXiv titled “Evolving Machine Learning Algorithms From Scratch”.

According to the researchers, AutoML Zero already outperforms its predecessor and similar machine learning-generation tools. Now, the team will attempt to set a more narrow scope for the AI and seeing how well it performs in more specific situations using a hybrid approach that creates algorithms with a combination of ‘Zero’s’ self-discovery techniques and human-curated starter libraries.

The Google Brain team members who collaborated on the paper said the concepts in the most recent research were a solid starting point but stressed that the project is far from over. There still exist a few drawbacks in AutoMl, for instance - one has to manually create and tune several algorithms to act as building blocks for the machine to get started. This allows it to take the work and experiment with new parameters in an effort to optimise what has been done. Novices can get around this problem by using pre-made algorithm packages, but Google’s working to automate this part too.