A crucial part of Google's strategy for the coming years revolves around artificial intelligence. The company's platforms are rapidly absorbing new capabilities, but according to CEO Sundar Pichai, the only way to move forward is to delegate the creation of new learning structures to existing models. Google Search and Google Assistant would be the primary beneficiaries, yet the company also envisions applications in medicine and chemistry.
Code That Writes Code
Many engineers believe this is inevitable, and that artificial intelligence will be a fundamental piece. In fact, artificial intelligence is shaping up to be a fundamental piece in dozens of markets far beyond programming. We've seen it draw, recognize objects, create films, edit trailers, and drive cars. We also released one on the Web, with disturbing results. But an artificial intelligence capable of creating another and optimizing it based on what it learned before... that's on an entirely different level. As expected, Google plans to achieve it.
Google's Answer: Google.ai and AutoML
CEO Sundar Pichai took the stage on the first day of the I/O 2017 conference, essentially admitting that creating new learning models is getting increasingly difficult. At the same time, the pool of experts is limited, so the next step is the automation of certain tasks. As a starting point, Google announced its new initiative Google.ai, and opened access to its Tensor Processing Units so that hundreds of thousands of developers can use machine learning to accelerate and train their own projects. From there, the new neural networks the company creates will make other neural networks with the help of AutoML. The system takes a group of "candidate networks" and subjects them to a special training method until the best one is discovered.
One of the plans is to channel these developments into optimizing Google Search and Google Assistant. There's even speculation that in the future they will cease to be separate products, but there's more. Pichai spoke about creating tools specifically designed to assist doctors and health researchers. Faster diagnoses, precise analysis of the progression of certain diseases such as cancer, and the formation of new drugs are some of the possibilities. It will take time, but they're already on the way.
