If you’re curious about how Google’s artificial intelligence works, you’ve probably come across the company DeepMind. This UK-based firm has made a big impact in artificial intelligence. The company’s code for its neural network, called TensorFlow, is open source. The company has published numerous papers online. It has helped Google reduce its cooling costs by using fewer data centers. But what can you do with DeepMind?
The company’s scientists are keen to apply DeepMind’s techniques to other areas. In particular, they hope to apply their research to the field of reinforcement learning. They are already working with Google Maps. As a test bed, they are focusing on Google Maps. However, they hope to apply their algorithms to a wide variety of domains. If DeepMind can make maps, then they have achieved one of their goals.
DeepMind’s AI technology has already been successful in beating humans in Quake III Arena Capture the Flag games. They have also developed partnerships with human teams to train their algorithms. In the process of developing these programs, they must learn procedurally generated environments, cooperative behavior, and competition. The resulting programs are referred to as ‘For The Win (FTW) agents’ and competed in a tournament with 40 human players.
Another notable DeepMind project has already shown great promise for healthcare. Google is already using the company’s AI to personalise app recommendations in Google Play, which has become one of the world’s most popular stores. In addition, the company has developed a patient safety alert app called Streams that reviews test results for signs of illness and sends an instant alert when it needs to be evaluated. It’s worth mentioning that DeepMind’s AI has made headlines in medical circles as well.
DeepMind’s ethics unit is working to develop ethical AI algorithms. In 2014, the company established an ethics board, which is supposed to study the ethical issues associated with artificial intelligence and superintelligence. DeepMind’s ethics board, which includes Demis Hassabis and Mustafa Suleyman, has not publicly disclosed its members. However, the ethical board will continue to investigate these issues and will be able to help you decide how to use DeepMind’s technology.
One of the most interesting examples of DeepMind AI using Google Street View images is a video that shows how it navigates the streets of major cities without a map. The AI was rewarded when the agent reaches its destination location. In a similar experiment, DeepMind used the Google Street View images as first-person views of the city. It memorized routes and ways to get to different locations, and rewarded when it got there.
The team also trained the AI to walk, by giving it torque-controlled virtual bodies and parameters to walk. They also trained the AI to create realistic images from nothing. The researchers used the ImageNet database, which contains real-world examples of images, and trained it with that. It then learned how to walk in those bodies. The AI learned to overcome obstacles, and the results were impressive. The neural network had a new capability to create and recognize realistic images.
The AI is also capable of predicting protein structures. Proteins’ three-dimensional structure determines how they interact with each other. This information could help scientists better understand a range of diseases. Since proteins are key ingredients of new therapeutics, unlocking their structures may speed up the development of new drugs. DeepMind has plans to address many of these issues in its next version. However, until then, it is important to note that it is still not yet perfect.
Another area of application for DeepMind AI is weather forecasting. It is possible to predict the intensity of a storm or heavy rain up to two hours in advance. In this project, the UK Meteorological Office has worked with DeepMind to develop “nowcasting,” which is an artificial intelligence program that allows forecasters to identify the time and location of high-resolution precipitation up to two hours ahead. This information can be vital for emergency services and event organizers to prepare for impending bad weather.