Canine Tech: Artificial Intelligence Trained to Behave as a Dog

We live in exciting times: from self-driving vehicles to cloning, there seems to be a new scientific breakthrough or a technological advancement with each new day. Now, researchers have something canine-tastic to wow us with, as a team working out of the University of Washington and Allen Institute for AI taught a neural network to behave like a dog.

The machine-learning algorithm was learning from the best our world has to offer, which would be doggos. The goal of the study was to see if it would be possible to model AI behavior based on visual data, so the process was focused on how man’s best friend sees the world. The research wouldn’t be possible, I might add, if it wasn’t for Kelp, an Alaskan Malamute who was the artificial intelligence’s teacher.

The process looks pretty fun: scientists used a GoPro camera and motion tracking sensors strapped to this very good boy. Then, the neural network used the collected data to, essentially, think as a doggo does. Pooch thinking actually helped AI overcome the issue of determining which surfaces are walkable, which is a big deal for AI researchers. Until the neural network got a “peek” inside a canine brain, it wasn’t able to discern where it should walk and where not, resulting in frequent falls and wall-hitting.

In addition to helping scientists improve machine learning algorithms by providing it with a better understanding of visual intelligence, this canine-taught AI study could have other applications. For instance, having a neural network that essentially thinks like a dog could make it possible for scientists to make lifelike pet robots or “preserve” a dog’s conscience by training an AI to act like them. Who knows what the future holds? (In any case, as long as there are dogs in it, I think we’ll be OK.)