Artificial Intelligence: Unravelling the Present by Predicting the Future
This past June, Google’s DeepMind developed a neural network capable of creating a three dimensional rendering from a single image. Named the Generative Query Network (GQN), the artificial intelligence (AI) can infer the entire layout of a virtual room without human interference. To do so, it uses the lighting, texture, and locations of objects to make its predictions on spatial relationships.
DeepMind Object ©️ DeepMind, CC-BY 4.0
Through this, it can figure out the configuration of a maze based on images taken from only a few different viewpoint.
DeepMind Maze ©️ DeepMind, CC-BY 4.0
According to the predictive coding theory, this methodology mimics the functioning of the human brain; generating models based on given information in order to create predictions on its surroundings. Then, the predictions are compared with sensory input to conclude the reasonings behind any discrepancies. Unexplainable errors are sent through feedforward signals to decide whether or not they are important, and to adjust accordingly. As Karl Friston, a neuroscientist at University College London said, “the game is now about adjusting the internal models, the brain dynamics, so as to suppress prediction error”. While the theory has yet to be strongly evidenced, it greatly resembles the design of GQN.
Technologically, GQN will be greatly applicable in the navigation of AI-driven robotics in both digital and physical spaces. Additionally, neuroscientists may one day be able to train AI to perform tasks using this algorithm. I personally find it incredibly fascinating that – whether or not it mimics our cognitive function – this AI software is capable of doing something so seemingly abstract. As I believe that our future may be in the hands of AI, it is an amazing opportunity to witness their development. While most AI currently only have the IQ of a toddler, the future is fast approaching: who knows how soon it will be until AIs are fully developed?
Technology/Biology
Leave a Reply