The possibility of a machine built by humans that can understand, analyze and interact with the world around it may seem as unimaginable as traveling to the moon was some decades ago. But now, the field of artificial intelligence, or AI, is making this just as feasible as man walking on the moon.

Philipp Krähenbühl, UT computer science associate professor, said scientists are drawn to this area of computer science because there are many challenging problems to be solved and a lot of room for discovery and innovation. Technological companies, such as Facebook, invest money, time and efforts on the development of AI and machine learning. 

Krähenbühl said he was attracted to UT because of the University’s commitment to being at the forefront of AI evolution. He said UT’s machine learning and AI department continues to expand and has plans to recruit up to four more faculty in the next few years. 

Krähenbühl’s main field is in computer vision, which involves “educating” computers to recognize images and objects. He said one main problem is that computer vision requires vast amounts of data, such as thousands of images of the same object. 

Krähenbühl said the problem is that it is infeasible to collect that much information for every object that you want the computer to be able to identify. He added that one possible solution is based off of video games. 

“Through video games we can get a lot of this labeled information, at least for low-level computer vision, for free,” Krähenbühl said. “In that way, you can play a video game for a day long and have the millions of images that you need to train your computer.”

Krähenbühl is also working on optimizing neural networks to require less data. 

The goal of computer vision and AI in general is to mimic the functions of the human brain. However, Krähenbühl said computer scientists should focus more on how well they can train computers to understand the world, even if that means surpassing human abilities.

“If a computer can do better in a specific aspect, it should do better” he said. 

Krähenbühl said one example of this is software that can magnify small movement; this type of computer could measure a person’s pulse by monitoring their facial movements

Krähenbühl said applications of AI have the potential to change the social and economic balance of society, but the extent and type of this change depends on how humans use AI. He said some new dangers of AI include privacy, security and safety. While debates about the ethics of AI are not far-off, Krähenbühl said people should not worry because computers are unlikely to dominate the human world, as long as scientists remain wise about the way they build and control the systems.