Anima Anandkumar joined Ben Taylor, Chief AI Evangelist at DataRobot, on the More Intelligent Tomorrow podcast to discuss the future direction of AI technology and its possible enhancement by the addition of more human capabilities.

Bren Professor of Technology at California Institute of Technology (CalTech), Anima joined Nvidia three years ago as the Director of Machine Learning Research. 

Nvidia is now at the center of the AI ecosystem, Anima points out, as its advanced Graphics Processing Units (GPUs) act as the brain of computers, robots, and self-driving cars capable of understanding the modern world. It has also promoted its Common Unified Device Architecture (CUDA) open source framework, which enables the use of GPUs for general-purpose computing. CUDA currently has more than 2.3 million developers.                                                        

Even today, however, many activities are beyond AI’s reach. Because the world has so many dimensions, it’s easy for AI systems to have trouble trying to decide what to consider as a theory and what to try as an experiment. To overcome these limitations, some scientists are trying to transform AI into Artificial or Applied General Intelligence (AGI) software, which would incorporate human cognitive abilities. 

Anima says, “To me, intelligence is really the ability to learn and adapt to different environments, to any changes. I think that aspect is missing in many of the current robots. For instance, a robot may be able to do impressive dancing, backflips, and open the door, but it’s pre-programmed to do precisely that. It’s not learning on its own by observing others.”

Today’s robots lack the ability to discover and explore. But if robots in the future have AGI capabilities, the situation may be different. Consider swarms of flying drones that have social awareness. Like birds, they could fly in flocks and handle changes in turbulence effectively. 

“When driving, good citizens give way to others. Are we going to lose all that when self-driving comes along, or will there be social bonds created through new means?” Anima asks. “Human collaboration is important to enable us to make advances in AI.” 

People often think that AGI will involve massive computer brains. But they overlook the probable miniaturization of AI that will occur during its progression to AGI. Anima believes that as time goes on, computation will gradually move to the edge and be characterized by small devices, with limited battery capabilities and highly intelligent processing units. For example, a CalTech researcher is investigating the possibility of using tiny probes during robotic surgery that can monitor nearby organs when the operation is happening. 

In 2020, the top AI domain was healthcare, but many challenging problems remain, such as drug discovery. Anima says, “If we think about all possible molecules in this universe, I think that’s 10 to the 60….it’s mind-bogglingly large. You can never go through all of them. So how do we come up with accurate predictions of what molecules have the desirable properties?”  

For the immediate future, she plans to continue her work as co-founder at CalTech of the AI4Science initiative, which focuses on how to apply AI concepts to specific disciplines. For example, one project uses AI to solve partial differential equations, which drive most scientific simulations and involve massive calculations. In the broad area of AI for science, a lot of exciting, groundbreaking discoveries remain to be made. 

To hear more about Anima Anandkumar’s work at Nvidia and CalTech and the possible future developments in AI, check out the More Intelligent Tomorrow episode. You can also listen everywhere you enjoy podcasts, including Apple Podcasts, Spotify, Stitcher, and Google Podcasts.