Two breakthroughs bring us closer to achieving AGI

The biggest technology game changers don’t always grab the biggest headlines. Two emerging AI developments may not go viral on TikTok or YouTube, but they represent an inflection point that could radically accelerate the development of artificial general intelligence (AGI). That’s AI that can function and learn like us.

Coming to our senses: WildFusion

As humans, we rely on all sorts of stimuli to navigate in the world, including our senses: sight, sound, touch, taste, smell. Until now, AI devices have been solely reliant on a single sense—visual impressions. Brand-new research from Duke University goes beyond reliance only on visual perception. It’s called WildFusion, combining vision with touch and vibration. 

The four-legged robot used by the research team includes microphones and tactile sensors in addition to the standard cameras commonly found in state-of-the-art robots. The WildFusion robot can use sound to assess the quality of a surface (dry leaves, wet sand) as well as pressure and resistance to calibrate its balance and stability. All of this data is gathered and combined or fused, into a single data representation that improves over time with experience. The research team plans enhance the robot’s capabilities by enabling it to gauge things like heat and humidity. 

As the types of data used to interact with the environment become richer and more integrated, AI moves inexorably closer to true AGI. 

Learning to learn

The second underreported AI technology game changer comes from researchers at the universities of Surrey and Hamburg. While still in the early stages of development, this breakthrough allows robots that interact socially with humans (social robots) to train themselves with minimal human intervention. It achieves this by replicating what humans would visually focus on in complex social situations.

For example, we learn over time as humans to look at a person’s face when talking to them or to look at what they are pointing to rather than at their feet or off into space. But robots won’t do that without being specifically trained. Until now, the training to refine behavior in robots was primarily reliant on constant human monitoring and supervision.

This new innovative approach uses robotic simulations to track, monitor, and importantly, improve the quality of the robot interactions with minimal human involvement. Robots learn social skills without constant human oversight. This marks an important step forward in the overall advancement of social robotics and could prove to be a huge AGI accelerator. Self-teaching AI could lead to advancements at an exponential rate, a prospect some of us view as thrilling, others as chilling.

AI signal over noise

Amazing as they may be to watch, dancing humanoid robots and mechanical dogs can be characterized as narrow AI—AI designed only for a specific task or purpose. The feats of these purpose-built tools are impressive. But these two new developments advance how AI experiences the world and how it learns from those experiences. They will dramatically change how technology exists (and coexists with us) in the world.

Taken together, these breakthroughs and the work of other researchers and entrepreneurs along similar paths are resetting the trajectory and the timetable for achieving AGI. This could mark the tipping point that turns the slow march toward AGI into an all-out run.