It's no secret that Facebook makes heavy use of AI, indeed the company is infamous thanks to those who feel they are hard done by the notoriously opaque algorithms utilized by the social network to create targeted ads or moderate harmful content. But Facebook recently shared some information on its efforts to make artificial brains for actual physical machines.
According to a recent blog post and subsequent reports, Facebook AI Research or FAIR is working on robots that use AI to navigate the real world. Key aspects that researchers are keen to develop is curiosity, adaptability and self-learning. From FAIR's blog:
"Much of our work in robotics is focused on self-supervised learning, in which systems learn directly from raw data so they can adapt to new tasks and new circumstances...in robotics, we're advancing techniques such as model-based reinforcement learning (RL) to enable robots to teach themselves through trial and error using direct input from sensors."
You may already be familiar with this concept through virtual training programs, such as AIs learning to navigate video game like environments filled with challenges and obstacles. But Facebook gets more physical here. One particularly exciting (or terrifying, depending on how you feel about robots crawling around) project involves a hexapod, or six-legged robot, teaching itself how to walk in the real world.
The insectoid-looking contraption above is born as a blank slate but equipped with various sensors through which it must learn to understand and move around its environment. Using a reinforcement-learning algorithm, this computer critter should discover via trial and error what its own movement capabilities are, what the objects around it are like and invent the best way to get itself around. Like a newborn baby, these robots will learn a sense of their own weight, balance and strength (though thankfully, they are not growing bigger and stronger alongside this).
If you don't succeed, try and try again...
That's where curiosity comes in. The AI needs to be rewarded for discovering new things about itself, which it might never learn to do if its hyper-focused on completing a specific goal. That way, we become ever closer to develop an AI that has a kind of 'general intelligence', even if the trial and error-process leads to a lot of seemingly hapless flailing around of mechanical limbs until the robot figures something about. Speaking to Wired, FAIR researcher Franziska Meier put it like this: “Although it didn't achieve the task, it gave us more data, and the variety of data we get by exploring like this is bigger than if we weren't exploring.”
- AIs are being raised on video games...not to beat us, but to join us
- How to log out of Facebook Messenger
All well and good, though it's as yet unclear why exactly Facebook, which specializes in social networks and the digital sphere, is developing these self-learning walking robots. At the moment, Facebook says this research isn’t connected to a particular product for the future, but we can look at recent hardware efforts from the social network titan. Something you can control remotely via Oculus VR? A home cleaning robot to complement its Portal smart home range?
It's fair to say that Facebook still has to recover some public trust before unleashing many-legged machines upon the world. As a colleague immediately voiced upon hearing the news of Facebook's robot research: "what could possibly go wrong?"