Facebook on Tuesday said that it has made advancements in its artificial intelligence research to better spot objects, tie natural language understanding to image recognition, plan and better utilize predictive learning.
The advancements were outlined in a blog post by Facebook CTO Mike Schroepfer. The findings were also presented at the Dublin Web Summit.
Facebook's AI Research (FAIR) has been working on improving machine learning for years. Initially, the focus was on algorithms and better spotting images for user photos. Ultimately, this AI knowledge will make its way into Facebook's virtual reality efforts.
Coupled with other AI efforts from the likes of Google, Facebook and IBM you eventually get to cognitive machines.
Also see: AI, Automation, and Tech Jobs |
As for Facebook's developments, the latest findings highlights the ability to:
- Train computers to better identify objects. FAIR will present a paper in December that outlines a system that taught machines to distinguish objects in a photo 30 percent faster with 10 times less training.
- Combine object identification with natural learning so people can ask a machine what's in a photo.
- Learn and make predictions without supervision. Machines are learning to make inferences.
- Make plans. FAIR created a bot to play the board game Go.
If you bucket those four items, the first two yield returns for Facebook quickly and the latter two will take more development and play into the broader cognitive computing race.
A few key points worth pondering:
Learning time for AI is a big hurdle. Facebook's research to improve speed and cut the learning time is a worthwhile effort. And if you had to pick one you'd go for any advancement that cuts the time to learn. Machines take time to ingest data as well as learn from it. In the enterprise context, you could implement a system and then hear "well the system is still learning" for months.
The FAIR team said that its systems can make predictions correctly 90 percent of the time. That batting average is better than most humans and could have wide implications for weather and economic data.
Predicting and planning ultimately is the Holy Grail and could ultimately be the more scary developments for humans. Schroepfer said in Facebook's engineering blog that building models for predictions and plans is challenging. He said:
There are also some bigger, longer-term challenges we're working on in AI. Some of these include unsupervised and predictive learning, where the systems can learn through observation (instead of through direct instruction, which is known as supervised learning) and then begin to make predictions based on those observations. This is something you and I do naturally -- for example, none of us had to go to a university to learn that a pen will fall to the ground if you push it off your desk -- and it's how humans do most of their learning. But computers still can't do this -- our advances in computer vision and natural language understanding are still being driven by supervised learning.