Facebook AI Hunts & Removes Harmful Content
Facebook announced a new artificial intelligence technology that can quickly identify malicious content in order to make Facebook more secure. The new AI model uses “few shot” learning to reduce the time to detect new types of malicious content from months to weeks.
Learning in a few shots
Few learning has similarities to learning without a shot. Both are machine learning techniques that aim to teach a machine to solve an unseen task by learning to generalize instructions to solve a task.
Low-shot learning models are trained on a few examples and from there unseen tasks can be scaled and solved, in this case the task being to identify new types of malicious content.
The advantage of Facebook’s new AI model is to speed up the process of taking action against new types of harmful content.
The Facebook ad stated:
“Malicious content continues to evolve rapidly – whether it is driven by current events or people looking for new ways to evade our systems – and it is critical that AI systems evolve alongside it.
But it usually takes months to collect and catalog the thousands, if not millions, of examples needed to train each individual AI system to detect a new type of content.
… This new AI system uses a method called “learning with a few shots,” where models start with a general understanding of many different topics and then use far fewer – or sometimes zero – labeled examples to learn new tasks. “
The new technology is effective in a hundred languages and works on images and text.
Facebook’s new low-key learning AI is meant to add to existing methods for evaluating and removing harmful content.
Although it is an addition to the existing methods, it is not a small addition, but a big plus. The effect of the new AI is one of great magnitude as well as speed.
“This new AI system uses a relatively new method called ‘little shot learning’, where models start with a high general understanding of many different topics and then use much fewer, in some cases no labeled examples, to learn new tasks.
If traditional systems are like a fishing line that can catch a certain type of fish, then FSL is an extra net that can catch other types of fish as well.”
New Facebook AI Live
Facebook revealed that the new system is currently deployed and live on Facebook. The AI system has been tested to detect harmful misinformation about the COVID-19 vaccine.
It has also been used to identify content intended to incite violence or simply walk to the edge.
Facebook used the following example of harmful content that doesn’t stop at incitement to violence:
“Does this guy need all his teeth?”
The ad claims that the new AI system has already helped reduce the amount of hate speech posted on Facebook.
Facebook shared a graph showing how the amount of hate speech on Facebook decreased with each new technology implemented.
The graph shows Facebook’s success in detecting hate speech
Learning ensues in just a few shots
Facebook calls their new technology, Entailment Few-Shot Learning.
He has a remarkable ability to correctly label written text as hate speech. related research paper (Reasoning as a Learner of Short Shots (PDF) It reports that it outperforms other low-shot learning techniques by up to 55% and achieves an average of 12% improvement.
A Facebook article about the search used this example:
“…we can paraphrase an apparent input of sentiment classification and label pair:
[x : “I love your ethnic group. JK. You should all be six feet underground” y : positive] Below is the sample text requirement:
[x : I love your ethnic group. JK. You should all be 6 feet underground. This is hate speech. y : entailment]. “
Facebook is developing human-like artificial intelligence
The announcement of this new technology explained that the goal is human-like “learning flexibility and efficiency” that will allow it to evolve with trends and enforce new Facebook content policies in a rapid amount of time, just like a human.
The technology is early stage and in time, Facebook envisions it will become more sophisticated and widespread.
“A teachable AI system like the Few-Shot Learner can dramatically improve the flexibility of our ability to detect and adapt to emerging situations.
By identifying evolving and malicious content faster and more accurately, FSL has the promise of being a critical piece of technology that will help us continue to develop and address malicious content on our platforms.”
Read Facebook’s announcement of the new AI
Our new AI system to help tackle harmful content
Article about the new Facebook technology
Malicious content can evolve quickly. Our new AI system adapts to deal with it