If you frequent Facebook or other online platforms you know conversations can get out of hand quickly. Facebook is hoping to use an AI program to keep things calm. The social media platform is testing a machine learning program that would detect fights in Facebook’s Groups and report it to the group admins so they can intervene.
Facebook will be rolling out a number of new tools to help the 70 million users who run and moderate groups on the social media platform. Facebook has 2.85 billion monthly users, 1.8 billion participate in groups each month and there are tens of millions of active groups on the platform.
Along with these other tools, AI will decide when to send out a “conflict alert” to those admins and moderators who maintain groups on the platform. When the AI finds a conversation to be “unhealthy” or “contentious” the AI will send out a message and the human administrators of the group can decide from there what to do.
Social media platforms have increasingly relied on AI to filter what users see and to help remove hate speech from their platforms. However, AI has not always treated users fairly when left to its own devices. So, Facebook is hoping this new combination an AI watch dog and allowing the human administrators of the group to make final decisions will be a more fair way to handle things.
Twitter has pretty much abandoned an image-cropping AI after discovering that automated system was biased. Users had complained it had a preference toward showing pictures of white people in previews of tweets.
A blog post from Rumman Chowdhury, software engineering director at Twitter, stated that Twitter concluded the AI was biased after testing it for both gender and race biases. The post and research essay attached detailed that the cropping system when tested on random images demonstrated a preference for white people over black people and men over women.
Chowdhury stated that their conclusion is some tasks are best left to people and are not a good task for AI.
The research conclusions come months after the company said it would investigate the issue with its AI image cropping program.
Twitter, in March, started testing a new method to show a full image, rather than auto cropped previews, on mobile devices when a user tweeted a single image. The program would auto-crop the preview image people saw before clicking the preview for the full image. However, the program was prone to errors.
Now they rely on saliency which requires knowing what area of the image the eye is drawn to first.
There’s a new personal robot ready to help you. It’s called Ballie and is made by Samsung. It was introduced at CES2020.
It is a small rolling robot that reacts to your needs. In the video, you can see some of the capabilities. From waking up and opening your drapes, helping you with your exercises, your pets, connecting its camera through an app, turning on the TV for your pet, knowing to clean the floor with the vacuum.
MIT researchers have created an AI that can do your knitting for you! Create your own design and have it made for you. Even if you have never knitted a stitch before, with this system you can create hats, gloves, or whatever you want. Even patterned designs. You can create a template for the computer to follow.
The only missing ingredient is the joy you can get from making each stitch by hand…
AIs are supposed to be ever learning. However computer vision can be a different story. The Verge has an interesting article that shows some pictures and what AI sees it as.
In the article are a lot of pictures labeled with what the AI says it is vs. what it really is. In one group they are showing 4 different pictures of a dragonfly but the AI doesn’t identify any of them as a dragonfly.
It is important to get computers to see what is really there especially in the realm of self-driving cars. We don’t want them to see stop signs where there aren’t any or vice-versa. In another article on The Verge, it was reported that Google’s AI saw a rifle and not a turtle. If this had been something where they were looking for adversarial attacks, this could be a problem.
With all the copyright laws out there, if you need music for videos, here’s a site where you can generate AI music. At Jukedeck, you can choose genre (instruments), moods, and length. They have also have a list of pre-made downloads. Every track is unique and royalty free.
If you set up an account, you can save your downloads for future use. This was originated at Cambridge University by a team of composers, producers, engineers, academics, and machine learning experts.
The downloads are free for small businesses or individuals if you give them credit or $0.99 per download without credit. For businesses with 10 or more employees the cost is $21.99 per download. If you need to buy the copyright, it is $199 and you own it outright.
Here are some samples: