AI is Being Used to Create Fake Celebrity Porn, Because of Course it Is

By Tom Pritchard on at

AI is one of the new big things in the tech industry, with an immense amount of hype surrounding the concept of machine learning and utilising it for the greater good of mankind. And then there are the people using AI to make more convincing fake celebrity porn – using AI to stitch celebrity faces into porn videos with greater accuracy than ever before.

The tool was developed by Reddit user deepfakes, using publicly accessible tools and a face swap algorithm that he developed himself. While the final result isn’t perfect, the tool can pull images from YouTube clips and Google and use AI to create a somewhat convincing video of a particular celebrity.

It doesn’t have to be a celebrity either. It could be a random person you saw on social media or someone you know. Like the ridiculously accurate lip-syncing technology we saw earlier this year, the prospects of how this might be used in future are horrifying. Particularly since The Next Web points out that deepfakes’ algorithm is mostly a combination of readily-available tools like TensorFlow and Keras, and it wouldn’t take a genius to figure out how to stick them together.

So far deepfakes has produced fake GIFs featuring Gal Gadot, Emma Watson, Aubrey Plaza, Maisie Williams. They’re obvious fakes, with glitches occurring throughout the clips, but they look advanced enough to make this a pretty scary prospect for the future. Particularly since AI researcher Alex Champandard told Motherboard that these clips could take a consumer-level graphics card a few hours to produce. Apparently a CPU could also do the job, but it would take days to complete.

Obviously an algorithm would need a plentiful supply of images to stitch something basic together, which might not be as feasible for non-public figures. But given all the high-profile hacks over the past few years, coupled with the abundance of pictures some people love to post on social media, the source material might not be as difficult to find as you’d hope. [Motherboard via The Next Web]

New Robots Imagine Future Actions To Figure Out How To Manipulate Objects They’ve Never Encountered Before

Credit: UC Berkeley

By Alton Parrish | Ineffable Island

UC Berkeley researchers have developed a robotic learning technology that enables robots to imagine the future of their actions so they can figure out how to manipulate objects they have never encountered before. In the future, this technology could help self-driving cars anticipate future events on the road and produce more intelligent robotic assistants in homes, but the initial prototype focuses on learning simple manual skills entirely from autonomous play.

Using this technology, called visual foresight, the robots can predict what their cameras will see if they perform a particular sequence of movements. These robotic imaginations are still relatively simple for now – predictions made only several seconds into the future – but they are enough for the robot to figure out how to move objects around on a table without disturbing obstacles.

Crucially, the robot can learn to perform these tasks without any help from humans or prior knowledge about physics, its environment or what the objects are. That’s because the visual imagination is learned entirely from scratch from unattended and unsupervised exploration, where the robot plays with objects on a table. After this play phase, the robot builds a predictive model of the world, and can use this model to manipulate new objects that it has not seen before.

The robot that knows its future

“In the same way that we can imagine how our actions will move the objects in our environment, this method can enable a robot to visualize how different behaviors will affect the world around it,” said Sergey Levine, assistant professor in Berkeley’s Department of Electrical Engineering and Computer Sciences, whose lab developed the technology. “This can enable intelligent planning of highly flexible skills in complex real-world situations.”

The research team will perform a demonstration of the visual foresight technology at the Neural Information Processing Systems conference in Long Beach, California, on December 5.

At the core of this system is a deep learning technology based on convolutional recurrent video prediction, or dynamic neural advection (DNA). DNA-based models predict how pixels in an image will move from one frame to the next based on the robot’s actions. Recent improvements to this class of models, as well as greatly improved planning capabilities, have enabled robotic control based on video prediction to perform increasingly complex tasks, such as sliding toys around obstacles and repositioning multiple objects.

“In that past, robots have learned skills with a human supervisor helping and providing feedback. What makes this work exciting is that the robots can learn a range of visual object manipulation skills entirely on their own,” said Chelsea Finn, a doctoral student in Levine’s lab and inventor of the original DNA model.

Credit: UC Berkeley video by Roxanne Makasdjian and Stephen McNally

With the new technology, a robot pushes objects on a table, then uses the learned prediction model to choose motions that will move an object to a desired location. Robots use the learned model from raw camera observations to teach themselves how to avoid obstacles and push objects around obstructions.

“Humans learn object manipulation skills without any teacher through millions of interactions with a variety of objects during their lifetime. We have shown that it possible to build a robotic system that also leverages large amounts of autonomously collected data to learn widely applicable manipulation skills, specifically object pushing skills,” said Frederik Ebert, a graduate student in Levine’s lab who worked on the project.

Since control through video prediction relies only on observations that can be collected autonomously by the robot, such as through camera images, the resulting method is general and broadly applicable. In contrast to conventional computer vision methods, which require humans to manually label thousands or even millions of images, building video prediction models only requires unannotated video, which can be collected by the robot entirely autonomously. Indeed, video prediction models have also been applied to datasets that represent everything from human activities to driving, with compelling results.

“Children can learn about their world by playing with toys, moving them around, grasping, and so forth. Our aim with this research is to enable a robot to do the same: to learn about how the world works through autonomous interaction,” Levine said. “The capabilities of this robot are still limited, but its skills are learned entirely automatically, and allow it to predict complex physical interactions with objects that it has never seen before by building on previously observed patterns of interaction.”

The Berkeley scientists are continuing to research control through video prediction, focusing on further improving video prediction and prediction-based control, as well as developing more sophisticated methods by which robots can collected more focused video data, for complex tasks such as picking and placing objects and manipulating soft and deformable objects such as cloth or rope, and assembly.

Read more great articles at Ineffable Island.

Facebook’s New Suicide Detection A.I. Could Put Innocent People Behind Bars

(Activist Post) Imagine police knocking on your door because you posted a ‘troubling comment’ on a social media website.

Imagine a judge forcing you to be jailed, sorry I meant hospitalized, because a computer program found your comment(s) ‘troubling’.

You can stop imagining, this is really happening.

A recent TechCrunch article warns that Facebook’s “Proactive Detection” artificial intelligence (A.I.) will use pattern recognition to contact first responders. The A.I. will contact first responders, if they deem a person’s comment[s] to have troubling suicidal thoughts.

Facebook also will use AI to prioritize particularly risky or urgent user reports so they’re more quickly addressed by moderators, and tools to instantly surface local language resources and first-responder contact info. (Source)

A private corporation deciding who goes to jail? What could possibly go wrong?

Facebook’s A.I. automatically contacts law enforcement

Facebook is using pattern recognition and moderators to contact law enforcement.

Facebook is ‘using pattern recognition to detect posts or live videos where someone might be expressing thoughts of suicide, and to help respond to reports faster.’

Dedicating more reviewers from our Community Operations team to review reports of suicide or self harm. (Source)

Facebook

Facebook admits that they have asked the police to conduct more than ONE HUNDRED wellness checks on people.

Over the last month, we’ve worked with first responders on over 100 wellness checks based on reports we received via our proactive detection efforts. This is in addition to reports we received from people in the Facebook community. (Source)

Why are police conducting wellness checks for Facebook? Are private corporations running police departments?

Not only do social media users have to worry about a spying A.I. but now they have to worry about thousands of spying Facebook ‘Community Operations’ people who are all to willing to call the police.

Our Community Operations team includes thousands of people around the world who review reports about content on Facebook…our team reviews reported posts, videos and live streams. This ensures we can get the right resources to people in distress and, where appropriate, we can more quickly alert first responders. (Source)

Should we trust pattern recognition to determine who gets hospitalized or arrested?

Pattern recognition is junk science

A 2010, CBS News article warns that pattern recognition and human behavior is junk science. The article shows, how companies use nine rules to convince law enforcement that pattern recognition is accurate.

A 2016, Forbes article used words like ‘nonsense, far-fetched, contrived and smoke and mirrors’ to describe pattern recognition and human behavior.

Cookie-cutter ratios, even if scientifically derived, do more harm than good. Every person is different. Engagement is an individual and unique phenomenon. We are not widgets, nor do we conform to widget formulas. (Source)

Who cares, if pattern recognition is junk science right? At least Facebook is trying to save lives.

Wrong.

Using an A.I. to determine who might need to be hospitalized or incarcerated can and will be abused.