Web Log: MIT works out how predict the future
Sophisticated algorithms parse videos and create simulated footage based on analysis
MIT’s predictive video: The algorithm can create simulated footage of what it deduces to be likely to happen next – but only for up to 1.5 seconds into the future
An algorithm to predict the future
You’ll never guess what happens next . . . but new software from MIT will, because researchers at the Computer Science and Artificial Intelligence Laboratory have created a deep learning algorithm that can predict the future. Sort of. After being “trained” on footage from two million videos, the algorithm is learning to predict what actions the subjects in the videos might take next.
The algorithm can create simulated footage of what it deduces to be likely to happen next but only for up to 1.5 seconds into the future. This is done using what is known as “adversarial learning”; one algorithm does the guesswork and creates new video footage while a second algorithm judges this against real video footage, helping it to become more accurate over time.
Because past behaviour is predictive of future behaviour, once the algorithm has fully understood how elements are behaving in a given scene, it can give an educated guess as to what may happen next. Applications include generating “fresh” scenery footage based on video of waves lapping against a seashore or people walking upon grass on a golf course.
Amazon tightens up online review rules
In an effort to discourage false product reviews, Amazon has changed the nature of its review submission process. Amazon users are now limited to submitting five reviews per week unless the reviews are attached to Amazon verified purchases. There are also exceptions to this rule if it is suspected that certain products are suddenly being flooded with new reviews.
“If we find unusually high numbers of reviews for a product posted in a short period of time, we may restrict these products to Amazon verified purchase reviews,” says Amazon, on the updated customer review creation guidelines section of its website.
These new rules follow events earlier this year where Amazon started suing sellers who were found to have purchased fake reviews. According to tech industry news website Techcrunch, Amazon has sued more than 1,000 individuals since 2015 who were found to have posted fake reviews in exchange for payment.
Associated Press wades in on ‘alt-right’ usage
A new blogpost by John Daniszewski, vice president for standards at Associated Press, has tackled the need for clarity around the fast-growing use of the term “alt-right”. We are seeing mentions of “alt-right” in our social media feeds and across online publications and there is a concern that it is a sanitised synonym for racism.
Daniszewski says the term is “a name currently embraced by some white supremacists and white nationalists to refer to themselves and their ideology”, adding that it should be used in quotes with the option of the modifiers “self-described” or “so-called” as well as a definition that makes clear what this extreme group is.
In a world where we are becoming accustomed to hearing about post-truth and fake news, tech-savvy, social-media engrossed media consumers may enjoy the clarity AP brings to this. “The term may exist primarily as a public-relations device to make its supporters’ actual beliefs less clear and more acceptable to a broader audience,” Daniszewski warns.