This guest column is by Pavel Cherkashin, Director at GVA Capital
If you’re over 30 years old and are honest enough to admit that you watch TV, I want you to think back before streaming video went mainstream. We were all in an unspoken competition for the biggest DVD collection of movies.
The winner was not the owner of the largest collection, but the one who could memorize the best scenes and give the best recommendations on what to watch. You give great movie advice 2 or 3 times, and you’re a guru. Give a few bad pieces of advice – and your friends think you have bad taste.
When BitTorrent allowed billions of people to access the world of free content, too much content quickly became a time consuming, painful problem. You can buy another Terabyte drive, but you cannot unsee that crappy video or return a spoiled night.
Now those Terabyte drives have been replaced with Popcorn Time and TV streamed through Netflix and a slew of other services. But the problem of choice is still there, and it’s costing us hours of valuable time. Meanwhile, the content hoarding is still ingrained in us and Hollywood is wise to it. TV and movie content is being produced quicker than ever before. In 2009, there were 211 primetime scripted television shows in the US. In 2015, there were over 400. Add to that anything outside of prime time and you’ve literally got thousands of TV shows alone, not including movies.
Do We Really Want Freedom Of Choice?
Tell me you haven’t spent way too many minutes searching for something good to watch. Research shows the public is searching for over 800 hours a year to find the next thing to watch.
It’s the classic “my-wife’s-wardrobe-dilemma”. More clothes make it harder to choose what to wear tonight. The more TV and movie choices Hollywood produces, the more time we waste watching trailers deciding what to watch.
There are 2 possible solutions to this problem.
1. The most obvious conclusion: the need for freedom of choice is an illusion.
Reduce freedom of choice and consumers will be happier watching whatever is offered. Back to linear programming, like in the early days of television, when you only had few buttons to choose from, but more personalized
Linear TV Versus Freedom Of Choice
HBO still runs most of their on-demand content in the old programming format. Once per week a new episode in a series comes out, everyone has to wait for it. Netflix, Hulu and the like are releasing many show’s a season at a time, allowing people to binge watch shows.
Although many like to have shows released seasons at a time, the choice paralysis is real. We like having options until we get to our couch and waste 20 minutes on reading synopses and watching trailers. The truth is people want to be programmed, to be told what to watch. Marvel’s superheroes, hobbits and New Mexican drug dealers are all turning into weekly programming that resembles soap operas.
The future of linear programming is personalization. It won’t be the opinion of the crowd deciding what I should watch next, but a neural network carefully analyzing my previous interests and views to predict what I want to watch next.
These challenges create an entire new opportunity for innovation. Indexing, rating, and recommending of TV and movie content across multiple platforms, websites, companies and networks is the new era that’s just starting to be explored in Silicon Valley.
This will not just be episode after episode of TV show content.
Because of “free” content delivery technologies like Torrents, Popcorn Time, Kodi, and other innovations, the TV industry will be pushed to make content available across platforms. But that’s not enough.
Imagine asking your TV to “Show the funniest boss jokes from all my favorite movies in the last century”. Humanity is speeding up, consumers have shorter attention spans and want to be entertained faster.
Watching a whole episode for a couple of good jokes is an unaffordable luxury.
Somebody has to teach the neural network to understand all videos frame by frame, word by word, to be able to pull up the right clips and group them together in new linear programs.
“Show me all the episodes where Angelina Jolie is naked. No problem, it will also show other similar looking actresses in similar scenes”.
“Remember that quote from Terminator about your clothes and motorcycles? Show me similar phrases in classic westerns of the mid century.”
There is some innovation around indexing sound, dialogue and scenes of video content like Ooyala, but that’s being tackled from an advertising frame. Soundhound promises meaningful conversation with TV with their speech-to-meaning technology and home appliance integration. Cross platform, by scene video recommendation engines aren’t here yet, but will be soon.
Now, Google doesn’t know much about the video on the web. The entire video, plus sound, category, locations and dialogue will all need to be indexed. This can’t be done manually. If a few great developers put their heads together they could solve this in a few months.
People innovating in Neural networking and AI space will be the future of this industry, as James Crowder points out in this TechCrunch piece. Technologies like Siri, Echo and Houndify will provide the interface to this newly indexed content and the future of home video.
Neural Networking Video Is Here, here’s a few pieces of tech that are working now.
Automatic Audio Indexing
Speech And Dialogue Research Group are using technologies that apply Context-Dependent Deep-Neural-Network to real life audio indexing across various sources. Through speaker independent transcription of phone calls, with 7 layers reduces word error as much as one third.
Using traditional class systems for indexing, entire volumes of subtitle or searchable audio content are being re-created. Here’s an example of a class system.
Automatic Video Indexing
Terrance Broad from Goldsmith, University of London has taken auto encoding to the next level. He’s reconstructed the movie Blade Runner to autoencode each frame from a downsized further compressed video using a convolutional autoencoder neural network.
His work shows the implementation along with a learned similarity metric. It’s capable of distribution of natural images, and reconstructs the videos sequences by passing each frame through the encoder and re-sequencing the output in order.
In the book, Multimedia Database Retrieval, video indexing innovation is constructed as adaptive video retrieval, which highlights where innovation is starting in the news, and moving towards any video content.
In most video retrieval systems, content is broken up into a few small segments called shots, and indexed by context, and other tags. Since this has proven not as effective as the tech we’re talking about, new ways to adaptively index are being created.
Through embedding representations of dynamic content in video, we will be able to search, index and call almost any piece of information covered in video.
This image shows how this will revolutionize tagging and searching (thus recommendations) across all video content.
How It Works Like Your Brain
When I say “That scene with Stallone and Schwarzenegger fighting in the desert”, it only takes a few bytes of information to transmit, but it triggers a lot of images in your brain. Imagine the same, but working with Neural networks.
You could transmit and re-create any content on the fly with very little information. It’s not only finding meta information with indexing, it’s actually producing new content by neural networks.
It will change the video industry and Hollywood forever. It will give new life to old content, usher in a new approach to linear entertainment programming and bring people back to TV again.
As an investor, it would be a dream to be a part of the new wave of home video technologies. If you know of anyone tackling this problem, have them send me a tweet and let’s chat. I’d love to get involved before it becomes a part of history.