On a recent business trip to San Francisco and Silicon Valley, I saw fascinating examples of machine learning in action, which inspired me to write this blog.
Everybody’s been there, and I don’t mean on vacation
I visited the Bay Area along with 15 colleagues of mine. Californication by Red Hot Chili Peppers became something of my mental soundtrack on the journey, so I’ll let it play in the background here, too.
Firstborn unicorn
Google and Facebook are perhaps the best known and greatest ‘unicorn’ companies – startups that quickly became multi-billion enterprises. Both address a market demand, but they were not alone, nor the first to do that. However, they were able to outpace and replace the competition because they use machine learning effectively to serve their users better.
A teenage bride with a baby inside getting high on information
An article in Forbes reported that the retailer Target had identified a high school girl being pregnant before her parents knew, and – to her parents’ dismay – sent her coupons for baby cribs and diapers. Be that as it may, forward-thinking companies rely heavily on predictive analytics to decide about actions, for example on users that are likely to click, buy or churn. Facebook estimates it makes 2 to 3 trillion predictions every day.
Sicker than the rest, there is no test, but this is what you’re craving
The success of Silicon Valley’s tech icons relies on their ability to suggest and deliver content tailored to the users’ interests. For example, Netflix gets 100% of its revenues from subscriptions, so it must keep its users hooked month after month. To that end, it uses a highly sophisticated model for personalized video recommendations. And there is a test. Lots of tests. Netflix runs hundreds of A/B tests all the time to see what works and what doesn’t, both on the user interface as well as on the content.
Celebrity skin, is this your chin or is that war you’re waging?
Facebook users know that the site recognizes individuals from images. While that’s impressive as such, I found the video analytic capabilities even more fascinating. Markku Mäkeläinen showed us by examples how five years earlier, they could distinguish the shape of a human in a video, but now they can pick several persons at a time in a live stream. In addition, they interpret the persons’ bodies as vectors, which makes it much easier to analyze their positions and movements, and ultimately what they are doing on the video.
Hard core, soft porn?
Formerly, platform providers like YouTube, Periscope and Facebook relied on users to report disturbing content, but after some shocking incidents, legislators have required them to censor offensive material before it reaches any audience. Given that every minute, 500+ hours of video is uploaded to YouTube alone, monitoring would be impossible without machine learning.
Psychic spies from China try to steal your mind’s elation
The battle against trolls, bots and fake news is another example of internet companies using machine learning to protect their reputation and comply with regulations. Political players, lobbyists and foreign powers try to manipulate the moods and opinions of the public. Unsurprisingly, Facebook uses machine learning to combat fraudulent accounts and trolling, deleting millions of accounts per day. Fraudulent accounts may be easy to detect from traits such as mass postings, but fake news remains a wicked problem. They typically can’t be identified by key words, and users endorse and forward them, rather than report them as inappropriate.
And Cobain can you hear the spheres singing songs off station to station?
The internet behemoths aren’t meeting their growth expectations in existing markets, so they’ve decided to expand the market – which means bringing the internet to billions of new users by means of high-altitude drones, balloons, laser beams, solar powered base stations – and machine learning. On the other hand, the companies want even more revenue from wealthy first-world users. To fulfil their visions about VR, AR and intelligent traffic – which merit blogs of their own – the tech giants need next-generation 5G networks to spread fast. For example, Facebook sponsors the Telecom Infra Project, and opened the source of its own network research. We were told that the firm’s researchers had decimated the time and costs of building a 5G network. In their approach, they use commodity components to build base stations, as well as machine learning to lay a mesh network that uses existing buildings and light poles to host base stations, eliminating the need to build new masts.
Destruction leads to a very rough road, but it also breeds creation
And tidal waves couldn’t save the world from Californication
So how does that apply to Finland? Here, too, physical goods and services are replaced by digital ones. “Software is eating the world,” as Marc Andreesen put it. Customers expect quality, delivery times and prices that require advanced data processing.
So what can you do? Learn from the best. One way is visiting the Bay Area.
Dream of Californication
Dream of Californication
Dream of Californication
