12 May 2016
What you need to know about machine learning – part 4: how does machine learning affect your life?
Part 4: how does machine learning affect your life?
Now that we understand the key concepts of machine learning as an idea, it is inevitable that we should begin considering the possibilities for such a technology.
Those in the know already have…
Machine learning is all around us
The question is; exactly which parts of our day-to-day lives are already being directly effected by machine learning and algorithms?
The surprising answer is that machine learning and algorithms are currently having a direct impact on many aspects of our lives. Some of these impacts we are very much aware of, others we are less so, and some border on the Orwellian.
I’m sure most of us have already figured out that when we use websites with recommendation engines such as YouTube, Netflix or Amazon, every selection we make from the videos we like and dislike to how long we watch a film for and the kinds of products we purchase are all being monitored and recorded. Driven by machine learning, these sites are using this data to “recommend” or “suggest” other similar products, videos, or films that we might like.
Recommendation engines are an entirely benign use of machine learning in our day-to-day lives. We (generally) give our implicit consent to such uses for machine learning, and sometimes even find the results quite useful.
We here at Phrasee were recently introduced to a series called “BoJack Horseman”. The Netflix recommendation engine thought we might enjoy it. Turns out, it was right. Great show.
That learning machine deserves a pat on the shiny, metallic head.
*** Note: The same learning machine also keeps recommending “Fuller House”, but we’ll let that slide and assume the machine is still learning. ***
There’s also Siri, a machine learning speech recognition program designed to let us talk to the internet and interface with the data it contains without the awkwardness of a keyboard.
Siri’s deep learning algorithms have allowed the program to overcome our idiosyncratic speech patterns in order to understand us. In turn; Siri helps us access the information we want more efficiently with every passing interface (although that voice still creeps us out a bit).
“Siri, show us another example”
Our own machine learning tech, here at Phrasee, optimises the language of email subject lines to increase email marketing revenues, which has consistently been shown to outperform human writers by as much as 417%.
So, from waking up, turning on the TV, checking your email inbox and getting directions to that place we are supposed to meet our friend for brunch, the list of examples of relatively harmless, beneficial machine learning technology in our day-to-day lives goes on.
But there’s more.
As we move along the machine learning spectrum from the “benign and harmless” towards the “disturbing and insidious” end, we find that machine learning and algorithms affect us in ways we might not be quite so comfortable with.
“Targeted advertising” falls somewhere in the middle, inundating us with social media advertising which seems to know a bit more about our shopping habits and online activities than we are ready to completely accept.
When we log on to Facebook and see advertising for hotels in the same Italian city, which was the subject of that interesting article on frescoes we were just reading, it is no coincidence.
We are being watched.
Still, we’re just talking about a few hotel ads, right?
But what if, instead of our potential holiday destinations, learning machines had a major influence over, say, our pensions?
It’s already happening.
In the financial markets, the vast majority of trades are now controlled almost entirely by algorithms and learning machines. A few people are making a LOT of money off of this fact.
The bad news? That’s your money the machines are learning with. In all likelihood, your pension and retirement savings are currently in the hands of micro trading machines conducting millions of trades per day.
This has proven to be a highly effective strategy, except that time it didn’t.
In May of 2010, the Dow Jones Industrial Average plummeted 1,000 points in the space of mere moments, only to regain equilibrium about 20 minutes later.
This event became knows as the “flash crash“.
The culprit? A trader named Navinder Singh Sarao, with his own pesky learning machines, exploiting a glitch in Dow’s micro trading strategy.
Thankfully, humans were able to intervene and correct the problem before any permanent damage was done and we all had to work into our 80’s to clean it up.
Image credit: ABC
Then, there’s C.R.U.S.H.
In the vein of “Minority Report”, 2002’s sci-fi blockbuster in which psychics were linked up with computers to allow Tom Cruise to arrest people before they could commit “future crimes”, comes a learning machine that is changing the way society is policed.
CRUSH, or Crime Reduction Using Statistical History, is an algorithm designed to use predictive analysis to reduce crime in a given geographical area.
The CRUSH IBM predictive analytics program was designed as a pilot project in Memphis, Tennessee in 2005. Within its first three days in operation, the Memphis police department had set a new record for arrests in a single day, with 1,200 arrests.
Over the next 6 years the city’s crime rate fell by over 24% and police departments across the globe began to eye it with envy.
The concept was simple enough; use historical data on exactly where/when crimes have been committed in the past to predict where they are most likely to be committed in the future. Increase police patrols in these identified areas and then watch the arrests roll in.
This approach has raised concerns with civil rights groups, which claim that in many ways it can become a self-fulfilling prophecy that perpetuates misery in historically underprivileged areas with populations who have been disproportionately targeted by police.
Like that ever stopped them…
CRUSH is now rumoured to be used by dozens of police departments world wide, including 2 in the UK, although most will not acknowledge using the program publicly.
In addition, parole boards in more than half of all US states use predictions founded on data analysis as a factor in deciding whether to release somebody from prison or to keep them incarcerated.
If this approach to decisions about release from prison produces similarly encouraging results, look for it to be viewed as a new gold standard and spread across the globe as well…
So buckle up.
The machine learning revolution is already well under way. It is all around you. The machines know where you live, they know what you like and they know what you want.
They might even know it better than you do yourself.