THE BLOG

Do You Like People Watching? My Computer Does

10/10/2017 17:28 BST | Updated 10/10/2017 17:28 BST

2017-10-06-1507288009-9107936-pexelsphoto260907.jpeg

https://static.pexels.com/photos/260907/pexels-photo-260907.jpeg

Every day I wake up at around 6am. I go downstairs, put the dog on his lead and take him for a walk. We go to the same park just around the corner. The one with the old oak tree by the lake where he does his business as he always does. He loves that tree. Then I walk him back home, have a bowl of cornflakes, read the news and check my emails.

After that, I leave the house and go to the tube station. On my way, I visit the local café where I order a latte from the same woman I order my latte from every single day, the one that always wears her hair in a bun and never wears lipstick. I don't even need to tell her my order, she usually has it ready as soon as I get there - which is ideal, as it means I don't have to speak to anyone in the morning before I've had my caffeine fix.

I'm a creature of habit. Most of us are and our behaviour is predictable. In fact - and you may find this uncomfortable to read - human behaviour is 93% predictable.

So recently one morning, when the woman handed me my coffee, I noticed that her hair was down and she was wearing bright red lipstick. I didn't think too much about it but I did fleetingly think it was strange. It suited her. I forgot about it and went back to my routine. I had a train to catch.

Then guess what? A week later I saw her chatting to a good-looking bar man. My memory kicked in and I put two and two together. There was a new relationship in the air. How lovely.

Of course, it was only by chance that I noticed this change in the woman from the café, but you know who would have noticed this change instantly? A computer.

Changes in mood and behaviour can be subtle, so subtle that the brain often cannot detect these changes or save them in our memory. They often don't stick and if they do, they only leave a slight trace. A computer however sees it as a big black binary mark. They have total memory recall and can be designed to retain information, analyse, compare, predict and interpret data. Computers can be taught to find patterns, learn from them and alert someone when something changes. In short, they can detect anomalies in behaviour patterns. They can spot the lipstick!

So, what are these patterns of behaviour? And can a computer really predict what we are going to do?

We are predictable

"Similarity breeds connection," according to sociologists Miller McPherson, Lynn Smith-Lovin and James Cook in their paper Birds of a Feather: Homophily in Social Networks. Homophily means the tendency within social groups for similar people to connect on some level, across various relationships including marriage, friendship and colleagues from work. People of different genders, races, ages, classes, education and ethnicities all have different characteristics that impact who they choose to socialise with. To paraphrase Aristotle in his Nicomachean Ethics: "people love those who are like themselves."

I'm specifically interested in how this principle can help us to safeguard children online. This principle, whereby similar people come together and form social groups helps machine learning systems to detect anomalies in patterns of behaviour and analyse when a discrepancy has occurred. For example, if your 14-year-old Zoella-obsessed daughter starts chatting to an 18-year-old punk metal fan or if your 12-year-old Chelsea Football Club supporter son suddenly connects with a 16-year-old boyband super-fan from Sydney, it could see that something was out of the ordinary and potentially not right. When these changes occur and they don't make sense, computers are able to flag the connection and look deeper into the data patterns to see why this behaviour is breaking the norm.

The cadence of communication is important too; children and young adults tend to brag and banter with each other, which is a very distinctive pattern of communication. It helps us to pick up when people are posing as children and attempting to conduct grooming. Similarly, this cadence is well established during the build-up to a sext: it's a 'go on, go on, go on, send it' style of exchange - very quickfire, much like children daring each other to do something they shouldn't offline. The patterns of behaviour are distinctive and when statistical and probability analytics are overlaid, they can start to provide accurate patterns that can lead to predictable outcomes. For example, sexting is more common between the ages of 13 and 17 and increases with age. Boys receive more images than girls, exchanges commonly happen in a one-to-one situation and 89% happen at the beginning of a new relationship. The application of statistical analysis significantly reduces false positive outcomes.

But isn't this all common sense? Well, not always. We are frequently too busy to pick up on these things, or are simply trying to give our children space online to explore, have fun and be themselves. With children on average sending and receiving over 200 messages a day, it becomes impossible for a human to spot delicate and intricate patterns. Not so in the world of AI and machine learning.

Mum and dad, for example, can drop the need to snoop, pry and spy on their children as the robots do all the hard work helping to safeguard children whilst also ensuring their privacy. It is this latter point that I get excited about. Children have rights too and apps that disclose to a parent what their children are sending and receiving are not only intrusive and invasive but do the safeguarding industry damage. Children simply find ways to avoid and bypass them.

The computer will see you now

This principle of automated discrete behavioural patterning is well-established in medicine. For example, Andrew Beck from Harvard Medical school conducted some research on women with breast cancer. He ran their details through a machine learning algorithm to see if the AI could identify whether the given biopsy he provided was cancerous or not. This would then determine what course of treatment the patient should go on. The only extra information he gave the AI was the longevity of the patient, whether they died within a week or are still alive etc. The AI was able to identify 11 signs that the given biopsy was cancerous. However, the medical community only knew around eight of them, the other three signs were unknown to the human eye, having never been picked up by a human before.

Similarly, when we look at behaviour and patterns of behaviour, we can see certain traits that crop up. Machine learning can spot a troll, predator, bully and a groomer by observing, patterning and analysing, it can then flag if it suspects a potential threat. We know that statistically, comments on YouTube that are between three and eight words have a 72% chance of being abusive and we know that an increased number of capital letters, exclamation marks and the use of first person pronouns is synonymous with the language of a troll.

By analysing these patterns of behaviour, machine learning algorithms can spot any changes that go against the grain. Regardless of whether they're typical or not.

Let the robots lend a hand

The ability to track and analyse human behaviour is vital when it comes to being able to detect harm or potential harm. Computers can make sense of the confusing, emotive and sometimes scary online world. The algorithms can understand chaotic human behaviour and find patterns in linguistic traits, social media content and even likes on Facebook.

Using these systems, personalities can be patterned online, allowing for proactive and pre-emptive action to be taken to help reduce abuse and hate online, to spot that the quiet and shy girl is suddenly loud and flamboyant, or to flag cancerous cells before a human can even see them.

We all love to people watch. We poke our heads out of the living room window because we heard some people on the street or we sit outside a café watching people aimlessly wander by. We may notice that a person is wearing a green hat or is eating an apple, but we won't observe much more. We need our computers to watch alongside us, to learn and analyse information in order to help us understand things we never would have seen - so that we can blithely get on with ignoring the majority of the world while enjoying a latte.

Now, if you'll excuse me, the dog needs to find that tree.