Two Magic Boxes

People talk about Artificial Intelligence (AI) and Machine Learning (ML) these days a lot. I'll join the crowd as I also believe that's the future. However, the power of artificial intelligence would not be the point here. Rather, I'll focus on something natural to all human beings, our intuition, and how similar (arguably, of course) it is to neural networks (NN).

Deep neural nets are, perhaps, the most powerful predictive tools up to date, and are well known for producing highly accurate models but could be hard to interpret. You have a choice: a simpler ML model that is easily interpretable but only 90% accurate. Or, you can train a NN that makes 99% accurate predictions -- but shrug your shoulders when asked why you get certain results. Doesn’t it sound just like that “gut feeling” when you know you are right but can’t quite make it out? With a trade-off between interpretability and accuracy, the favor is often given to the former.

NN have been inspired by the human's brain. We still don't understand a great deal of the processes inside our heads but we do not distrust our senses. Well, except one.

We don't trust intuition. Personally, I don't trust my intuition.

Logic, our knowledge, and the five natural senses help us build judgments about the world the way most ML models produce their predictions, e.g. (Non)-Linear Regression, Decision Trees, k-means clustering, etc.

Neural nets may have the same input but are not as transparent with all their hidden layers, weights, and backpropagation. So assume that intuition works the same way: its logic is hidden, its predictions are accurate but hard to explain. That is to say, I am convinced that intuition is, in fact, uses our experience and observations but in a more subtle way. Intuition assigns its "weights" while we see something out of the corner of the eye without even realizing it; when we hear something without listening. Intuition learns without our knowledge.

As I mentioned, I am among those who think Neural Networks and Machine Learning in general will shape the future of AI.  Yes, magic inside a black box is attractive but scary. However, we embraced this "scary" reality by welcoming voice assistants and face recognition technologies when we first asked Siri/Cortana/Alexa/Alice what the weather forecast was, tagged our friend on a Facebook photo, and unlocked our iPhone with the Face ID... 

Maybe we should treat our intuition with more respect as well?

What is behind: Social Networks

Dear all, 

I know that many of you worry about your personal information and would prefer if it stays private. For the past three years, I have been doing research on social network analysis and have gained a bit of understanding how social networks function. Although cybersecurity is out of my scope, there are numerous ways of obtaining personal information; not all of them are illegal.

Just think about it: Why are there so many web-pages that tell who you were in 1917, or which actor you look like the most, or which of your friends is your personal angel/demon, or what you did in your past life, etc.? Some of them are simple — yes, stupid, but isn’t it fun to know which character you are in the Harry Potter stories. Others are more complex and are based on existing psychology tests — sure, I would like to know more about my personality, and it’s free! These tests/quizzes may seem fun and innocent, but they have a certain purpose apart from entertaining: they collect (note: not steal) personal information. In fact, they ask your permission to do so. Of course, you may say, Why not, this is only “to make the test more precise”, besides,  I have nothing to hide.

Yes, some personal information — including your name, schools you attended, your current location, past and present jobs, even photos — is accessible with a little effort if one needs to find it. However, these quiz-pages do not require any effort at all: they are very fast and efficient in collecting large volumes of data from thousands of people.

What happens next: well, who knows. It depends on who created the web-page: the test may be collecting statistical data for research purposes, or in order to shape targeted marketing strategies; it may be sponsored by the government and used for the sake of security, or, alternatively, by an organization for malicious purposes.

Big data is not just a fancy term; it is indeed a powerful tool. Some of you probably read the controversial article about how big data won the U.S. election (the original version was published in Das Magazin, Switzerland). One may agree or disagree with the article, but the methods of predicting/analyzing what people like and believe based on their activity on social networks do exist and are quite impressive.