As Europe is expecting a new law on artificial intelligence (AI), and many other countries are also considering similar laws, with Canada hitting the news most recently, I finished reading a book about AI that came to me highly recommended. It is a book by Kate Crawford, “Atlas of AI”.

This is great read for those who do not have big understanding of AI and want to get some basic understanding of the main concepts. The books helps to make it clear that AI is actually not “magical” or “all about robots”. Spoiler: it is all man made. For those who are already familiar with AI, the book has a lot of interesting research materials and history bits; and helps to fill some knowledge gaps. The books is a “simple read”, does not provide solutions, but leaves you wondering about where our society goes next.

Some ideas I had after reading the book

We aim to collect all the data. Not even considering if we need it

Techno-evangelists aspire to efficiency, speed, standartization, but not to diversity, complexity, interdependence. We want to gather all the data, calling it “new oil”, and strive to make our economy faster, more efficient, more digitalised. When “Google Street View” was collecting their data, it was notably said “anything you see in the world, needs to be in our databases”. “A collect it all” mentality is now normalised, it is seen wasteful not to collect everything.

and we see enormous data collections. Same issue arises when we collect data for adverting, or when we analyse user behaviour on the website. We collect it all, not even trying to find a way to achieve same results with less data, maybe achieve a less standard, a less efficient, but more unique result.

Problem is, in the world of data, no one wants to be first to gather less.

We should understand data sets

We often do not know which data is used to “train” AI. You may think you upload a picture on your personal blog, but alas, it is used to train an algorithm. Very often images are extracted right from the search engines. Some people say that data is a “natural source”, same as oil, waiting to be extracted. That is why privacy and AI discussions are often happening together. I know, we want to kill all the lawyers first, as Shakespeare’s Henry VI famously asked for; and legal protection of our data may not be perforce but is a great starting point to control the pipe of the data extraction.

Is a mistake worth it?

Book also talked about alarming examples of bias. Many say, it is okay to discriminate if we achieve greater good. Is it though?

I think the hiring examples are quite common, when women did not come recommended in the automated processes of hiring: since based on previous years of history, men usually were hired. Algorithm read it as “oh, men are better than females” and would exclude females, when gender was removed, algorithm would find other “feminine” attributes and made sure to exclude such candidates.

Book also talked about failed attempts to train AI to recognise faces, and AI not working on difference race, in different gender. Thus, in one research, it was shown that algorithm generally marked black women as “angry” more often, based on the way their face looked when they smiled or expressed a neutral emotion, since it did not consider different bone structure of the face.

This particular example was also interesting for me as an artist, since the art tutorials often try to break down emotions by providing similar simple tips – if eyebrows are up, the expression is surprised. Made me wonder if some of my art tutorials could also be biased?

Lines between public and private sector get blurry

We are facing the blurring of lines between private and public sector blur: Facebook and amazon now have so much data they may become of the great value to the governments. Book talks about examples of governments cooperating with tech giants; and its leaves us wondering: if data is a new gold, what will happen when most of this gold is left not with the governments but with the private companies. and speaking of governments, do we even trust our own governments to have all that data on us?