Where to start: Machine learning 101 for human rights defenders

15 posts / 0 new
Last post
Where to start: Machine learning 101 for human rights defenders

Below is a list of questions to serve as a starting framework for the discussion in this thread:

  • What is machine learning and how does it work?
  • What is machine learning used for?
  • What are the benefits and risks of it?
  • What are tools and resources to get started?
What is machine learning?

In 1959, Arthur Samuel described machine learning as giving 'computers the ability to learn without being explicitly programmed'.
In other words: Computers were already used for the classification of data (text, pictures, geographical information, records of experiments, etc.) long before machine learning became a hot topic. Compared to machine learning, however, the big difference is that humans had to specify patterns to distinguish different categories. For example, documents about discrimination might also contain words like prejudice, harassment, racism, etc. Searching for these terms increases the probability of identifying discrimination documents that do not explicitly mention this word. However, it might also lead to documents of other topics as these word are also used in different contexts, e.g. harassment is highly associated with documents about sexual assault for which we might want to choose a different category, such as women rights. Manually defining good patterns for classification tasks is a very time-consuming and cumbersome process leading to highly-customized solutions.
The power of machine learning is that it uses statistics for the detecting of such patterns. By feeding lots of sample texts that are labeled with their corresponding category (e.g. discrimination, women rights, .. ) to a machine learning algorithm, it is able to learn which words or word combinations are relevant to identify a document belonging to the category discrimination.

resources to get started

Here is a really nice tutorial on machine learning by Vishal Maini: https://medium.com/machine-learning-for-humans/why-machine-learning-matt...
'Machine Learning for Humans - Simple, plain-English explanations accompanied by math, code, and real-world examples'

Do you know other good resources or tools to get started?

This is really good, thanks

This is really good, thanks for sharing it. I particularly like the bit on "Why machine learning matters", which states that "Artificial intelligence will shape our future more powerfully than any other innovation this century. Anyone who does not understand it will soon find themselves feeling left behind, waking up in a world full of technology that feels more and more like magic." I think it brings up the fact that we need to be careful, if we care about societies, of who is or will be left behind. (This may belong to a different topic on this online dialogue, so I'll reference it later)

I like the approach of

I like the approach of Distill (https://distill.pub/about/) It is focused on clarity and transparency. They even have a  prize for clarity in Machine Learning aimed at "recognizing outstanding work communicating and refining ideas in Machine Learning and adjacent topics"

Members tagged in this comment: 
understanding ML

This is really cool! I like their explorable explanations and think the prize for clarity is a great initiative.
Google's AI experiments (https://experiments.withgoogle.com/collection/ai) also offers lots of interactive games and visualizations to understand the concepts and methods in machine learning.
And there is the r2d3 visual introduction to machine learning: http://www.r2d3.us/visual-intro-to-machine-learning-part-1/

This one is really

This one is really interesting you are right! -> http://www.r2d3.us/visual-intro-to-machine-learning-part-1/

And this is in 10+ languages,

And this is in 10+ languages, pretty cool.

That's really cool!

Comment originally posted by Natalie

That's really cool!

Commercial and Open Source products

I have the sense that most practitioners will have a better chance to enter the ML space now that there are services, products, libraries and hardware available at realtively low cost. From Tensorflow(https://www.tensorflow.org/tutorials/) and DeepLenshttps://aws.amazon.com/deeplens/) to AWS Machine Learning (https://aws.amazon.com/machine-learning/) and Google's Cloud AI(https://cloud.google.com/products/machine-learning/), there seem to be a growing number of projects lowering the barrier to ML. How about a curated list of those resources? Does it exist already? 

Members tagged in this comment: 
Commercial and Open Source products

Comment originally posted by Natalie

A curated list of ML resources (both one for learning and another one for tools) would be very helpful.
Personally, for Natural Language Processing and ML in Python I use nltk (http://www.nltk.org/) and scikit-learn (http://scikit-learn.org/stable/). For Java there is Weka (https://www.cs.waikato.ac.nz/ml/weka/) available.
Has anyone used Accord.NET (http://accord-framework.net/)? And what are you working with, Adam, Micaela, Will?


Where to learn about ML?

What are good starting points to learn about Machine Learning. I know about Coursera (see https://www.coursera.org/courses?languages=en&query=machine+learning&use...), for example, that offers a wide number of specialized courses. What else is out there? How about for non-English speakers, where can they access content like this?

More tools for human rights defenders

Comment originally posted by Vivian Ng

Something else I have come across is the 'security in a box - digital security tools and tactics' by the Tactical Technology Collective and Frontline Defenders. I thought it might be useful to share as they were developed specifically for human rights defenders in the context of technological tools for human rights defenders.



ML Interpretability and Fairness

Comment originally posted by Natalie Widmann

Patrick Hall just published an 'awesome-machine-learning-interpretability' repository with papers, tutorials and software packages to build interpretable ML tools and to measure fairness: https://github.com/jphall663/awesome-machine-learning-interpretability

AI Pattern Language

Comment originally posted by Natalie Widmann

Another super interesting project is the Data & Society AI Pattern Language by M.C. Elish and Tim Hwang:

'we present a taxonomy of social challenges that emerged from interviews with a range of practitioners working in the intelligent systems and AI industry. In the book, we describe these challenges and articulate an array of patterns that practitioners have developed in response.'