Explainable AI

With ever more powerful and ubiquitous artificial intelligence (AI) questions arise about how these inscrutable black-box systems arrive at their decisions, which can have serious legal or health-related consequences, such as in autonomous cars or tumor recognition.

In 2016 the US Department of Defense (DARPA) started an initiative on explainable AI (XAI) to investigate how to answer these questions. In May 2018 the EU’s General Data Protection Regulation (GDPR) will come into effect, with especially article 21 “Right to object” and article 22 “Automated individual decision-making, including profiling” giving subjects the right to object against ‘any decision based solely on automated processing[, meaning AI], […] which produces legal effects concerning [the subject] or similarly significantly affects him or her.’ (cited from article 22)

In the light of these initiatives to investigate and to regulate the usage and misuse of AI, we will have a look at and discuss both the technical concepts and solutions at play, and the regulatory law and how to devise it such that it works to the greatest benefit of all.

Therefore, please read at least the material of level 1 to prepare for our next meeting on Wednesday 31st January 7:30pm 2018 in Tunnel Vienna, Florianigasse 39, 1080 Vienna.


Level 1

Read this very accessible article on the topic by the the New York Times.

Level 2

Also read these articles critical towards the GDPR and the DARPA viewpoint, and have a look at the Wikipedia articles on XAI and the Right to Explanation as attempted to be installed in the GDPR.

Level 3

Dig deep into the legal framework of the EU’s GDPR (especially Recital 71) and see for yourself how this affects future AI development.


Picture credit: https://tctechcrunch2011.files.wordpress.com/2015/11/humanrobotoverlap.jpg?w=738

Prisoners Dilemma Tournaments

The next meeting will take place on the 15.11. at 20:00 at the Tunnel Vienna.
The topic this time is going to be: Prisoners Dilemma Tournaments

Please have a look at the literature provided:

1)  Definition and Basic Game Theory of the Prisoners Dilemma


This website contains a lot of information on game theory in general and gives an introduction. You can read the linked page but also proceed to further topics or clarify certain terms in previous basic chapters. In the NWC however we will focus on the concept of the Prisoners Dilemma.

2) Expand the concept: Iterated Prisoners Dilemma

The case becomes interesting once the Dilemma is iterated several times and/or played among different people using their own strategy. The optimal strategies in such tournaments lead to different interpretation and have significance in many fields. Here are 3 articles presenting different successful strategies. All shortly review what the Prisoners Dilemma is ( you might want to skip these parts )

Tit for Tat and altruism:
Zero Determinant Games and the evolution of cooperation:
Tribalism and the importance of communication:

Additionally: These are some online Simulators to check the payoff of your own strategies.  Hope you manage to navigate through the Interface.


3) Sourcecodes

Stanfords Plato:https://plato.stanford.edu/entries/prisoner-dilemma/

A generalization of the IPD is allowing strategies chosen due to probabilities. The so called Stochastic Iterated Prisoner Dilemma has quite a mathematical body (see also zero determinant games). Take a dip with google!

Additional historic information on IPD Tournaments:


If you proceed on this page with NEXT you will find all the different strategies together with their code which took stage during Axelrods first IPD Contest.

Social implications of AI

The next and last meeting for this semester will take place on the 30.6. at 18:00 at the Tunnel Vienna.
Since we probably don’t see each other till October, we figured it would be nice to have a “home exercise” for the summer (like last year). Therefore I’d like to ask you for suggestions. If there is any literature you think is interesting, let us know. On the next meeting we will then choose together which one we want to read over the summer. Besides that we also will send out a collection of all suggestions.

For our this times meeting Nino prepared the topic Social implications of AI. Please have a look at the literature provided:

Kurzgesagt Video:
CGP-Grey Video:
Wait-But-Why Blog Part 1&2:
Paper on AI-Safety:
“The focus of the discussion should not be if and how a general AI is technological feasible but on the social implications of automation and the rapid improvement of AI’s”

P vs. NP Problem

At our next meeting on 28th November we will discuss the P vs. NP problem of complexity theory presented by Thomas Zauner. There are several levels of readings.

Level 1
Watch this video explaining the basics about the P vs. NP problem and complexity theory.

Level 2a
Read this article explaining not only the concepts, but also shedding some light on recent developments in research.

Level 2b
Watch these excellent and well explained lecture video on an introduction complexity. 
(For further studies see: https://www.youtube.com/watch?v=eHZifpgyH_4)

Level 3
For further reading consult the always-insightful Stanford encyclopedia of philosophy.