Contents
Goals
- Understand how AI applies learning from interaction with humans.
Steps
1. Individual play for 15 minutes (20 minutes in an online setting).
2. After playing 20-30 fast rounds, the players share their results.
Reflection
- When did the human players realise how the AI first explores different strategies and subsequently adopts a successful strategy?
- What was your strategy?
- Is Menace intelligent? What is a simple AI process?
- How does it learn from you? What makes it more intelligent?
- Would AI be able to replace (your) human action?
- How much do AI models in general rely on human real life experiences?
- Where are the necessary large amounts of data derived? Communication platforms, Facebook, Twitter, social media, picture databases, fitness apps, transport apps, other digitally mediated interactions?
Reference
- Menace was created by Matthew Scroggs. https://www.mscroggs.co.uk/menace/
- Mayer-Schöneberger, V. (2015). What is Big Data? Accelerating the Human Cognitive Process. In: Zimmermann, N. (ed.) (2020). The Internet, Big Data and Platforms. Digital Transformation in Learning for Active Citizenship. DARE Democracy and Human Rights Education in Europe, Brussels. https://dttools.eu
Experience
The game provides a relatively easy to understand experience on AI learning. It might be a good basis for a discussion on the dependence of AI to harvest data from human reality, for example, from communications, facial recognition databases, tracking, etc.
Variation
By choosing different modes, e.g., the “professional mode”, or “AI against AI”, one can see how the machine learning process adapts to different opponents.
If all are in an online setting, the game can be accompanied by conversations, in in-presence meetings participants can play in pairs with a strict focus on the game.
Exkurse Algorithms and strong and weak AI
The tasks above also raise the question of what machine intelligence actually is. First of all, the tic-tac-toe programme is steered by an algorithm. Algorithms are a set of instructions for computers, letting them conduct various tasks, as opposed to processing limited calculations. The algorithms respond to the input and might even apply themselves. Better hardware and more complex programming allows them to model complex situations and even behaviour.
Artificial intelligence requires these kind of complex, learning algorithms. However, one can distinguish between strong and weak AI. Strong AI works like a learning system, increasingly applying its routine, gaining insight through various data from different contexts and responding to its changing environment – the ideal of strong AI is deep learning in the way of the human brain: find information, connect these, come up with insights and new solutions (independently).
Most AI systems, although, also use machine learning. However, since they do not learn deeply they are recognized as weak. They heavily rely on the algorithms humans program, improving themselves in a pre-determined way. Even if the tic-tac-toe game were to be played 1,000 times, this would not lead the underlying algorithm to invent a new game. At this point, it learns only to improve its strategy inside the rules of the game. They appear to be intelligent.
Big Data is a method of gaining insight on the basis of quantitative data by building statistical correlations and relations. It requires a variety of data types and a massive amount of data (=big) for modelling social reality (with intelligent algorithms) through statistical approximation (compare Mayer-Schöneberger, 2015).
Inspiration: Board game “Mensch Maschine”
The board game Mensch Maschine aims at better understanding how deep learning AI works. Based on traditional “pawn chess”, up to five players can experience how the machine's learning progress increases round by round – and thus understand how human thinking differs from the way the machine works. The board game in a German language version can be downloaded at: https://www.wissenschaftsjahr.de/2019/jugendaktion/
