Deepmind, AlphaStar and Cake – getting messy with Material Semiotics
A recurrent theme that’s popping up a lot in this blog is CONTEXT. Understanding how things physical and philosophical fit together is crucial to a nuanced understanding of the world and the subsequent creation of art that has something to say.
This week we were introduced to Material Semiotics through the work of Donna Haraway, rock star philosopher of computational art, feminist technoscience and speculative fabulation.
Her methodology for Material Semiotics gives artists a practical set of questions that can help “untangle” the webs of relations related to a topic. We were shown a film in which she explains the topic using metaphor; a bundle of string (entanglement) and a messy cake (showing we can’t escape getting involved with a topic, getting messy with it, and getting enmeshed with it).
Most important – material semiotics is grounded in empirical case studies. It requires a case study for the framework to be applicable. As an example of how it might be useful in practice, I’ll take a recent story regarding developments in AI as the case study to “untangle”.
In late 2018, professional StarCraft Player Grzegorz “MaNa” Komincz was invited by Google’s Deepmind research team to London to vs their latest AI, “AlphaStar” in a best of 7 competition. Starcraft is a real-time strategy game, played on a virtual map with limited space and resources. Victory is secured by destroying all of the other player’s building structures.
The game is the latest challenge for the Deepmind team. Previously they had developed their AI to play – and ultimately defeat – world champions of both Chess and Go. Starcraft was seen as the next big test for them as it’s a game with exponentially more variables in play than Go.
Needless to say, Deepmind was victorious, and dominant throughout. So with this example, let’s have a look at this even with the questions supplied by Material Semiotics:
What is going on here?
The training of an AI agent to learn a complex computer game well enough to beat top human players.
Which are the agents implied?
Agents – the AlphaStar AI, which actually consists of millions of smaller AIs that take on different strategies and methods and are then put into competition against each other millions of times. The human player. The developers of the Alphastar AI algorithm. The corporate actors who own both DeepMind, and Starcraft. The well-known personalities (“Casters”) of Starcraft who were employed to front the live stream of the games.
The real: Deepmind headquarters, London. The room in which MaNa plays, and the adjoining meeting room in which Deepmind’s staff observe the match progress. The home of MaNa where the pre-tournament practicing took place. The locations of millions who tuned in to see the live stream. The network infrastructure linking up the thousands of CPU and GPU’s powering both the AI of AlphaStar and the Starcraft game.
The virtual: The virtual training ground for AlphaStar’s myriad sub-agents to do battle within. The virtual map of StarCraft’s game area. The “probability maps” generated by Alphastar to aid it decide on the correct move to play next.
The moments of the final series being played. The thousands of hours played by MaNa to elevate his level of play to world-class. The millions of virtual hours of competitions as Deepmind’s AI’s battled against each other, a high speed evolutionary process. The decision time of around 100 milliseconds it would take Alphastar to calculate the probability of millions of potential next moves, and decide on the optimal one to play. The near future as AI discourse evolves to reflect this new reality (see point below). The distant future when we all are wiped out at the barrel of a laser rifle wielded by a robot (speculative).
Something interesting related to the “Whens” aspect of this – human players have already observed the strategies devised by the AI and are now incorporating it into the way the way that they approach the game. It’s fascinating to watch the effect of Alphastar’s “intervention” into the existing world of Starcraft, as it provides us with a real example of how a technological (literal) game change may change society in the future.