Digit Geek
Starcraft
Digit Geek > Recent Articles > Culture > AI now closer to mastering StarCraft

AI now closer to mastering StarCraft

After beating the best Go player in the world, Artifical Intelligence’s next goal is to master classic RTS StarCraft, and there’s been progress.

The world was taken by surprise when Alphabet’s DeepMind was able to best humanity’s best Go player. Understandably so, this feat wasn’t expected to be achieved any time soon, in fact, what experts predicted was decades off! This should give you a good idea of how rapidly artificial intelligence is progressing.

The DeepMind division then announced that its next goal in gaming was to master StarCraft, the classic PC RTS game. One that has been a staple in competitive eSports for a while now. It was a race now, with even Facebook joining the fray, making its framework open-source so that developers could make use of it’s AI toolkit in their attempts to master StarCraft with AI.

The latest entrant is a China-based team from Alibaba who have published a paper reporting their own AI software, BiCNet. As written in the paper, “BiCNet can handle different types of combats under diverse terrains with arbitrary numbers of AI agents for both sides. Our analysis demonstrates that without any supervisions such as human demonstrations or labeled data, BiCNet could learn various types of coordination strategies that are similar to these of experienced game players.”

This system is already capable of executing a number of Starcraft strategies which high-level players of the game use. To add to that, it doesn’t require supervision and is capable of taking action during combat without the need for specific instructions. As is the case with most deep learning systems, the AI learned through trial and error and is now able to adapt to changes during combat.

Starcraft

The authors also go on to mention how BiCNet approaches different in-game scenarios and handles multiple types of units, “BiCNet is easily adaptable to the tasks with heterogeneous agents. In our experiments, we evaluate our approach against multiple baselines under different scenarios; it shows state-of-the-art performance, and possesses potential values for large-scale real-world applications.”

That last bit is interesting. Despite being developed with a particular purpose in mind, developers always hope that their AI will have a broader range of uses in real-world applications. “Real-world artificial intelligence (AI) applications often require multiple agents to work in a collaborative effort. Efficient learning for intra-agent communication and coordination is an indispensable step towards general AI.”

Imagine an AI designed to control space marines going rogue, though. Nothing could go wrong there… right?

Manish Rajesh

Manish Rajesh