Ask HN: Is OpenAI research the latest wargames? Why not vs. climate change?

2 points by DrNuke 7y ago ↗ HN

OpenAI released their latest algo baselines to the general public yesterday. Their AI is also playing top Dota 2 human teams at The International from later this day. My question as a member of the general public is pretty simple: are these the latest wargames instalments (you know, the Matthew Broderick kool-aid from his 1983 film and so on)? Is it possible to make these AIs play army vs army other than synthetic army vs synthetic army? AIs playing Civilization by Sid Meier is fascinating enough and Youtube is full of demos, but real-life scenarios? At the very least, the weakest links in real plans would be discovered. If so, why not let these latest AIs play vs climate change, for example? Just interested in HNers tech wisdom. Thanks.

3 comments

[ 3.0 ms ] story [ 16.4 ms ] thread

stealthcat 7y ago ↗

Their github repo codebase is too unreadable to be useful. Some of worst software engineering out there. The irony of some org trying to push openness but giving near unusable code.

mikert5671 7y ago ↗

is this by design or it just doesnt matter having bad code?

DrNuke 7y ago ↗

Adding a bit of internet research from this morning, hopefully it helps. Of course, AIs are not a magic wand and we should not expect this. Deep reinforcement learning for stochastic dynamic systems is still extremely difficult, most of all it is the Markov chains modeling / unmodeling to make scenarios ultimately unreliable. Running a number of these scenarios to achieve the statistical grade maybe? Weather forecasting seems a viable analogy and a first order approximation, at this time, even if time-series predictions are passive while rl agents interact and optimize (that's why climate change with human's effect, agents playing humans or any other stakeholder). We will see.