mike cook 🏝 Profile picture
Royal Academy of Engineering AI Research Fellow: AI Creativity, Generative Systems, Automated Game Design. Also @rogue_process @angelinasgames @procjam. He/They
30 Apr
Just want to observe a couple things about Salesforce's latest AI romp, this time into economics. Fortunately it's not been too widely reported on, so I was tempted to just leave it, but a couple of things need pointing out.
Before I start let me preface this by saying that I know this is "preliminary" work but it's also heavily backed by Salesforce's PR and has been widely reported in the press, and I think it should be discussed as seriously as it might be taken.
The first is that, like all models, this is based on assumptions about human behaviour, and simplifications to make them calculable. The addition of AI does not make it less of a model, even if it's very nicely presented. AI is not a crystal ball, it is a calculator.
Read 10 tweets
22 Apr
I've been working on a new kind of AI game designer, one that explores a game codebase directly and experiments on it. It's led to some surprising conclusions about how even the simplest coding decision can change how an AI works. A short blog + paper: possibilityspace.org/ai-game-develo…
The blog is a brief primer on this new research strand of mine, but the main output for now is a CoG paper I wrote about how software engineering, and specifically game development, affects how an AI understands a codebase. You can get it on arXiv here: arxiv.org/abs/2004.01770
This work is exciting and interesting to me, but more importantly - as I hint in the blog - it's also an effort to shepherd research from AI game design and make sure it benefits the people who make games, not the bottom line. I want to work with devs of all kinds to guide this.
Read 4 tweets
31 Oct 19
Huh, I missed this detail from the AlphaStar news yesterday. That's... extremely odd. This thread contains a bunch of interesting criticisms I'd not seen before.
Overall Florian's tweets (and excellent gifs) remind us that this is not about beating humans on even ground, because it can't ever be, because the idea does not make sense. We can twist the setup, get signed notes from whoever, but an even matchup does not exist. It's nonsense.
Games are a really fun domain to work in, and I'm glad DeepMind, OpenAI and so on are playing in the space with us. But I'm still yet to understand where any of these labs see their work actually going. "The real world" remains a vague and nebulous idea.
Read 8 tweets
24 Apr 19
This is incredible: Huawei's latest phone claims to have an AI-enhanced mode for photographing the moon that defies belief. Then someone conducted an incredible investigation and concluded that it's completely faking the entire thing. androidauthority.com/huawei-p30-pro…
Here's the original investigation report (in Chinese, but Google Translate does an okay job) with amazing examples and detailed explanations: zhihu.com/question/31998…
Huge thanks to @eegnsma for sending this my way, this is one of my favourite AI stories of the year so far. Why on earth would you do this. So bizarre.
Read 3 tweets
13 Apr 19
Three hours from this tweet, @OpenAI's DOTA 2 bots will be taking part in what OAI are calling their "final" live game. I'll be tweeting about the game in this thread, aimed at non-DOTA players.

πŸ“Ί Livestream: twitch.tv/openai
🀫 Hashtag to mute: #OAI
@OpenAI Last year the bots played live on stage at The International, and lost to two human teams. Here's a thread I wrote during their first game, which was played against the professional team paiN Gaming:
@OpenAI And here's a thread I wrote about their second game, which they played against a team of ex-professional "old boys". The overall skill/experience level here was arguably higher than paiN's, and the strategy employed by the humans is slightly different:
Read 163 tweets
27 Nov 18
Can't agree more with Folmer here. A similar article went up on Gama recently about Jon Blow and Jason Rohrer, I think, on selling without Steam or something, and it's just the most disingenuous kind of "business" writing you can do.
It's not just the advice given is inaccurate - it also covers up problems like compounding success and survivorship bias, which we could be trying to solve but instead are ignoring in favour of the mysterious allure of "if you make something special enough, you'll be famous".
imo the answer to "too many developers on Steam" or whatever is not "get rid of the chaff" or "be smarter than everyone else" but to re-examine what we want the indie community to be, what sustainability means, and how we can stop the problem of feast or famine.
Read 8 tweets
5 Sep 18
I know you already know this, but today's thread from Jack is the perfect crystallisation of the modern tech belief that neutrality is achievable, and specifically that machine learning can be made neutral. It's so obviously wrong, and will be looked back on with incredulity.
Obviously Jack himself is far right of centre, but this doesn't stop people (including, probably, Jack) from believing he's aiming for the perfect, balanced middle point of blind justice.
This isn't unique to the right-wing either, I meet a lot of people in CS/tech who would describe themselves as left politically, but believe that perfectly rational technology is the answer to the world's political ills, if only they could implement their version of it.
Read 3 tweets
23 Aug 18
I'll be livetweeting OpenAI's second bots-vs-humans DOTA 2 match at The International in this thread. Once again I'll be trying to keep it light on AI & DOTA jargon. The pre-game intro should be starting soon!

πŸ“Ί Livestream: twitch.tv/dota2ti
🀫 Hashtag to mute: #OAI.
First we'll be having winner/loser interviews, analysis of the game that just ended, so you probably have 10 minutes before we move into the OpenAI game. Go put the kettle on! #OAI
The bots lost yesterday, and under the rules of a best of three, this means a loss today will be the end of their International appearance - there won't be a game 3. That makes today's game an elimination match! Can the bots pull through and win? Get those predictions in now.
Read 74 tweets
22 Aug 18
I'll be livetweeting OpenAI's (first) bots-vs-humans DOTA 2 match at The International in this thread. I'll be keeping it light - will try and focus on big ideas and explain or avoid jargon.

πŸ“Ί Livestream [will start soon!]: twitch.tv/dota2ti
🀫 Hashtag to mute: #OAI.
The game won't be starting just yet, but they should start the intro soon. We still don't know a lot about the match today. Big questions are: who's on the human team? How many matches are they playing today? And most important of all: have any more restrictions been lifted? #OAI
It's extremely hard to know what we'll see today. OpenAI have been training more than ever before in recent weeks, according to their project lead. They were already using a vast amount of compute to train, so this is incredible to hear: #OAI
Read 60 tweets
24 Nov 17
Trying to draft a tweet about this week’s research event at Dagstuhl that is not a subtweet aimed at the entire history of academia. Good things undoubtedly came out though, and seeing the best people feels warming and empowering. And yet.
What I will say is this: the status quo is not a hill you want to die on.
And a lot people don’t actually want that, it just takes some time to recognise that that is what is going on? I think everyone understands that feeling. That’s how the status quo works. It sneaks up on you.
Read 7 tweets