Programmatic Conquest

May 21

SatStatSim: A SpaceGambit funded adventure.

As I alluded to earlier, we’ve got a new project for the next few months, some sexy space simulations we were funded for by SpaceGambit.

Lacking any creativity, our Satellite and Station Simulator is currently called SatStatSim, and it’ll split the difference between games and hard math simulators, so that students and New Space enthusiasts alike can enjoy it.

The initial version will support a range of orbital objects from cubesat to small manned stations. Put together your own pet designs using modular parts, hit the button, and see what kind of performance you can get. What sort of problems plague satellites and stations in orbit, what features can you get?

I’m really excited about this project, not only is it a great project I think lots of people will have fun with, but it’ll be useful math for building better stations for our sci-fi rpg, because we’re crazy completionists.

SatStatSim will be an open-source project, and we’ll be setting up a separate website for support and discussion soon, as well as hosting public discussion at CrashSpace. Excelsior!

May 9

Coming Soon to a Bookshelf Near You

I just got an alert reminding me that DSM-5 is going to be here in a few weeks, and barring any last minute retractions from the APA, looks like that last draft is the final word.

It’s the tried and true disease model of psychological disorders for at least one more edition, although more fundamental and scientific approaches like Research Domain Criteria made up a lot of ground in the nasty ten year fight over the 5th edition, there’s too much inertia and too many clinicians who need straightforward answers for patients presenting specific symptoms.

Here’s hoping the next edition can move beyond the dueling symptom lists and spectrums psychopathology is currently defined by.

But, while I’ve disagreed with every draft of DSM5, I’m still going to get a copy of the final and read it cover to cover, because while I’m a contrarian, I’m not THAT contrarian.

There’s a lot of good updates and stats and science to know! So here’s to Psychiatry, slowly moving forward.

Apr 30

Intro to Arduino: Reloaded

outlawpoet:

Intro to Arduino is a single two hour course to learn all the basics of how to set up, program, and wire up your arduino to anything. The May 30th and June 1st classes are identical, so feel free to choose the day that works for you!

A laptop and Arduino is all you need to interface with the real world.

Arduino comes with libraries to interface with hundreds of different electronic devices and gadgets. You can use it to connect physical things to everything from simple sensors to internet feeds, leave it running standalone or hook up to computers, other arduinos, custom radios, and much more.

We’re running some more Arduino classes for all the people who couldn’t get into the last one!

Crashspace prevails.

Seriously, there is massive pent-up demand for these classes, all I get these days are emails begging to get in.

Next one will be a more advanced Arduino class.

Apr 23

Sci-Fi to Real-Fi

Something I haven’t really talked about is the details of the sci fi rpg we’re making for the kickstarter. I have a lot of mechanics and a design document put together for it, skeletons and frameworks in Unity, but it’s felt too unfinished to talk about in detail.

Visuals will make it easier to discuss, and we plan to have a separate video focusing on the game. Science fiction games are in a weird place right now, Halo, Mass Effect, and Half-Life are giant science fiction franchises, but it still feels like the sci fi concepts are in the dog house. Everything reverts to dressing on manshooters.

We wanted some real hard sci fi concepts in our game, and we started looking around for some simulation code on space stations. How atmosphere, heat, power, waste, food handling all work in a closed environment. And we couldn’t really find anything. The math itself is understandable, and there are references from studies by NASA and whatnot, but no one had put it together in a usable way.

It seemed like we’d have to hack together something ourselves, to whatever level of detail we wanted, and leave it at that.

Enter a group called SpaceGambit, run by some friends, who are looking for projects to further their goal of community-led projects for the advancement of space development in a open-source, for all humanity, kind of way.

I realized that if I couldn’t find any build-a-station simulation, neither could anybody else. And what else could inspire more dreams of space expansion than making your own little home on the asteroid range?

So we wrote up a proposal for making a full fledged simulator of satellites and space stations with real math behind it, where you could make your own out of modular and parametric parts, and then see what happens, what capabilities and flaws it has, and what it would cost to get such a thing into orbit using various launch systems.

We’re going to try to straddle the line between real educational tool, and fun clickyness. As much as I love Kerbal Space Program, the total inaccessibility of numbers and systems keeps it from being everything my nerdbrain wants it to be.

So alongside the simulator will be an open library of all the space parts you can use, along with math and stats on how they work and what their characteristics are, so that people can add more and more detailed options to use.

Or that is the grand plan if we end up being selected by SpaceGambit, which will necessarily push the kickstarter back some for the four months of SpaceGambit, if not, our development here will be limited to what we can use in our kickstarter game, which will be a little sad but still useful. And perhaps we can carve it out an open source library others can use for their own scifi games or code projects.

Apr 10

Being Fun to Play

While most of my expertise, and a lot of the initial impulse for this kickstarter are the AI elements and underlying theory, I know for most backers, the critical part is the game. Not only is it the most tangible reward, but it’s the medium through which the AI toolkit can be judged. You can’t have intelligent behavior in a vacuum, and more than that, it’ll be the proof that improved AI can make better games.

I’ve probably played hundreds of videogames, on computers, consoles, handhelds, phones, and arcade machines. While some pundits may disagree, I don’t think there’s any one thing that can make a game fun to play, but it’s the most crucial characteristic of a successful video game. Things like graphics, storyline, and other individual features are good, and I obviously believe that AI can make a better one, but nothing will make a good game by itself.

The quality of being fun to play depends on understandable controls, a sense of progression and agency, and a reward system that pushes you through the game. This mix of things is what I’m most concerned about in the project, and part of why we’re launching a kickstarter in the first place. We need a community, for a deep pool of ideas in the development process, and wide ranging feedback as we change things. Also, if things go well, we’ll need to hire artists and programmers, and recruiting within the group is better than getting disinterested parties up to speed. I want a group of people who are invested in the game, who have been a part of the initial stages and helped make it better.

Every backer of any level will become part of the beta, and the higher levels will be on a mailing list for development and part of the alpha, which we hope to have open by January. Our plan is to get to a functional prototype as fast as possible, and then enter a long, open development process, where we can both add content and improve the experience.

I think there are a lot of people who are as frustrated as we are about the state of game AI, and I hope you will want to be a part of things.

Apr 6

ToyDecider: a kickstarter that’s hard to explain.

There are two things that trip people up when I try to explain this in person. The first is “ToyDecider”?

The story behind the name is long, but the short of it is that it’s sort of a joke, the way that mathy terms tend to be uber-macho, problems are “trivial” solutions are ‘obvious’, regardless of how difficult they actually are to implement.

I’ve been doing AI research for a long time, and while we were trying to come up with actual deep frameworks capable of supporting generic skills, (what’s sometimes called AGI(artificial general intelligence) to contrast it with narrow AI abilities, I would be thinking about the straightforward, pragmatic way to do something, if you didn’t care about the system being conceptually clean, or weren’t willing to waste cycles. How would I implement a system if I didn’t have to worry about it fitting in with other systems? It might still be a lot of work, but being conceptually straightforward, I started to call the hypothetical “known tricks” system ToyDecider, to contrast it with the “real” Decider which was too complicated to even start implementing.

Now that I have a chance to do it, I worry that it makes it seem like ToyDecider won’t be capable, so I think we’re going to spend some time contrasting it with existing AI in games more strongly, because while from an AI research perspective ToyDecider might not be exciting, it has some features gamers have never seen before in a production game.

The second thing that doesn’t seem to grab people is the open-source interface we’re designing for AI controllers in general.

To me, this is the most critical part of the kickstarter. The goal of all this is to kick over the anthill of assumptions and lowered expectations of AI in video games. More than personal success I want community success. And that only works if devs see a path forward. We’re never going to be the only provider in the universe for something, but we can be the first in this specific niche. There are AI toolkits, but none as wide-ranging as we want to provide.

The interface isn’t just about using ToyDecider in your game, it’s about establishing standards for level creation, what meta-data your game objects need to have for the AI to perceive them properly, what interfaces your game needs so the AI can communicate with other NPCs and your players.

There are a lot of advantages to building games with more organized meta data and standardized interfaces besides making it easier for AI to behave better, but people won’t do it without obvious advantages. We need to sell those advantages and provide for the possibility that other people will want to use the interface for their own AI toolkits and mods.

Apr 4

Coming soon: a Kickstarter about AI in video games

So I’ve spent a lot of money on kickstarters. It’s a place where I’ve gotten to have input on all kinds of crazy things you’d never be able to buy in a store(unless you got really lucky).

I’ve been planning a project of my own for months, and a friend of mine finally had to tell me to set a deadline or get a new job, because I’ve been moping endlessly, vacillating between going back to academia, getting a job in Big Data, or doing this fantasy project I know would be amazing, but I need help doing it.

So I’m launching a kickstarter on the 24th of this month.

I’ll blog a bit about the process of setting this up here, but mostly I’m going to be trying to put together as good a demo of the technologies I’m going to use as possible. My friend Joshua Truett is helping me, he’s focusing on the features and presentation, so people can actually understand what we’re talking about.

It’s a complicated project, but I hope we can outline our goals clearly.

AI in video games is terrible. So proverbially terrible that everybody has gotten so used to it we don’t talk about it anymore. Actually, the fact that some of it is even called AI is actually kind of embarrassing for an AI researcher like me.

It doesn’t have to be like this. Actual AI research, in Academia and Enterprise has leapt ahead in the last ten years. Most of it can’t be used directly in video games because it’s too resource-intensive. Grad students think nothing of throwing gigs of RAM and minutes of analysis at problems games would need solved every frame. But sparse implementations could be made.

We’re going to make an AI toolkit. A 3rd party tool you can hook up to your engine and make your NPCs behave in complex, trainable ways. Giving them a wide array of basic skills that you won’t need to endlessly script.

And to prove it works, we’ll have a game alongside it. An RPG you want walk around in and see thousands of characters interacting, seamlessly between players and NPCs.

When you aren’t playing, your character will remain in the game, acting as you’ve taught them to act. When you log back in, it’ll tell you the story of what it’s been up to while you were away, so you can fine tune their tendencies, characteristics, and intelligence.

To make all this work we need a community. A pool of interested people to play the game, and tell us in detail what’s smart and what’s dumb. This is a task I think gamers are pretty good at, maybe even better than most AI researchers, judging by forums. Gamers are ruthless. And they all want different things, for some, smart will be game-theoretic perfect gold grinding, for some it’ll be never breaking character and roleplaying every interaction, for some it’ll be killing every misbegotten soul unfortunate enough to show their back to them.

If we can get enough people into our alpha and beta, collect the data and tune the algorithms, I think we can really shock people with the intelligence and lifelike behavior we can show. We can be in alpha probably in 8 months, a usable beta in a year.

I want to start an arms race in the games industry. Whether they end up using our toolkit, or they build their own and outcompete us, I don’t really care. I just want games that have characters that live and act with some sensibility.

And of course we really need to have rock solid solutions to the niggling idiocies, like pathfinding and threat detection, where the solutions are known, but everybody implementing their own halfass version means a different bug in every game.

We can, to some extent, modularize behavior, the same way that physics, graphics, and networking have been modularized, so that we don’t have to throw out all the progress we’ve made every time we upgrade game engines.

More to come on the project.

For those unfamiliar, kickstarter is a crowdfunding site, where you can contribute to individual projects. The hook and best feature is that if the project doesn’t reach the funding level they have set as necessary, individual contributors are not charged. So you can feel safe pledging some money that the project won’t get some money but not enough to succeed, and you’ll have wasted your twenty-five dollars. Some amazing projects have happened only because of kickstarter, and I have put a lot of money into it.

Apr 2

Crash Space » More Intro to Arduino Classes on May 2nd and 4th

introtoarduino

I finally scheduled some more classes so you can learn all about the easiest microcontroller platform there is. Come and take home a new understanding of Arduino, a few functional projects we’ll make in the class, and a kit of parts to do more.

Mar 11

“interesting in places, fairly readable, moderately informative, but wildly overstated.”

Homunculism by Colin McGinn | The New York Review of Books

Some much needed sobriety on Kurzweil’s latest book, without descending into any dismissive or broadly denialist language.

Ray Kurzweil is exactly the best and worst friend the AI field has. Best because an unknown but large number of people have been inspired by him over the last fifteen years to go into AI and associated fields, the worst because his is the worst kind of unfounded dilettantism, overconfident nontechnical speculation outside his specialty by an engineer. His language is precise but expressing imprecise and ungrounded ideas.

I always imperceptibly wince when he comes up in conversation, but I’m glad he exists.

Mar 10

(via http://twitter.yfrog.com/z/mm9asxuj)

Another great picture by matt, a panorama of shop one at CrashSpace. This is where we do light work, electronics or hand tools, and the 3d printers and epilog laser cutter you can see at the end of the room there.