Introducing 'Goal State'
My new series of online courses for game AI coming to Kickstarter in Q3 2024
Today I’m proud to announce my latest project, Goal State, is coming to Kickstarter in Q3 of 2024. Goal State will be a series of online courses and educational content on artificial intelligence for video games, designed to provide an accessible and informative set of materials that consolidates the many disparate threads of AI applications in games.
Having amassed an audience who enjoy my musings on AI for games, I wanted to find a way to build atop my years of experience as an award-winning university lecturer to provide lectures, tutorials, homework exercises, and more that helps teach people interested in the field to learn to do what I do.
My goal - pun intended - is for multiple courses to emerge under the Goal State brand. The Kickstarter will help finance the first course, Game AI 101, which I will detail in a moment.
Today, I wanted to highlight my plans for Goal State, what I hope to achieve in the long-term, and critically why I’m running a Kickstarter for support.
Every (Good) Idea Has a Story
My work building AI and Games, be it the YouTube channel, this Substack, the Branching Factor podcast, or my broader industry activities, has been a real labour of love. I have spent over a decade sharing what I can about the world of artificial intelligence for video games, to a point that it has fundamentally changed the course of my career. Transitioning out of my work in academia, to engaging more with the broader games industry, and publishing a variety of materials through my media platforms.
With over 100 videos launched on the main AI and Games YouTube channel, dozens of articles online, and tens of millions of views to boot, it's been a pleasure to see how people engage with each new episode, and critically learn something new about how AI intersects with games. Either as a student, a researcher, or as someone who works in the industry, who will take these ideas and apply them to their own projects. And of course, for all the good folk out there who tune in simply to learn more about the games they love and how they actually work.
Goal State is planned as a series of courses: teaching people how to build AI for games, without any background or knowledge of the subject.
But to my mind, it's still not enough. AI and Games was designed to be a high-level overview of these topics - after all, it started out as a supplementary material for my students during my lecturing career. As a result, we don't dig into the technology as far as we could. Often in every YouTube episode, there's plenty more details left on the cutting room floor: the underlying theory, the written code, the challenges of making this come together.
Now as far as I am concerned, AI and Games satisfies its remit. It’s ‘edutainment’; it’s aiming to provide an informative yet concise overview of complex subject matter. But what about those who need more help to get started?
It has been a great pleasure of mine to receive messages from viewers around the world whose interests in AI and/or game development emerged from a chance encounter with one of my YouTube episodes. Regardless of the many other metrics of ‘success’ that may be prescribed to my work, learning that people the world over have began their own journeys and emerged as university graduates and professional game developers as a result of AI and Games is as humbling as it is edifying.
But while it is wonderful, this correspondence also reminds me of the ‘gap’ in my work: that more could be done to provide richer, deeper, material that helps onboard new learners. I want to be able to help others who want to start from scratch, knowing that what they’re learning is accurate, correct, and will help support them in their journey.
Plus it’s clearly something that a subset of my audience have been asking for. For years I've had my community ask me: “Why don't you run online courses? Why don't you teach game development to your audience?”
Goal State is my plan to make this a reality.
Goal State: The Series
Goal State is planned as a series of courses, built around teaching people how to build AI for game development, without any background or knowledge of the subject. My goal is for anyone who watches an episode of AI and Games to be able to start a course on Goal State and over time learn how to reproduce that which they've seen in their favourite episodes.
But of course, if I’m going to build a series of online courses, I want to do it right. Those of you who have followed AI and Games for a while, will know that I spent a decade as a university lecturer in the UK. In fact, spread over my career I have over 15 years experience in teaching AI, computer science, and game development. I've won numerous awards for my teaching over the years across multiple institutions, and the university courses I've led have been nominated for national awards.
And so I put all this experience to practice: I've spent the past year researching every facet of AI for games that would appear somewhere in the Goal State syllabus: from the foundations of AI, to innovations in deep learning, modern generative architectures, classic game AI techniques, and all the other weird and experimental stuff that deserves it's time in the light.
I want Goal State to be the source of learning about AI for games:
A resource for people with zero background and experience in AI or game development to learn to do what I do.
I want it to become a training resource that studios can use to help onboard new team members.
I want it to be a tool that academics working in community colleges and universities around the world can utilise to support their teaching.
I want Goal State to support the next generation of developers who want to get into AI for games. And to do that, I need your help.
I’ll detail the specifics of the first course momentarily, but first I wanted to dig a little more into why I feel this initiative is sorely needed.
Building a One-Stop Resource
AI for game development is something of a niche. Video games have a variety of unique challenges that are bespoke to them. When AI is taught in academia, it seldom arms students and graduates with the right information on how to build AI for games. Game AI - the practice of using AI to build parts of a gameplay experience, such as opponents, adversaries and companions - is very much its own beast. This is a sentiment that has simmered below the surface for a very long time, and there are things I wanted to try and address with Goal State’s creation that would satisfy my frustrations.
Theory Meets Context
Having spent close to 20 years working in and around AI and game development in higher education, I’ve been frustrated by how most educational resources lack the necessary features needed for people to truly be immersed in the field. Academic texts and classes are often too theoretical, or don't focus on the needs of game development. Meanwhile many programming textbooks and similar sources built for game developers lack the theoretical backing. Yes, I’m complaining about too much theory, and not enough theory. But that’s largely because you need to balance it out: theory needs context; practical applications need to be reinforced by theory such that you can see how it ‘fits’ in the grand scheme of things.
If you’re a long-time fan of AI and Games, you’ll have noticed that I have the AI 101 series of videos/blogs. These emerged as a result of my need to build a stronger theoretical underpinning to the rest of the ‘main’ series, in which I typically focus on one games’ technical merits. The massively self-referential way in which my YouTube channel runs is both a desire to ensure viewers understand that not only is everything fundamentally connected - that no great ideas exist in isolation - but is also a means to prevent repeating myself. Given I suspect if I had to establish what a behaviour tree is for 57th time, it’s going to annoy repeat viewers.
Now AI and Games’ structure is a retcon: an attempt on my part to better organise my output. Largely addressing two issues that arose from its success:
I didn’t expect AI and Games to survive for 10 years
AI and Games wasn’t built with the intention that it become the educational resource it has become, and at the scale it has achieved.
With Goal State, I want to address this from the very beginning: every theoretical concept we introduce is tied into existing industry practice, or it is made clear why that is not the case. While also highlighting the potential of established concepts in academic literature that could, with time, become industry practice.
We will start from the theory and build forward to the context, rather than AI and Games, which talks about the application in context and has to squeeze some theory in there where it can. Goal State will refer to AI and Games, because why wouldn’t it? We have over 100 videos of useful information accessible to you. But now we will build the theoretical underpinnings, and then highlight the context both by looking at industry examples, but also in practical assignments as we get you to build these systems yourself (if you so wish).
Consolidation of Knowledge
Returning to an earlier point - that academic work is often lacking in context - I want to take a moment to defend my peers in higher education. This isn’t a slight on them and their hard work. Rather, the lack of context has historically been due to the often limited and sporadic exchange of information within the industry of how AI ‘fits’ in games, or how AI is used in the industry.
Now of course, sometimes that’s because the academic in question isn’t interested or familiar with game development. And until perhaps 10 years ago, games were not considered a serious application of AI in academic circles. The research field I was part of in the late 2000s was very small - a point I discussed at length in my recent 10 year retrospective video. And critically the application of AI in games as opponents and NPCs was seldom considered ‘real’ AI by scholars - hence the creation of ‘Game AI’ as a sub-discipline of the field, given game developers needed a way to share ideas and methodologies with one another.
And this brings us to the other big problem: given the games industry was talking amongst themselves, there was little emphasis on broader communication that could later be utilised by academics and those keen to join the industry. It was and still is hard to find the information and resources that provide context to students. It’s rather telling that when you ask people who work on AI in the games industry what books they should read, they’ll largely mention the same ones: AI Game Programming Wisdom and then the Game AI Pro series, both of which edited by Steve Rabin. These books are fantastic resources, but most of those volumes are long out of print and won’t be re-released. It’s why everyone was so excited by the announcement of the Game AI Uncovered series edited by Paul Roberts, given we finally (FINALLY) have another series of books to share our ideas with one another.
But even then these books don’t consolidate everything. They expect you’ve got some theoretical underpinning. That you’ve read the likes of ‘Artificial Intelligence: A Modern Approach’ to learn the basics. The same can be said of the ‘Artificial Intelligence for Games’ book written by professors Togelius and Yannakakis. Again, let me stress that these resources are all fantastic - I wouldn’t pay them lip service lest I felt it was warranted. But there’s so much content spread out by different people who work in AI, and in different corners of the sector, that it can be difficult to find all of the information you might need.
And of course, there’s the events! Conferences are a great resource for the exchanging of ideas. But how many of these do we really have for AI in games? The Machine Learning and AI summits at the Game Developers Conference (GDC) are fantastic*, but also prohibitively expensive (in both time and cost) for developers to engage with. Either because of the cost to reach San Francisco, the time and effort to gain entry to the United States, the cost of the ticket, or even if you don’t attend, the cost of the ‘Vault’ pass to access recordings of the talks.
*Full Disclosure: I am on the advisory board to the AI Summit, meaning I work in a volunteer capacity to help curate speakers and run the event.
Meanwhile in Europe, there’s been little to show for the community since the fantastic Game AI Conference (later renamed nucl.ai) that ended in 2016, and Game AI North that ran for one brief (but brilliant) iteration in 2017. Many of these talks are often lost to time. Sure, the information might be out of date, but it’s still valuable.
Why do I bring all of this up? Well, a big part of what I’ve done with AI and Games - first accidentally, but now by design - is tried to consolidate this knowledge as much as possible: providing links to resources, finding articles and blogs that were thankfully saved either in Archive.org, or can be accessed user their Internet Wayback Machine. Episodes of AI and Games become starting points for viewers to engage in deeper research on a given topic. Plus I try to maintain up to date links to resources, and where I can, update them in the event they go dead.
As much as I dislike blowing my own trumpet, AI and Games is - to my mind - the second-longest running entity dedicated solely to the communication and dissemination of AI practices in video games, with GDC and the AI Summit (and it’s previous incarnations) comfortably in the top spot. A big reason for that is GDC is the only one not being driven by a sole individual.
If you look at all of the book series, events, and programming websites out there that support the games/AI community, their existence was largely predicated by a single individual. Without Steve Rabin’s hard work, almost a dozen books on AI for games wouldn’t have been published. Alex Champandard’s work in fostering AIGameDev.com and the Game/AI Conference built a community for game/AI professionals in Europe. Plus really good websites such as Alan Zucconi’s tutorial series and Amit Patel’s Red Blob Games - both of which cover more than just AI for games - have been delivering great content for 10 and 20 years respectively. Sure, AI and Games is courtesy of one foolish albeit handsome Scotsman, but that’s also why I feel it is, perhaps arrogantly, my responsibility to build something like Goal State to ensure this information is shared such that it can exist without me in the future. I know a lot about this field, but it seldom gets shared as thoroughly as it could.
One of the big reasons I work with a lot of games companies is because of the breadth of knowledge I have on AI in video games. I won’t sit here and tell you I am a hardcore expert on any one subject - given I could quickly list people who I would consider the definitive experts of any particular subset of the games/AI taxonomy - but you won’t meet many people who know the space as broadly as I do. Again, this was more accident than design, a by-product of building up AI and Games for a decade.
As such, I feel it’s important that I try and make an effort to deliver on this in a way that will have lasting impact for my audience, and to do it in a way that nobody else does. I could go and write the Goal State book, but all it would do is point you to the books I’ve already listed. Instead I should focus on what I do best: videos, and university style teaching.
Oh and speaking of universities…
The University Paywalls
If you're lucky enough to study a university course that has a Game AI class, it is often treated as a one-off, and doesn't dig into the myriad of applications and sub-disciplines sufficiently. This doesn't bode well for people interested in building AI for games. It feels like we need to build a more comprehensive overview of AI for games, and allow for a broader understanding of theoretical underpinnings, applications in context, and discussing the merits of different approaches for different problems. Of course it’s not a requirement that a university education provide all of this at the level of detail I think is relevant - I won’t get bogged down in the discourse of what is the ‘right’ things to teach in higher education - but I do think Goal State can deliver university-quality education outside of that environment.
But I also want to make it accessible without it being tied to university regulation. Higher education institutions are increasingly protective of their educational material, often with lectures and other materials being held behind in virtual learning environments (VLEs) that can only be accessed by registered students. Naturally active students are the priority for a given institution, but current practices prevent any information from being more widely shared. This makes it increasingly difficult for anyone interested outside of higher education to find out more at this level of expertise. In fact, it also makes it difficult for university staff to share their own curriculum, given so much of this information is not made publicly available.
In fact, I can speak to this, given I have written hours upon hours of content all about AI and game development over the years for several universities. That material is owned by those institutions. This both highlights my capability to craft Goal State, but also why it is needed. I can’t simply share that which I’ve done before with you all, but also I know what I want to write about, given I have done much of it already.
With Goal State I want to provide more readily accessible materials, at a quality befitting a university-level education, but with production values that fit with my existing work. This will enable not just for aspiring AI developers to learn at their own pace, but for university staff to utilise these materials in ways that best suit them.
Coming to Kickstarter: Game AI 101
I'll be launching a Kickstarter in Q3 of 2024 for the first course Goal State has to offer: Game AI 101. This course will provide a full length university-style curriculum with three classes:
Foundations of AI [~30 hours]: The theoretical fundamentals of classic AI, and the variety of subdisciplines within it that can be used in games.
AI for Games [~30 hours]: A deep-dive into specific AI applications historically used in games, and how it can be applies to solve game design problems.
Designing AI for Games [~10 hours]: Rounding out this introductory course, is a design-focussed segment dedicated to challenges that emerge when using AI in practice in your own games.
Right now, my plan for this course is to offer over 50 hours of content, with video lectures and tutorials, plus quizzes, homework assignments and more. The equivalent of a university semester's worth of material that you can pursue in your own time, and at your own pace. But to build all of this context to the same standard and quality you’ve come to expect of AI and Games, this will take time. And time - as the saying goes - is money.
Producing the full series of lectures, tutorials and homework assignments for the course will be more or less a full time job for about a year - bearing in mind this will run alongside my existing content production pipeline and other commitments. While I know the subject matter well, it will be a significant time investment to write, record, edit, and prepare each individual segment of Game AI 101. As such, I need your financial support to get this project off the ground.
Follow Goal State on Kickstarter
Goal State will go live on Kickstarter in Q3 of 2024, but in the meantime you can register your interest and be the first to know when it goes live via this link.
In the meantime, I will be sharing more updates in the weeks and months to come as we build up to the launch. I'd really appreciate if you could help spread the word and share it among all your networks to help drive awareness.
I spent 10 years building up AI and Games into a trusted resource for people the world over on how artificial intelligence works in video games. It is my dream, to not only continue to build on this foundation and deliver new episodes of AI and Games to you every month, but make Goal State the definitive resource for everyone who wants to learn how to do it themselves.
Game AI 101 is only the first course planned for Goal State, in fact I've written the syllabus for three courses thus far, and my hope is that by making this Kickstarter a success, we can build an ever expanding series of content for people to access over time.
I won't lie, I'm excited, but also terrified. This would be a dream to pull off and make it happen. Frankly the only way I can do that, is with the help of my wonderful community who have supported me thus far.
Thank you all for you continued support, and I look forward to sharing my on Goal State as we approach the Kickstarter launch.