Contemplating a total career change.

[semaphore]

The info on the website isn't up to date. I asked them for full details about the course, and calculus as well as applied linear algebra is included. If I do take that programme I'm planning on taking maths as a minor during the 3rd year.

I'm still torn between that and compsci. If I do go the compsci route, it's only 3 years til bachelors and further 2 years til masters. A lot of thought will still have to be put into this.

1 year till masters from bsc, honors is 1 year.

And listen if you want to do AI, you need to be very good at math and stats like cguy has stated, taking it as a minor really is not going to bode well for you.

 

[cguy]

The info on the website isn't up to date. I asked them for full details about the course, and calculus as well as applied linear algebra is included. If I do take that programme I'm planning on taking maths as a minor during the 3rd year.

I'm still torn between that and compsci. If I do go the compsci route, it's only 3 years til bachelors and further 2 years til masters. A lot of thought will still have to be put into this.

You may want to check to see if an hons degree is require before an MS, or alternatively if they may expect something higher than a 12 year school career before starting (E.g., 3 year UK BSc's often require the equivalent of a "13th year").

 

[semaphore]

You may want to check to see if an hons degree is require before an MS, or alternatively if they may expect something higher than a 12 year school career before starting (E.g., 3 year UK BSc's often require the equivalent of a "13th year").

Generally speaking an honors is usually always required, maybe if you do exceptionally well in undergrad they may let you skip. But doubtful. Universities love to rip people off.

 

[cguy]

Generally speaking an honors is usually always required, maybe if you do exceptionally well in undergrad they may let you skip. But doubtful. Universities love to rip people off.

In the UK a 3 year BSc is often (regional, not in Scotland for instance) considered equivalent to a US 4 year BSc or SA BSc(hons). The regions where this apply usually have a 13-year school curriculum. I.e. There is no 4-th year BSc here - the hons designation means that you did better in your degree, which can certainly be s factor, but generally isn't a requirement for starting an MSc.

Some quick googling seems to indicate that holland doesn't have a notion of an honours degree, so as long as a wo (vs hbo) qualification is taken, Masters entry should be possible.

 

[semaphore]

In the UK a 3 year BSc is often (regional, not in Scotland for instance) considered equivalent to a US 4 year BSc or SA BSc(hons). The regions where this apply usually have a 13-year school curriculum. I.e. There is no 4-th year BSc here - the hons designation means that you did better in your degree, which can certainly be s factor, but generally isn't a requirement for starting an MSc.

Some quick googling seems to indicate that holland doesn't have a notion of an honours degree, so as long as a wo (vs hbo) qualification is taken, Masters entry should be possible.

Ye i was speaking locally, not abroad.

 

[HavocXphere]

game programming (AI programming in particular

I'd really suggest dabbling in it a bit first on a hobby level because you're proposing a sizable gamble here. e.g. You'd have to move overseas...game dev in SA is +- dead aside from an occasional indie project.

Here's a suggestion: Try writing a snake game - from start to finish, incl a bit of polishing. Like this. Seems easy enough, right? I promise you you'll have quite a few epiphanies along the way (I sure did).

Very personal opinion: Super interesting field & great fun...wouldn't touch it with a 10 foot pole as a career though.

And listen if you want to do AI, you need to be very good at math and stats

He is asking about game AI in particular - from what little I did see it didn't seem to involve any heavy maths or stats. Quite the opposite in fact...because its such a ridiculously complex area and resources are limited <10% of available horsepower for AI usually...so the end result is a lot of scripting to make it appear "real" and not a whole lot of actual AI in the true sense.

e.g. In the original Unreal Tournament (or maybe 2004?) the bots pulled off crazy sideway dodge/roll and just cleared your shot. Looked impressive...except it was +- hardcoded to you mouseclick. Impressive "real" intelligent behaviour...zero actual AI.

 

[semaphore]

He is asking about game AI in particular - from what little I did see it didn't seem to involve any heavy maths or stats. Quite the opposite in fact...because its such a ridiculously complex area and resources are limited <10% of available horsepower for AI usually...so the end result is a lot of scripting to make it appear "real" and not a whole lot of actual AI in the true sense.

e.g. In the original Unreal Tournament (or maybe 2004?) the bots pulled off crazy sideway dodge/roll and just cleared your shot. Looked impressive...except it was +- hardcoded to you mouseclick. Impressive "real" intelligent behaviour...zero actual AI.

I suggest you go do more research then, AI is very math heavy matrix algebra, bayesian statistics, calc etc. Look at the AI in crysis it has machine environmental learning in it.

HDS:
http://web.media.mit.edu/~jorkin/aibooks.html

 

[stoymigo]

Hey guys, so as it stands I am strongly thinking of changing my career. My current job pays good but it sucks - boring as hell. Working with so much incompetence in this country is also not helping.

I have my eye set on software development, as well as game programming (AI programming in particular, but that will require additional courses after the initial study). My parents are moving abroad this September to EU,and seeing as they are getting old I'd like to be near them. Netherlands would be my preferred choice of country to study in, they have high standards of living, hard working people that get things done, and very good standard of higher education.

This is what I want to study (programming is done in C++ during the core phase, not C). That programme is 4 years, and leads to a bachelor degree. After I am done with it, I plan on doing Masters part-time (+2 years).

Now, could the IT guys here comment on that programme I linked above? Any feedback will be greatly appreciated and would help me make my choice.

I am absolutely dead-set and 100% sure that I want this. I do study C++ in my spare time(got 2 books that I'm currently busy with), as well as some UML.

The university programme does cover various aspects about Software Development, but I am not in the game programming business.
The programm will help you towards software development, as for game development you should start developing little apps for starters and get a track record online while you trying 2 get a job at a studio, as a beginner you'll learn much more when you get a job and how the industry works , but I am not in the game programming business.
And try to get freelance projects online, even if it's for very little money and build a reputation.

If you prepared to be disciplined and learn constantly so you can become competent then go for it.

 

[HavocXphere]

I suggest you go do more research then, AI is very math heavy matrix algebra, bayesian statistics, calc etc. Look at the AI in crysis it has machine environmental learning in it.

What I was getting at is this:

Whereas "real" AI addresses fields of machine learning, decision making based on arbitrary data input, and even the ultimate goal of strong AI that can reason, "game AI" often consists of a half-dozen rules of thumb, or heuristics, that are just enough to give a good gameplay experience.

Look at the AI in crysis it has machine environmental learning in it

Never heard of the phrase. Nor has Google apparently. Taking a wild guess I'd say you mean:
http://en.wikipedia.org/wiki/Reinforcement_learning

Look obvious maths & stats will help but we're talking games here meaning that the aim is to create a convincing illusion, not to write a thesis on strong AI. In practice convincing illusion seems to involve scripts 99% of the time with a sprinkling of fancy AI terms on top of it for marketing purposes.

 

[HDS]

I'd really suggest dabbling in it a bit first on a hobby level because you're proposing a sizable gamble here. e.g. You'd have to move overseas...game dev in SA is +- dead aside from an occasional indie project.

I am planning on studying and working abroad. The moment I'm out of here, this country is dead to me.

Here's a suggestion: Try writing a snake game - from start to finish, incl a bit of polishing. Like this. Seems easy enough, right? I promise you you'll have quite a few epiphanies along the way (I sure did).

Thanks for the suggestion, will try it in my free time. I'm already busy on a small project of my own.

He is asking about game AI in particular - from what little I did see it didn't seem to involve any heavy maths or stats. Quite the opposite in fact...because its such a ridiculously complex area and resources are limited <10% of available horsepower for AI usually...so the end result is a lot of scripting to make it appear "real" and not a whole lot of actual AI in the true sense.

e.g. In the original Unreal Tournament (or maybe 2004?) the bots pulled off crazy sideway dodge/roll and just cleared your shot. Looked impressive...except it was +- hardcoded to you mouseclick. Impressive "real" intelligent behaviour...zero actual AI.

Sorry maybe my OP wasn't too clear, but I'm looking into studying CS/Software eng, and only afterwards attending additional courses for specialization. After obtaining bachelors I would like to keep my options open.

The above example of AI is just some tricks devs have up their sleeves. Take a game like Fear for instance, the AI were insane. But you are right, these days it doesn't look like there is much resources going the AI way...fancy graphics and effects seem to take priority.

http://web.media.mit.edu/~jorkin/aibooks.html

Thanks bud, bookmarked.

 

[semaphore]

What I was getting at is this:




Never heard of the phrase. Nor has Google apparently. Taking a wild guess I'd say you mean:
http://en.wikipedia.org/wiki/Reinforcement_learning

Look obvious maths & stats will help but we're talking games here meaning that the aim is to create a convincing illusion, not to write a thesis on strong AI. In practice convincing illusion seems to involve scripts 99% of the time with a sprinkling of fancy AI terms on top of it for marketing purposes.

Sorry i did not consult google.

Sorry but if you're going to do proper AI or advance the field of AI, maths is needed. If you're just going to regurgitate the same bull**** AI in existence then rather just stay out of the field. The purpose of learning is to advance not to stand still.

So OP if you gonna do what havoc is suggesting, please dont go into the field

 

[HavocXphere]

Sorry maybe my OP wasn't too clear, but I'm looking into studying CS/Software eng, and only afterwards attending additional courses for specialization. After obtaining bachelors I would like to keep my options open.

Ah OK. Thats a much more sensible course of action - by the time you're done with CS you'll also be in a much better position to evaluate thing. Its things like Fullsail that make me worry about people that develop a sudden enthusiasm for game dev.

 

[semaphore]

Ah OK. Thats a much more sensible course of action - by the time you're done with CS you'll also be in a much better position to evaluate thing. Its things like Fullsail that make me worry about people that develop a sudden enthusiasm for game dev.

My friend had a sudden passion for game dev, hes now a senior physics engine developer. There is nothing wrong with enthusiasm.

 

[HDS]

Ah OK. Thats a much more sensible course of action - by the time you're done with CS you'll also be in a much better position to evaluate thing. Its things like Fullsail that make me worry about people that develop a sudden enthusiasm for game dev.

Exactly.

Btw this wasn't an overnight thing. I was interested at the development process of games since I can remember. I'm a lot more interested in a single area of game development like simulated physics or graphics programming, where you can put all your effort, in contrast with developing your own game. Guess what I'm trying to say is the deep technical aspects interest me, not game development as a whole.

 

[HavocXphere]

proper AI or advance the field of AI, maths is needed.

I don't recall OP being on a mission to "advance the field of AI".

software development, as well as game programming (AI programming in particular

I agree you need maths etc for AI in the academic sense. In the game world its all script and state driven though. Why do you think entire games get written in C/C++ and then they patch on a scripting language? Because scripting languages are ideal for scripting NPC responses, trigger events etc. Unreal, Source, even cryengine has a built in script lang.

I assure you they didn't add the scripting langs for the "math heavy matrix algebra".

My friend had a sudden passion for game dev, hes now a senior physics engine developer. There is nothing wrong with enthusiasm.

Jeesus dude. I was referring to Fullsail in particular. Thats why I put the word "Fullsail" in the sentence. If you read that and somehow conclude that I think there is something "wrong with enthusiasm" in general then I really don't see us finding common ground here.

Its things like Fullsail that make me worry about people that develop a sudden enthusiasm for game dev.

 

[semaphore]

Ok Havoc you're right im wrong AI is just scripts, BODMAS is okay OP, You got your answer now

 

[cguy]

I agree you need maths etc for AI in the academic sense. In the game world its all script and state driven though. Why do you think entire games get written in C/C++ and then they patch on a scripting language? Because scripting languages are ideal for scripting NPC responses, trigger events etc. Unreal, Source, even cryengine has a built in script lang.

A lot of game AI has just been set of arbitrary rules, but even in those cases, there are primitives that involve everything from good path finding algorithms, that account for flow and collision detection (think bridges, danger avoidance, primary vs secondary goal prioritization, etc.), to visibility and hearing driven reaction (very geometrically intensive) and player prediction. These are often very math and algorithm heavy, and thus far, far from solved.

I would argue that there is a lot of math involved in game AI today, much more so than there used to be, and that the simple dodge and chase mechanics that rely on scripted events are being phased out, or at least being used to incrementally augment more and more generic parameter based behavioral engines. Engines always want to generalize, and allow scripting to add the occasional trick when necessary, rather than leave the developers to navigate a maze of logic. The same trend has occurred in game graphics, physics, collision detection, art pipelines, etc.

Also from a career perspective, I really wouldn't want to be the guy that just writes up a ton of scripts in order to get something that fakes a bunch of behaviors - I would much rather focus on the unification of AI technology for games, and develop the skills to push the boundaries instead.

I assure you they didn't add the scripting langs for the "math heavy matrix algebra".

No, that's why they add native bindings to the scripting language. There will always be a need for scripting languages as long as games are unique, but their existence doesn't mean that a lot of the AI can't be generalized into a frame work.

What I was getting at is this:

Quote Originally Posted by wiki

Whereas "real" AI addresses fields of machine learning, decision making based on arbitrary data input, and even the ultimate goal of strong AI that can reason, "game AI" often consists of a half-dozen rules of thumb, or heuristics, that are just enough to give a good gameplay experience.

You know, the very next line in the wiki is:

Game developers' increasing awareness of academic AI and a growing interest in computer games by the academic community is causing the definition of what counts as AI in a game to become less idiosyncratic.

It's still very different, as they subsequently say in the wiki, but the trend has always been away from specialized hacks in every aspect of game development, and I expect this to continue.

 

[HDS]

Ok well just an update :

I have concluded that a degree in compsci is the most flexible in the IT world. With the core concepts in place, you can literally do anything. I'm sure the compsci guys in here will agree.

I will be leaving to Netherlands end of August, that's when orientation starts for most universities. Need to settle in, visit the place, get to know my surrounding etc.

First I will be doing the 3 year compsci course at University of Groningen. At the end you get a bachelors + honours degree. In EU those 2 are combined, so you don't have to do an extra year, but more material is squeezed into those 3 years. Bsc = accomplished.

After the Bsc it's straight to masters. I am planning on doing the master's at Tu/e, 2 years.

If I'm still alive and kicking after the masters, I really, really want to do a Phd. Another 4 to 6 years, but part-time. Big ambitions, I know.

Anyway, at the moment I'm doing some coding with C++ ( got some books from deitel ) and loving it. Don't know why I wasted my early days in the army while there is a whole world of computer science out there waiting to be explored.

 

[gkm]

Good luck. I hope you have a great time.

 

[MartinMorrison]

A lot of game AI has just been set of arbitrary rules, but even in those cases, there are primitives that involve everything from good path finding algorithms, that account for flow and collision detection (think bridges, danger avoidance, primary vs secondary goal prioritization, etc.), to visibility and hearing driven reaction (very geometrically intensive) and player prediction. These are often very math and algorithm heavy, and thus far, far from solved.

I would argue that there is a lot of math involved in game AI today, much more so than there used to be, and that the simple dodge and chase mechanics that rely on scripted events are being phased out, or at least being used to incrementally augment more and more generic parameter based behavioral engines.

On what observations do you base this argument on? I'd be interested if you have links to specific papers/companies that are trying to integrate more complex AI into their games.

I've just graduated with an Artificial Intelligence degree and I, like the OP, had a strong interest in AI for games and tried quite a bit to find relevant research projects/work in this. The key point is that the purpose of AI in games is not to build the most realistic experience, but always to make the game fun. The reading I did revealed that most pathfinding was simple A* search, and the whole idea that you can 'cheat' in the game (ie the NPCs actually know where the character is even if they're not meant to) makes most of the 'AI' trivial. I read a nice piece on AI in Left4Dead2 where basically the whole dynamically generated levels/difficulty are simply hard coded based on map and player position.

I suggest you go do more research then, AI is very math heavy matrix algebra, bayesian statistics, calc etc. Look at the AI in crysis it has machine environmental learning in it.
HDS:
http://web.media.mit.edu/~jorkin/aibooks.html

I would say the most content heavy books on those links and most definitely not the ones that deal exclusively with game programming, but rather NLP, planning, and data mining (reading some of those textbooks now actually). Like you correctly say, AI is very math heavy and a lot of success stories come from Bayesian statistics. But Bayesian statistics is really only applicable when trying to reason about uncertainty with prior assumptions (see no free lunch theorem) - how and why would need to apply this in a game environment? The game engine knows where the player is at any time, what he is doing, and this is relayed to the engine to make the game fun. I don't think games strive to be realistic by trying to imitate human behaviour - realism is conveyed more through lovely textures, graphics, and story as a work of art.

I think these misconceptions come from the idea that AI always has to do with imitating human behaviour. In the academic world today, there is at least a clearer breakdown as to what AI encompasses, see: http://en.wikipedia.org/wiki/Artificial_intelligence#Goals. The big fields of research today are very different from what you may expect, especially in the state of the art approaches.

What I was getting at is this:
Never heard of the phrase. Nor has Google apparently. Taking a wild guess I'd say you mean:
http://en.wikipedia.org/wiki/Reinforcement_learning

RL is a very active field of research with some great applications to robotics, trading agents, but I really haven't seen RL applied to computer games. It's a very tough field and I doubt extending NPCs with RL is going to make a game like Crysis fun.

The application of AI to Crysis I think you are referring to is the Bayesian skill rating system TrueSkill. This is a nice example of some 'hardcore mathematics/stats needed for AI'. In multiplayer games, like Crysis or Halo, teams of players play matches and either win or lose. Image that the 'skill' of a player is a hidden variable in all of this - you directly observe game outcomes and want to infer individual skill, to ultimately assign ratings to players on which the game system can carry out match making. It is clear that there is true uncertainty now that can be handled by a Bayesian framework.

Here is the TrueSkill paper: http://research.microsoft.com/pubs/67956/NIPS2006_0688.pdf , here's a more readable FAQ: http://research.microsoft.com/en-us/projects/trueskill/faq.aspx. To me this is a 'true' application of AI (more specifically machine learning) to games, and an active area of research in CS at Cambridge & Edinburgh. If I remember correctly, it is used by almost all Xbox Live games. To understand how TrueSkill works and build on it, you're going to need to understand probabilistic graphical models, bayesian statistics & machine learning, on top a basic foundation of LA. I haven't seen of work like this developed in industry - Microsoft is of course supported by Microsoft Research which does cool stuff like this.

Good luck HDS and if you are still interested in AI, consider it more generally, there are a lot of fun areas like Natural Language Processing(see: http://www.inf.ed.ac.uk/teaching/courses/inf2a/schedule.html), Machine Learning (http://www.inf.ed.ac.uk/teaching/courses/iaml/), Vision (http://www.inf.ed.ac.uk/teaching/courses/av/), Agent based AI (http://www.inf.ed.ac.uk/teaching/courses/inf2d/timetable/), Machine Translation (http://www.inf.ed.ac.uk/teaching/courses/mt/lectures.html). Consider doing a masters in AI at a UK university like Edinburgh, there are scholarships, and a very strong place to do a PhD.

Edit: Here's a MDP used in Civ: http://people.csail.mit.edu/camato/publications/LearningInCiv-final.pdf