Development Team

Currently, the development team consists of me:


Docent Dr. Manfred G. Grabherr


I have been programming since I was 13. Not that I wouldn't have done it earlier, but back then, it was difficult to get a hold of a computer. Unsurprisingly, the first complete software program that my friends and I wrote together was a game. A 'physics game', actually, in which the player had to land a multistage rocket on the moon, manually controlling the entire flight and getting constant feedback from a rather simple physics engine. No graphics involved, and in the end, none of us ever managed to land safely on the moon. Which might have been not too bad after all, since we never planned for any return trip. That would have taken us another couple of months, since we could only use the computer on Wednesday afternoons at school.

There is something incredibly fascinating about the combination of the latest advances in science, computer animation, and modeling reality, be it physics, graphics, sound, or behavior. It brings together very different aspects of human endeavors; exploration, achievement, interaction, competition, and perhaps most importantly: the awe of seeing something beautiful work. Granted, these elements can also be seen as integral parts of any software, but what sets games apart is that we play them for the pure fun of it, for the experience alone. While we do that, we learn many things about the world - and perhaps about ourselves as well - without even noticing, which predestines this approach for a number of educational purposes as well. And, having been an engineer and scientist for decades, it is a pleasure to watch kids teaching themselves Java or Lua so that they can write mods for sandbox games such as Minecraft or Garry's Mod. These might be the engineers and scientists of the future, and I think the outlook is more than bright.

But for now, there is one particular area in which further development is desired. Computers have come a long way since I was 13, and now even my phone is orders of magnitudes more powerful than the largest machines that existed back then. This opens up possibilities that we could not have imagined only a few decades ago, and one of them is modeling cognition. Consider the following: in the scientific area that I am working in, computational genomics, advances in biochemistry have made it possible to generate enormously large data sets, for example the whole genomes of dozens or even hundreds of individuals, or the tens of thousands of genes that they express. To put this in perspective, the human genome contains about 3 billion nucleotides, represented by the letters A, C, G, and T. Multiply this number by the number of individuals you sequence, and it is easy to see that the amount of data is, well, large. And in there, you now want to know which nucleotides make some crows black and others grey, how stickleback fish have adapted to different environments, how Darwin's finches change their beak shapes, or what happens in the pancreas of people with type I diabetes.

The classic approach, which is to employ hundreds of graduate students to manually look through the data, is not always feasible for these data sets. This is why machine learning, i.e. the concept of "let the computer figure it out by itself", has become very popular within the past few years in this area. What this approach does, in essence, is to identify distinct patterns in large data sets, partition the space accordingly, and, in some cases, propose models that describe the data. Computationallly, this methodology is by no means cheap, but with proper parallelization and a modern compute architecture, it is very well feasible to process tons and tons of data in reasonable time frames.

Let us now perhaps scale back the amount of data, but let us increase the complexity. While some genomic analysis software already has the concept of space, what happens if we add a component that deals with time? What we get is a system that automatically recognizes patterns in both time and space, and if we add another component that takes into account interaction and the consequences thereof, then these patterns can also be qualified. In other words, we now have a machine that remembers, learns, and determines its own motivations. It is not only able to follow ongoing action, but constantly comparing what is happening with its memory allows it to anticipate the future. It projects what will happen from its experience. Likewise, it can "think" through possible chains of actions and choose the best - or least worst - option, while able to revise strategy if things do not go according to plan. In short: now we have a system that models cognition, even if it is on a crude level compared to humans. But, if we take this system and let it control the non-playable characters in a computer game, even if these are just as smart as ants or chickens, then we get something that adds a whole level to game play.

At least, this is what I am convinced of. And, as always, there is only one way to find out.