Scientists Use Quake 2 To Study the Brains of Mice 185
An anonymous reader writes "In this week's issue of Nature, scientists from Princeton University trained mice to navigate around a virtual environment using a setup that resembles a combination of a giant trackball and a mini-iMax theater displaying a virtual world rendered using a modified version of the Quake 2 open source game engine. (Here's the academic paper, subscription required.) They hold the mouse's head still atop a giant trackball, which the mouse turns by running. The scientists use the rotations to move the mouse around in the virtual environment, and when he reaches certain places, he gets a reward. Because they are able to hold the head still, they can stick microscopic glass electrodes into individual neurons in the hippocampus of this mouse as it 'navigates.' They find the neural activity that resembles activity during real life navigation, and learned new things about the inputs and computations that are going on inside these neurons, which weren't known before. No word as of yet whether the scientists plan on giving the mice control of the gun. Wonder whether John Carmack ever envisioned this when he opened up the Quake code?"
Re:Damn! (Score:4, Interesting)
Open Source Helping Humanity (Score:5, Interesting)
This is a great example of open source really helping humanity learn new and interesting things.
It's great to see that the tools provided by open software can really help speed up research.
I wonder what would happen if old versions of adobe photoshop, 3ds max, or cubase were left to open source for research purposes. What kind of discoveries would scientists make with programs like these?
Hmm.... (Score:1, Interesting)
Human controlling fighter jet in fighter jet.
Human controlling fighter jet in console type game system.
Mice controlling console type game system.
Is no-one else seeing the dangerous possibilities?!
Frame rate? (Score:3, Interesting)
So what was the frame rate on the mouse's brain? Could it run crysis/linux?
Re:Damn! (Score:4, Interesting)
Hmm, yeah, and they should train the mouse to fire the gun by rewarding it – first, make it find a target and operate the "gun firing" switch (whatever that is). Then train it to find a person and target them and "fire". Finally, put it up against armed opponents, give negative feedback if the mouse gets shot, and see what happens. Will the mouse go for positive feedback (by shooting opponents), will it hide/flee from the opponents (to escape the negative feedback it learns to associate with being shot by them), or will it be intelligent enough to create a new self-defense mechanism whereby it learns that by shooting the opponents before they can shoot it, it can both avoid the negative feedback and at the same time get positive feedback?
Acknowledgements (Score:3, Interesting)
If anyone is wondering, the authors did thank Carmack and id software:
Acknowledgements We thank E. Chaffin for help with mouse behaviour,
J. Carmack and id Software for providing the Quake2 code
New definition (Score:3, Interesting)
Gives you a new definition for a head mounted display.
Re:Damn! (Score:2, Interesting)
Once you have these smarter mice identified, give them two guns and see if they learn to choose the right one for the situation.
Re:Damn! (Score:5, Interesting)
I suspect learning to use the in-game offense/defense mechanisms would tax the mouse's brain far beyond its capacity already without adding the complexity of choosing between different weapons.
Although that does make me think of another interesting thing. Mice are more of a foraging creature... they look for food and hide from predators, fighting back only when cornered. A FPS, on the other hand, lends itself to predatory tactics... seeking your prey and killing them without being killed by their defensive tactics. Is it even possible for a mouse to learn to exhibit predatory behavior using a reward system, if their prey tendencies to flee or hide are being simultaneously triggered as their target fights back? If a bunch of mice were put into such a simulation, would they all hide from each other? Would they actively seek and kill each other? Or would there be some of both, where some mice tended to behave in a predatory manner (aggressive personalities?) while others hid (passive/survival)?
forget Virtual Reality... (Score:3, Interesting)
Let the mice steer the beetles!
Re:Damn! (Score:3, Interesting)
The implication is that cats would be better FPS players than mice. The problem of course is that cats are sit-and-wait predators, sitting quietly and waiting for their prey to move before suddenly striking. In other words, campers. So playing on a server against a bunch of cats would probably suck. Dogs, which are pursuit predators, would be probably be better opponents. Probably better at playing cooperatively as well, given the pack mentality.
Re:That just seems sick. (Score:3, Interesting)
Re:Damn! (Score:3, Interesting)
That's the key, though... what you described is exactly a "foraging" tactic. Find the lever, hit the lever, get the cheese.
When the toon is able to interact with the mouse in a negative way (unpleasant) in addition to a positive way (pleasant), it's no longer just a maze or puzzle to solve. That's what would be so fascinating: would the mouse have the intelligence to plan and attempt to get the positive stimuli (reward) while avoiding the negative stimuli. (What you mentioned – the "target" is moving – is another point of interest, to be sure, but I'm more fascinated by the fact that it has both positive and negative associations to the mouse.)
It's sort of like the whole clever mouse vs. mousetrap thing. Is the clever mouse actually able to figure out how to get the bait without springing the trap? It's probably more sheer luck... the mouse is either lucky or it's dead, which doesn't lend itself to learning from mistakes. An immediate-feedback, non-lethal yet still unpleasant stimulus is much more interesting, because then we can see how the mouse learns and, in particular, how it reacts to a "lever" which is directly associated with both reward and discomfort.
A mouse that ends up either happy or unhappy is a whole lot more interesting to do follow-up research on than a mouse that is either happy or dead. Did it learn anything, and how will it react next time? Will it react so as to try to get the ideal outcome, or merely in the interest of avoiding the negative one?