Braaaaaains
Sep. 12th, 2009 06:08 pmI like living in the future. I was reading a blog a few weeks ago that reminded me of Heaven & Earth, the most excellent Scott Kim puzzle game that never got the publicity it deserved, so I dug out my copy and replayed it. Then, hitting the end, I had to Google to verify that the victory screen was actually the end and not an Easter Egg for a further challenge, which led me to a forum where someone was giving love to Bug Brain.
Wow. I've never had the opportunity to play much with practical logic circuit design. Electronics books get esoteric in a hurry (even when they're written for kids) and math books stay abstract. I've always wanted a platform that combined the relevance of robotics with the puzzle-oriented worldview of games like The Incredible Machine. Evidently, someone wrote this game for me in 2000 but didn't go to the trouble of telling me about it.

This is an example of an intermediate-level puzzle, designing an artificial brain for a worm that is supposed to crawl forward until it hits a door, and then back up for a few seconds (to allow space for the door to swing out), and then return to the forward crawl. The brain contains input nodes (the red circle, which fires if the head is currently against an obstacle) and output nodes (the four blue circles, which direct the motor skills for raising the middle segment, lowering it, grabbing the ground with the head segment, and grabbing the ground with the tail segment when they are charged). The green circles represent neurons, which do the very elementary computation of measuring whether the accumulated charge of all of the incoming synapses (which can be individually weighted and even negatively weighted to inhibit charge) and send a charge down all of its outgoing axons if the total is greater than a threshold you define. You can use that mechanism to create the common logical gates: the two neurons in the middle of the brain are an AND and OR gate, and combined with the two supplemental yellow nodes they form the logic of an XOR gate. In addition, you can set synapses to slowly lose their charge over time, which let you form natural constructs like the feedback loop in the upper left that allows the brain to briefly "remember" that it bumped its nose a few moments ago and the two neurons with mutual decaying inhibitors in the lower left that form the cadence that it steps to. As the worm "chapter" progresses, you get control of more input and output controls and have to design more complex brains that ultimately have you negotiating a 2-D landscape filled with obstacles to find mushrooms that you can sniff out.
The game is far from perfect. The learning curve can be very steep for some problems, and while there are hints and even full solutions available you just wind up building the author's model instead of conceptualizing your own. It would also be nice if you could "chunk" common components like this XOR gate into a single "bundle" of neurons because it can get tricky to read the brains as they grow. The game contains a final module that hit on the real apparent strength of neural networks, which is their capacity of adaptive self-programming, but the examples they give are either simple or don't solve their problems dependably. But the biggest flaw is that it's such a delight that it's over too quickly.
Wow. I've never had the opportunity to play much with practical logic circuit design. Electronics books get esoteric in a hurry (even when they're written for kids) and math books stay abstract. I've always wanted a platform that combined the relevance of robotics with the puzzle-oriented worldview of games like The Incredible Machine. Evidently, someone wrote this game for me in 2000 but didn't go to the trouble of telling me about it.

This is an example of an intermediate-level puzzle, designing an artificial brain for a worm that is supposed to crawl forward until it hits a door, and then back up for a few seconds (to allow space for the door to swing out), and then return to the forward crawl. The brain contains input nodes (the red circle, which fires if the head is currently against an obstacle) and output nodes (the four blue circles, which direct the motor skills for raising the middle segment, lowering it, grabbing the ground with the head segment, and grabbing the ground with the tail segment when they are charged). The green circles represent neurons, which do the very elementary computation of measuring whether the accumulated charge of all of the incoming synapses (which can be individually weighted and even negatively weighted to inhibit charge) and send a charge down all of its outgoing axons if the total is greater than a threshold you define. You can use that mechanism to create the common logical gates: the two neurons in the middle of the brain are an AND and OR gate, and combined with the two supplemental yellow nodes they form the logic of an XOR gate. In addition, you can set synapses to slowly lose their charge over time, which let you form natural constructs like the feedback loop in the upper left that allows the brain to briefly "remember" that it bumped its nose a few moments ago and the two neurons with mutual decaying inhibitors in the lower left that form the cadence that it steps to. As the worm "chapter" progresses, you get control of more input and output controls and have to design more complex brains that ultimately have you negotiating a 2-D landscape filled with obstacles to find mushrooms that you can sniff out.
The game is far from perfect. The learning curve can be very steep for some problems, and while there are hints and even full solutions available you just wind up building the author's model instead of conceptualizing your own. It would also be nice if you could "chunk" common components like this XOR gate into a single "bundle" of neurons because it can get tricky to read the brains as they grow. The game contains a final module that hit on the real apparent strength of neural networks, which is their capacity of adaptive self-programming, but the examples they give are either simple or don't solve their problems dependably. But the biggest flaw is that it's such a delight that it's over too quickly.