Obviously improving the parasitic loss is a goal, powering drones and such off world for construction, exploration and mining resources is the endgame. Imagine it as a temporary source of energy until the generators are up and running.

Apologies for delay, I am in France so it was 2AM when we started I am a blogger writing about very old computer wargames (zeitgame.net). One thing leading to another, I came to write about the people, companies, context, technologies of the era as long as they were related to computer wargames - usually after the After Action Report. Obviously, going through time, I reached your's (and Patricia's) very impressive Fortress, totally forgotten noways but the first game where the AI used neural network and trained against the player (something I expect see again soon, but then we are on the other side of the AI winter). Back then people who knew about computing were I believe extremely impressing. So, my questions: Who were you (you and Patricia, whom I did not manage to contact) in 1983? I am not sure if you had graduated yet or whether you were still students. Where did the idea of making a game using neural network come from? Why SSI? Do you remember a bit how the process went with SSI? Did they make feedback or accept the game as is? It is not really in their usual niche. How did you conceive the game rules. It seems a very simplified Game of Go. Was the ruleset designed to optimize the efficiency of the neural network, or not specifically? What was the split between Patricia in you in the conception/creation of the game? What happened to you after Fortress or rather is there a reason you did not double down with more game / game AI? Thanks !

Apologies for the delay, I am in France and it was 2AM I am a blogger writing about very old computer wargames, and one thing leading to another I started writing about the context around those wargames: people, companies, tech, etc - usually these are after the AAR if you find my blog (putting links would block my message). I reached 1983 and that peculiar game of yours (and Patricia's) called Fortress. It is absolutely unique, the first commercial game using Neural Network opponents, something that outside of research (AlphaZero, MuZero, etc) is still not done, though I expect this to change soon of course. I have several questions to write the article Fortress deserves. - Who were you in 1983? I am not sure if you were still students or had graduated? - How did you have the idea of making a neural network game? Was Fortress a way to showcase the neural network? - Where does the ruleset come from? It looks like a very simplified Game of Go. Was it optimized for the neural network? - What was the split of work between Patricia and you? - Why did you choose SSI as a publisher? - Do you remember if they did any feedback? - What happened after Fortress. Your research history answers a bit this question, but did you consider continuing in the video game industry. You had something no one else had.

@@wargamingscribe I invented the game and learning system as my senior project in EE/CS undergrad at GWU, back in 1977. It was originally called ‘Babel’ because of the falling towers. I’d already done a system that generated artistic geometric patterns based on how people graded each successive iteration. That was for a class taught by Professor Peter Bock, who introduced me to his approach to machine learning. Babel included the invention of the game, the move characterization scheme, and the adaptive learning algorithm. The algorithm was based on a transition matrix representation (treating it like a Markov process), the weighted adjustments were treated very much like the synaptic adjustments you see in neural networks (which weren’t all that new back then). What was novel was adjusting the weights based on how accurately the total collection of evaluations associated with each move predicted the score of the game’s outcome. The later was said to be: based on the expected outcome, rather than simply the end score itself (a sort of predictor-corrector numerical method). More specifically, it applied a root-mean-squared adjustment in distributing the synaptic adjustment weights. I was always viewing things in terms of neural networks, although the field was referred to as machine learning back them. Hebbian learning was introduced back in 1949. I thought of Babel as a compact, stacking version of the game of Go. Rather than surrounding the enemy you built upward as well as spreading out. Stacking was common in the wargames played on hexagonal grids printed on large paper sheets. The five square cross of influence made for interesting interwoven patterns. Unlike Go, the game could be played fairly quickly, which was essential for training the learning algorithm. And it also played up its duality to Go. Later when I was working in a small (2-3 person) company called ‘Iconographics’ we resurrected the game and recast the user interface to one of building castles rather then the ‘towers’ which originally were nothing more than concentric square ‘rings.’ The flags were added to make it easier for people to learn & follow the game. And the initial board position was adopted to keep the player with the first move from dominating the game. The computer graphics interface was originally developed for the Apple II and Atari computers by Patty Denbrook and myself, respectively. Ms. Patty Denbrook also did most of the hard-core programing for the Pointman user interface, and the fully immersive ‘Gaiter’ user interface for: walking-in-place to walk through VR (circa 1999). The cast of pre-trained computer players was also added for the commercial ‘Fortress’ version. Genghis, Maginot, and Vauban were all trained by me from scratch to exhibit aggressive, defensive, and balanced strategy, respectively. The ease and speed of training was the beauty of the system. We never knew how many people appreciated that aspect. The computer game was published by SSI (Strategic Simulations Incorporated) who mainly published computer games with an emphasis on war games. It was published in 1983, under the title “FORTRESS A Game of Strategic Deployment and Fortification.” Fortress appealed to a limited audience who appreciated novel games along classic lines. We got a minor award at one of the game conventions the year it came out. Just to keep things in perspective, ‘Frogger’ was the biggest, most successful computer game to come out that year. And its sales dwarfed that of Fortress. Fortress went on to be sub-licensed by Victor Musical (then a Japanese company) in 1988 & 1990. The game aficionados in Japan appreciated both the game and its adaptive capabilities. But it never became a really popular game. As an undergraduate in computer science, I was mainly interested I simulating neural function. But during my Master’s work Dr. James Foley came to George Washington University. I got to work with him on a number of Computer Graphics and User Interface projects. That’s also when I met Ms. Patty Denbrook. After graduating in 1979 I worked doing computer graphics and user interface design, first at NASA Goddard and then at Honeywell. Patty hired me to work at Honeywell. Later Patty and I and a third student of Jim Foley went off and started our own consulting company doing CG & UI projects, called Iconographics. During that time, we got to work with Paul Brainerd, a head of Atex Corp., who went on to found Aldus Corp. which created PageMaker. Dr. Foley started his own consulting company and we collaborated with him on a number of design and development projects. The consulting work tended to go from feast to famine at times, so we decided to develop Fortress into a commercial product. On top of the game design and machine learning the game also had a solid user interface for the time. Although we got it published it was never a great financial success. At this point I decided to go back to the university to get a doctorate. I majored in Computer Science with minors in Neuro Biology and the Psychology of Perception. You can find my publications and dissertation on Research Gate. For the dissertation I developed a model of how the visual cortex learns to deal with visual transformations through exposure to sequential views. It combines J. J. Gibson's psychological theory of active perception with Hubel and Wiesel's physiological analysis of the visual cortex. It’s my own take on what today would be called deep learning networks. I completed the dissertation and graduated in 1992. The field of neural networks way rather splintered back then and my work did not fit neatly into an existing ‘school of thought.’ so in looking for a job I found more success in returning to the field of user interface development. I was hired by Dr. Rob Jacob to work at the Naval Research Lab, doing ‘novel interface development.’ I got to work with him on his eye tracking based interface. I developed a novel panning and zoom system based on head tracking. And then I got into the field of simulating human locomotion, which I worked in for about fifteen years with five patents and two major interface developments. One, called Gaiter, used full-body tracking, HMDs, and a weapon prop for the Marine Corps. And Pointman took all the sessions we learned from Gaiter and put it into an advanced, yet low cost, seated control system, called Pointman. In retrospect, I should point out that as a teenager I read a lot of science fiction. I particularly liked Arthur C. Clarke's books. He wrote about HAL-9000 in 2001, and about VR in “The City and the Stars." Since I retired in at the end of 2016, I’ve been working on what he covered in “Childhood’s End.” And that’s were my mind is at currently.

​@@JimTempleman Thanks a thousand. I have a handful of follow-up questions, if you allow me: As I understand Babel was already a "final" version of Fortress except for the pre-trained opponents, the (optional) flags and the art / graphics. Did I understand correctly? What was Iconographics doing as a company? Is there a reason you chose SSI as a publisher specifically? About SSI => Indeed, I know them well due to the nature of my blog. I sent Joel Billings a link to this thread and he told me to "say hi for me and tell them we were always impressed with their game". I will report when the article is published.

@@wargamingscribe Yes, Babel was already the "final" version of the game of Fortress with its machine learning system in 1977. We then added the pre-trained opponents, the (optional) flags and the art / graphics, the input controls, and the tournament game facility. We also adopted the song “Men of Harlech” as the theme song and worked short phrases from the tune into the significant game events. (Da-da, da da da) (Free-dom, God and Right!) What was Iconographics doing as a company? We were computer graphics and user interface consultants. Those were the heydays for companies wanting to convert over to more intuitive, graphical user interfaces. Is there a reason you chose SSI as a publisher specifically? I’m sure we submitted our game to them because of their presence in the hobbyist computer and gaming magazines. I no longer remember all of the details surrounding the final choice. And yes, we dealt with Joel Billings. Hearing his name brings back memories. Say hi him for us as well! It’s great to see you so diligently pursuing this particular niche of history. Look forward to reading your article!

Well, great part of that loss comes from the solar panel that collects the power. I imagine once we get close to 100% efficiency with photovoltaic cells that the overall power loss will decrease to maybe 50%