devtrackers.gg
8 months
ago -
EVE Online
-
Direct link
Transcript (by Youtube)
| 0s | foreign |
|---|---|
| 4s | can real-world behavioral relationships |
| 6s | between predator and prey be applied to |
| 9s | Eve online here to answer that please |
| 12s | welcome capsulear zakaidon Esther mayor |
| 31s | ah there it is |
| 33s | awesome |
| 35s | all right so I'm an evolutionary |
| 37s | biologist uh it's a lot of fun because I |
| 41s | get to wear lots of different hats some |
| 44s | applied aspects I look at the commercial |
| 46s | Turtle harvesting so in the state of |
| 49s | Maryland you can commercially Harvest |
| 50s | snapping turtles and so when I started |
| 53s | grad school |
| 54s | they didn't have any regulations I had |
| 57s | very few regulations in the turtle |
| 58s | Harvest so uh being the the state then |
| 63s | uh convenient working group so they got |
| 65s | scientists and Harvesters and advocacy |
| 68s | groups and they put them all into a room |
| 70s | and they said come up with some kind of |
| 71s | regulations that would make this Harvest |
| 74s | sustainable and then me being the bright |
| 75s | out grad student in the room they all |
| 77s | looked at me and said go collect data uh |
| 80s | so I did I went out with commercial |
| 81s | Turtle Harvesters they met all the |
| 83s | measured Turtles hundreds a day that |
| 86s | they usually collect and then it led to |
| 88s | some kind of uh regulations |
| 91s | uh some observational aspects I look at |
| 94s | the evolution of signaling traits these |
| 96s | are lizards that the males will defend |
| 99s | territories |
| 101s | and so if an intruder comes in |
| 103s | they'll square off with each other and |
| 105s | then do a whole bunch of push-ups and |
| 107s | they'll show off these bright blue |
| 108s | bellies and probably convey some kind of |
| 111s | information with uh fighting ability |
| 113s | they don't want to fight fighting is |
| 115s | dangerous it leads to injuries probably |
| 117s | death and that's not such a good thing |
| 119s | so they try to figure out who's going to |
| 120s | win before they actually get to fighting |
| 122s | but with these lizards they can see into |
| 125s | ultraviolet they can see things that |
| 127s | that we can't and so I wanted to see if |
| 129s | there's any components in the signals |
| 131s | that they convey that had UV or if UV |
| 134s | would be important at all and it turns |
| 136s | out it's it is pretty interesting |
| 137s | there's a lot of UV information that |
| 139s | they have in these patches that they |
| 142s | might use to evaluate fighting ability |
| 146s | and then my favorite part about being a |
| 148s | evolutionary ecologist is the |
| 149s | theoretical aspect not specifically |
| 151s | Predator prey Dynamics uh predatory |
| 153s | Dynamics goes all the way back to the |
| 155s | latko Volterra models where they modeled |
| 157s | population uh Cycles yeah Cycles based |
| 161s | on population size and they can be |
| 164s | cyclical they can reach an equilibrium |
| 167s | and it's a really really good starting |
| 169s | point for things it's also an alliance |
| 172s | is anybody here from latke Volterra |
| 176s | nobody |
| 177s | it's kind of an old one all right |
| 180s | um but what are the limitations with |
| 182s | population cycling is that it doesn't |
| 185s | convey everything with this particular |
| 187s | conflict conflict there's behaviors that |
| 190s | take place and we want to see what |
| 192s | behaviors might be the best |
| 194s | um uh to determine this conflict and |
| 196s | optimize some kind of payoff right so |
| 199s | Game Theory comes into this game theory |
| 202s | is basically a study of conflict it's uh |
| 204s | just three major components you have |
| 206s | players that can be one player or or |
| 209s | many players |
| 211s | there's strategies that these players |
| 213s | can take to try to optimize what we call |
| 216s | a payoff so what what is the thing that |
| 218s | they're trying to to do the best at |
| 221s | um one of the Classic Game Theory uh one |
| 223s | of the classic games out there is called |
| 224s | the prisoner's dilemma the idea here is |
| 227s | that you and your partner just got |
| 229s | arrested for robbing a bank and they |
| 231s | have you in a room by yourself and your |
| 234s | partner's in another room somewhere else |
| 235s | and uh they say well if you confess |
| 238s | we'll let you go and put everything on |
| 240s | your partner and so you can either stay |
| 243s | quiet and if your partner stays quiet |
| 245s | then you'll both get off probably |
| 247s | scot-free uh but if you stay quiet and |
| 250s | your partner confesses you go to jail |
| 252s | for a really long time so you have two |
| 254s | players you have different strategies |
| 256s | you can either cooperate or defect |
| 258s | or yeah |
| 262s | so I play this with my students in |
| 265s | evolution class and uh I give them cards |
| 270s | and then they pair off for about 10 |
| 272s | rounds and they get to play either of |
| 274s | the two strategies and it's adorable |
| 276s | because the students usually start off |
| 278s | cooperating everybody's nice by Nature |
| 281s | as much as CCP doesn't want you to be |
| 283s | nice |
| 284s | but there's always that one student that |
| 287s | either kind of gets it or is just a |
| 289s | really mean person or wasn't listening |
| 291s | to my directions and played a random |
| 293s | strategy that starts off defecting and |
| 296s | then so they start accumulating a lot of |
| 297s | points and all the people who are |
| 299s | cooperating feel like suckers right |
| 301s | because this person just took all the |
| 303s | points and they got nothing for it so |
| 305s | about round five or six uh everybody |
| 308s | everybody is playing the the defecting |
| 311s | strategy the the selfish strategy |
| 314s | and so uh in 1977 |
| 317s | there was a tournament that was held |
| 319s | that I wanted to try to figure out what |
| 322s | would the best if if somebody could |
| 324s | write code that could win prisoners |
| 326s | Dilemma to get the best uh payoff there |
| 329s | was about 14 submissions this was |
| 331s | written in Fortran and basic I think uh |
| 334s | they put those uh scripts into a giant |
| 337s | made-frame computer waited overnight |
| 339s | each strategy was played 200 times |
| 342s | against all the other strategies and |
| 344s | then there was another strategy that was |
| 347s | just a random it would select random uh |
| 349s | either defect or cooperate |
| 351s | and then they were surprised by the |
| 353s | winner the winner happened to be written |
| 354s | by by this guy and he uh he's a expert |
| 359s | in peace which is kind of weird because |
| 361s | it's a con studying conflict and |
| 365s | his they expected that the code would be |
| 368s | really really complex this was the same |
| 370s | time that uh they were writing code to |
| 373s | beat people in a big humans in chess and |
| 377s | they thought it would be hundreds or |
| 378s | thousands of lines of code it actually |
| 380s | just happened to be a few lines of code |
| 382s | and basically the strategy was start off |
| 385s | by cooperating and then anytime you meet |
| 388s | that player again you do whatever they |
| 390s | did to you so if they cooperated you |
| 392s | cooperate again if they defected then |
| 394s | you defect again and that happened to be |
| 396s | the best strategy and there were two |
| 397s | tournaments and it happened to be the |
| 399s | best in both of those it's called Tit |
| 401s | for Tat |
| 402s | right uh we see prisoners dilemma all |
| 406s | over the place we see in in climate |
| 407s | change uh so if you have a whole bunch |
| 409s | of countries that get together and |
| 412s | they'll say we want to reduce emissions |
| 413s | right there's maybe one country out |
| 415s | there that says if I don't reduce |
| 417s | emissions I stand to to gain |
| 420s | economically |
| 421s | um and then all the other countries that |
| 423s | have been doing it uh to say well I |
| 424s | don't want to be a sucker so then |
| 425s | everybody ends up not reducing |
| 427s | submissions |
| 428s | uh it's also in dating when you go on a |
| 431s | date for the first time you can be |
| 433s | completely honest with your date and say |
| 435s | everything about you |
| 437s | um but if your date holds a lot of their |
| 439s | stuff back then you kind of feel like a |
| 441s | sucker so everybody on their first date |
| 442s | tends to hold everything back |
| 445s | uh there's a really great example of |
| 447s | prisoners dilemma in World War II uh the |
| 450s | Japanese wanted to reinforce some areas |
| 452s | in New Guinea and they had two routes |
| 455s | that they could send their ships uh |
| 457s | through either a northern route which |
| 458s | was uh pretty stormy and cloudy |
| 461s | um and then a Southern route which |
| 462s | wasn't and the Allied Air Force knew |
| 465s | this the Allied Air Force wanted to |
| 466s | maximize the number of days that they |
| 468s | could use to bomb the Japanese Navy and |
| 472s | so they thought well what would the |
| 473s | Japanese do if they could take the |
| 474s | northern route that has lots of cloud |
| 476s | cover lots of storms that's that's going |
| 479s | to inhibit the ability to bomb these |
| 480s | ships then if they go to the southern |
| 482s | route where if we go north they're going |
| 485s | to have a great time but if we go south |
| 488s | then we'll have lots of days to bomb |
| 491s | them so the payoff is more risky in the |
| 493s | southern route so the Japanese ended up |
| 495s | going to Northern route and and the the |
| 497s | Allies predicted that |
| 500s | uh there's a variation of the prisoners |
| 504s | dilemma it's called the hawk Dove gave |
| 506s | this kind of incorporates some kind of |
| 509s | uh |
| 510s | uh more conflict so uh if you play a |
| 513s | hawk then you'll be more aggressive and |
| 515s | you'll steal resources from who you play |
| 517s | with if you play a dove you won't |
| 520s | um so you could play a risky one because |
| 522s | if you're a hawk and you get paired up |
| 524s | with a dove then you take all the Dove's |
| 526s | resources and nothing happens to you but |
| 528s | it's more risky in that if you end up |
| 529s | meeting another Hawk then uh you're |
| 532s | gonna both lose resources both be |
| 534s | injured and that's that's a problem |
| 536s | right so one of the questions I ask in |
| 539s | in my research is can Game Theory |
| 542s | predict animal behavior can we use Game |
| 544s | Theory to figure out why animals make |
| 546s | the decisions that they do one of the |
| 549s | central tenets in in foraging ecology is |
| 552s | trying to figure out how we can explain |
| 554s | how foragers distribute amongst habitats |
| 557s | right so if you have a bunch of foragers |
| 559s | and you have uh patches with resources |
| 561s | in them how are they going to distribute |
| 563s | amongst those resources and what you'd |
| 565s | predict is that the foragers would go to |
| 567s | Patches based on how many resources |
| 569s | there is so the forger density ratio the |
| 572s | number of foragers should match the |
| 574s | amount of resources availability |
| 575s | available this is called the ideal free |
| 578s | distribution and you can see this all |
| 580s | over the place too if you go to a pond |
| 582s | and you put two people on either end of |
| 584s | the pond and they start feeding bread |
| 586s | out into the pond at different rates |
| 588s | you'll see more geese go towards the |
| 591s | higher rate than the lower rate and yeah |
| 593s | that's the ideal free distribution |
| 595s | but another question is how can we get |
| 597s | non-perfect matching for these patches |
| 599s | one way to do that is to introduce |
| 601s | predation risk or if a cloaky Loki jumps |
| 605s | into your system you might see a bunch |
| 607s | of resource gathers miners move out into |
| 610s | other less risky type of patches |
| 613s | so we wanted to study this using Game |
| 616s | Theory where uh specifically where the |
| 618s | players would affect each other's |
| 620s | decisions your decision depends on what |
| 622s | your opponent is going to do |
| 625s | uh limitations so far is that typically |
| 628s | the research will fix one side or the |
| 630s | other so |
| 632s | um you'll fix the Predator's behavior |
| 633s | and see what the prey do or you'll fix |
| 635s | the the praise behavior and then see |
| 637s | what the Predator does but if you let |
| 639s | them play simultaneously you should see |
| 642s | that the prey will allocate towards |
| 644s | being safe in a refuge or going out and |
| 646s | foraging and getting energy and |
| 648s | converting that energy into babies |
| 650s | and then the Predators should manage the |
| 653s | praise fear this comes from something |
| 655s | called a waiting game the idea here is |
| 658s | that you have a predator that chases a |
| 659s | prey into a refuge that's when the |
| 662s | waiting game starts so the Predator |
| 664s | knows that the prey is in the Refuge it |
| 666s | has to decide how long it's going to |
| 668s | wait for the prey to come out or it can |
| 671s | decide to leave and it can go off try to |
| 673s | hunt some other prey and get the element |
| 676s | of surprise or something like that the |
| 679s | prey also has a waiting game it needs to |
| 681s | figure out I just got chased in there I |
| 683s | know there's a predator out there or at |
| 685s | least was so how long am I going to wait |
| 688s | in this Refuge where I'm safe but I need |
| 691s | to get out there and I need to start |
| 692s | foraging I need to get that energy to |
| 694s | make babies |
| 695s | right one way that was this was tested |
| 698s | empirically uh was by some folks in |
| 700s | Israel they had this enclosure had a |
| 702s | bunch of kiddie pools around uh inside |
| 704s | those kiddie pools were goldfish so in |
| 707s | the center there's a refuge uh they can |
| 709s | hide in the Refuge or they can go |
| 710s | outside where they can uh feed and then |
| 713s | they would let a little Egret go into |
| 715s | the enclosure and little Egret would go |
| 717s | from pond to Pond and try to eat |
| 719s | goldfish so we could see that time |
| 720s | allocation for the prey and also the the |
| 722s | managing of the the Predators |
| 725s | so with our game model uh that's just |
| 728s | think of those those two players |
| 729s | goldfish and a little Egret |
| 732s | um and so with our game we have two |
| 734s | patches uh and the first patch is broken |
| 737s | up into two parts where the prey can |
| 739s | move either from a refuge where it's |
| 740s | safe from predation uh but it can't |
| 743s | forage or it can move out into what we |
| 745s | call the open where it can forage but |
| 747s | it's exposed to predation risk |
| 750s | the Predator can move between the open |
| 753s | where it can hunt prey or it can move |
| 756s | into the environment the environment is |
| 758s | a fixed intake uh if you think of it |
| 761s | that way so uh or just a goldfish bowl |
| 764s | that we can control the number of |
| 766s | goldfish so it's something static |
| 768s | something that the gives the Predator a |
| 771s | choice to either hunt or go to this this |
| 773s | expected amount of food just lets us |
| 776s | wiggle something |
| 777s | so with uh two places that the prey can |
| 780s | go to and two places that the Predator |
| 781s | can go to uh there's four states but we |
| 785s | wanted to be more explicit about uh |
| 786s | information so uh when the praise and a |
| 789s | refuge just on its own time allocation |
| 791s | that's different than it went and when |
| 793s | it was chased in there so we added two |
| 795s | more states where the prey is in the |
| 797s | Refuge because it just got attacked by a |
| 799s | predator right and then the a predator |
| 801s | can be either in the environment or the |
| 803s | open |
| 804s | with six states we can create a |
| 806s | transition Matrix uh it looks something |
| 808s | like this so State R23 is just uh while |
| 813s | the Predator is in the open the prey |
| 815s | moves from The Refuge to the open so |
| 817s | that's the rate that the prey leaves the |
| 820s | Refuge |
| 822s | uh State R uh R sub 2 0 2 is the rate |
| 827s | that uh |
| 830s | uh from state zero to two that's the |
| 833s | rate that the Predator enters the open |
| 835s | right so while the the prey is in the |
| 837s | Refuge we had a couple of assumptions |
| 839s | that there's no moving simultaneously |
| 842s | there's a non-zero probability that they |
| 845s | would move at the same exact time but it |
| 847s | was so small we just considered it zero |
| 849s | we also assume that the Predator would |
| 852s | never leave the open when the prey was |
| 854s | in the open it would never come in Sea |
| 856s | pray and then just decide to leave we |
| 858s | didn't allow that |
| 860s | so with a transition Matrix we can then |
| 862s | calculate eigenvectors to figure out the |
| 864s | proportion of time spent in each of |
| 866s | those Six States so Pi sub 3 in this |
| 869s | case is just the amount of time of the |
| 871s | total time that they're playing uh where |
| 874s | both uh players are in the open |
| 877s | and then from the eigenvectors we can |
| 880s | use uh we can use a proportion of states |
| 883s | to uh |
| 885s | sorry we can use the eigenvectors to |
| 887s | determine Fitness functions and fitness |
| 890s | is just the way that we uh |
| 893s | think of the the payoff right so each |
| 895s | player is trying to maximize their own |
| 896s | Fitness |
| 897s | so the fitness function is this massive |
| 900s | beast but it's just broken up into |
| 902s | several parts so this this first element |
| 904s | is the intake that it gets from the open |
| 906s | uh this is a movement cost we found that |
| 909s | if you didn't have a movement cost the |
| 912s | prey would just act like an electron |
| 913s | field just be a probability field it |
| 915s | moved back and forth really really fast |
| 918s | um and then this is the probability of |
| 919s | survival |
| 920s | the prey Fitness function is very |
| 922s | similar you get the intake from the uh |
| 923s | the open you've got the intake from the |
| 926s | environments and then it also has a |
| 928s | movement cost |
| 929s | so with these Fitness functions we |
| 932s | utilized a tool called adaptive Dynamics |
| 934s | this assumes a monomorphic population so |
| 938s | every individual in this population |
| 940s | exhibits the same exact Behavior but if |
| 943s | a mutant shows up and let's say it has a |
| 945s | slightly better Fitness than everyone |
| 947s | else given enough time that population |
| 950s | should evolve to that particular |
| 952s | Behavior |
| 954s | so we can calculate or we can figure out |
| 957s | one particular Behavior so in this case |
| 959s | with The Predator |
| 961s | the rate that it goes to the open we can |
| 964s | use the fitness functions to calculate |
| 965s | the fitness if it was a slightly faster |
| 967s | Behavior or if it was a slightly slower |
| 970s | behavior and then that will give us a |
| 972s | fitness curve we can calculate the |
| 975s | tangent of that curve and if it's |
| 976s | positive then we'll slightly increase |
| 979s | the behavior in the Next Generation if |
| 982s | the tangent is negative will slightly |
| 985s | decrease the behavior in the Next |
| 987s | Generation |
| 989s | right and then so if you let them play |
| 991s | you'll see something like this we always |
| 992s | wanted them to go to some kind of |
| 994s | equilibrium that wasn't zero a rate of |
| 996s | zero and so you'll see uh them kind of |
| 999s | start to play uh quite a bit and then |
| 1001s | they reached an uh equilibrium |
| 1004s | so we asked a couple different questions |
| 1006s | how does our model stand up to previous |
| 1009s | models right these static Predator prey |
| 1011s | models we wanted to be on par with those |
| 1015s | um what happens if they play |
| 1016s | simultaneously that's the really really |
| 1018s | interesting question and then also |
| 1020s | what's the importance of information |
| 1022s | so we had four different models static |
| 1024s | prey static Predator uh we had a game |
| 1027s | where there was no information so the |
| 1029s | prey went into the Refuge and then |
| 1030s | didn't consider whether it was attacked |
| 1033s | by a predator or not and then the full |
| 1035s | game is where there is information and |
| 1038s | uh where they can play simultaneously |
| 1041s | and so with some of the results uh you |
| 1044s | get something like this uh there's lots |
| 1047s | and lots of interesting things go read |
| 1049s | the paper it's it's pretty fun but I'm |
| 1052s | going to talk about the the third column |
| 1053s | so this is with a high missed |
| 1055s | opportunity cost so think of this like a |
| 1057s | giant Fishbowl full of fish so the the |
| 1060s | Predator always has that option to go |
| 1062s | get a high payoff by going to that |
| 1064s | environment |
| 1065s | we also wiggled the uh lethality of the |
| 1069s | predator so when her Predator is really |
| 1071s | really bad at capturing prey it doesn't |
| 1074s | even bother right that's pretty |
| 1075s | intuitive that makes sense and so the |
| 1078s | the red part of the uh the third graph |
| 1081s | on the right is uh them not playing at |
| 1084s | all but as a predator gets better at |
| 1086s | capturing prey it spends a little bit |
| 1088s | more time uh in the open trying to hunt |
| 1092s | prey but it'll go in if the prey is not |
| 1094s | there it leaves immediately or if it |
| 1095s | goes in it'll attack the prey it won't |
| 1097s | wait at all because it has that high |
| 1099s | missed opportunity cost and this this |
| 1102s | was pretty interesting this is a |
| 1103s | hypothesis I thought a lot about it like |
| 1106s | what where could we see this in nature |
| 1107s | and I was reading a paper that had |
| 1110s | nothing really to do with this |
| 1113s | particular project |
| 1115s | um it had to uh had to do with |
| 1117s | ultraviolet signals in in fiddler crabs |
| 1120s | and one of the things in the methods |
| 1122s | that they talked about was it was really |
| 1125s | easy to capture these fiddler crabs they |
| 1127s | would uh purposely chase them into their |
| 1129s | Refuge and then after just a couple of |
| 1132s | minutes the total cab would emerge and |
| 1134s | they would just grab it and it just that |
| 1136s | in this paper they even say that doesn't |
| 1138s | make sense if there's a predator a smart |
| 1141s | enough Predator it should have evolved |
| 1142s | to just pick these things off like |
| 1143s | popcorn it's a very predictable behavior |
| 1147s | um and so yeah that you shouldn't see |
| 1149s | this in fiddler crabs but our model |
| 1151s | predicts that you can see this if the |
| 1154s | Predator has a high missed opportunity |
| 1155s | cost right if it's something like a gull |
| 1158s | it's going to go in Chase it and then |
| 1160s | just take off right because it's going |
| 1162s | to go somewhere else |
| 1164s | and so the the neat thing is that we we |
| 1166s | see this a lot in uh in Eve online right |
| 1169s | so we do this every day uh you've got |
| 1172s | miners you've got people who like to |
| 1174s | blow up miners we see these kinds of |
| 1177s | behaviors quite a bit uh we also see it |
| 1180s | in the economics so there's a lot of |
| 1182s | great economics talks that took place |
| 1184s | today and I I liked it was it the Tycoon |
| 1187s | one where he's like I just set up shop |
| 1189s | in the systems that I could pronounce |
| 1191s | right it could be a little more refined |
| 1194s | you can use Game Theory uh to try to |
| 1196s | figure out the things that aren't |
| 1198s | necessarily intuitive but might maximize |
| 1200s | your your payoffs |
| 1202s | um but yeah it doesn't have to be very |
| 1205s | mathematical either there's lots of |
| 1207s | different ways to do it |
| 1208s | um or if you're into Transportation so |
| 1211s | uh |
| 1212s | if you need to get uh you know 50 |
| 1214s | billion worth of cargo from one place to |
| 1216s | another uh you can use Game Theory to |
| 1218s | try to maximize maximize this and again |
| 1220s | we do this intuitively we think about |
| 1222s | risks and duration and costs and and |
| 1225s | that sort of thing uh but but game |
| 1226s | theory formalizes that |
| 1228s | um and could put it more into context |
| 1230s | and show you things that maybe you're |
| 1231s | not thinking about right and so how is |
| 1234s | Eve uh online unique it's it's just an |
| 1237s | incredible ecosystem right you you guys |
| 1239s | do this it's as a scientist I want to |
| 1242s | see Predator predynamics out in nature |
| 1245s | but it's really really hard seeing an |
| 1247s | actual predation event is pretty |
| 1249s | impossible you have to be at the right |
| 1250s | place at the right time it usually |
| 1252s | happens very quickly and so you can't |
| 1254s | really observe that in nature and then |
| 1256s | if you take predators and bring them |
| 1258s | into an enclosure and you try to do an |
| 1260s | experiment that way predators are a pain |
| 1262s | in the ass they they don't like to |
| 1264s | behave when you want them to |
| 1267s | um and so yes anybody from CCP here |
| 1271s | can I can I uh do something like project |
| 1274s | Discovery but I don't know project |
| 1276s | Predator Maybe |
| 1278s | I just want to look at data because you |
| 1280s | guys are doing this and I want to I want |
| 1281s | to study you right I want to test my |
| 1284s | hypotheses because this this is just |
| 1285s | very very great way to do this |
| 1289s | um and then yeah if you're really really |
| 1291s | interested in this uh or even starting |
| 1293s | to be interested in this I highly |
| 1295s | recommend this book it's a how not to be |
| 1296s | wrong |
| 1297s | um so Jordan Ellenberg uses lots of |
| 1300s | examples of real life and and uh |
| 1304s | approaches them from a mathematical |
| 1306s | mindset and it's not mathematical at all |
| 1309s | um so if that's not your strong suit |
| 1311s | like like me I'm not uh yeah math isn't |
| 1314s | my strong suit I highly recommend this |
| 1316s | book it'll just make your thinking uh |
| 1319s | much much different uh in in everyday |
| 1322s | problem solving and and that sort of |
| 1324s | thing |
| 1325s | um so yeah that's my talk |
| 1337s | you have any questions feel free to buy |
| 1339s | me a beer I'll |
| 1341s | I'll tell you anything you want so all |
| 1343s | right thank you |