workshop-oee4 chat record of 1st session 3:00-5:45
15:02:38 From Norman Packard to Everyone : 1st session chat archived at Chat session 1 (12:30-2:30)
15:02:56 From Yuri Lavinas to Everyone : Thanks!
15:04:51 From Norman Packard to Everyone : How about using OEE not to just simulate, but to engineer new OE systems “in real life”?
15:06:01 From Olaf Witkowski to Everyone : @Norman I’m unclear on what that would mean?
15:06:18 From Olaf Witkowski to Everyone : What would be the difference?
15:06:22 From Dragana Laketic to Everyone : I’d also vote for “engineering” rather than only “simulating”.
15:07:35 From Norman Packard to Everyone : simulation typically connotes existing independent of interactions with “real life”. engineering connotes such interactions as being primary.
15:07:50 From Lisa Soros to Everyone : I think by “simulation” we mean something that runs in a computer, in case that is part of the confusion. It’s not necessarily about strictly simulating biology or something like that
15:08:05 From Dragana Laketic to Everyone : But engineering assumes a bottom-up approach which is far from straightforward. It would also have to account for the properties of the physical substrate.
15:08:51 From Andy Lomas to Everyone : Don’t humans using Minecraft introduce evolution into Minecraft?
15:08:59 From Norman Packard to Everyone : @Dragana: “far from straightforward”… yes! that is why it needs research focus.
15:09:36 From Norman Packard to Everyone : @Andy: yes.
15:09:45 From Olaf Witkowski to Everyone : @Norman I guess I like any hybrid systems, so I’m in 
Of course, it would be nice to identify which factors produced open-ended properties afterwards.
15:10:08 From David King to Everyone : There are mechanisms in MC that could be used for evolution, like red switch circuits.
15:10:35 From Dragana Laketic to Everyone : @Norman: yes, and if we are to engineer anything in that way, we have to account for the physics of the substrate.
15:11:33 From Norman Packard to Everyone : @NGuttenberg suggested entropic measure of explanation to escape “human bias” component?
15:11:38 From Odin to Everyone : I probably has missed it. If Open Ended mean “goal-less” then the open-ended in “open-ended evolution” may have a different meaning. The ecology of the simulation creates goals (the fitness is measured there). I thought that open ended was about “leave it happen and find its own problem to solve”
15:13:15 From Odin to Everyone : Great!!!
15:14:27 From Dragana Laketic to Everyone : Entropic measure and information theoretic approach - along Levels of Description (Mac Gregor & Fernando, 2005)
15:16:12 From Sharon Minsuk to Everyone : Odin, I think those are two different things. No doubt related though!
15:16:15 From Susan Stepney to Everyone : keep as much as poss; amount you keep limited by computational capacity?
15:16:48 From Nicholas Guttenberg to Everyone : One thing I’ve looked at is something like ‘predict as much as possible, such that the things you predict can’t predict each-other’
15:17:00 From Nicholas Guttenberg to Everyone : its the second bit of that sentence that causes trouble
15:18:03 From Emily Dolson (she/her) to Everyone : Lisa’s dissertation is actually a great read, though!
15:18:15 From Susan Stepney to Everyone : necessary and sufficient
15:18:18 From Dennis Wilson to Everyone : I want to read your dissertation @lisa! where can we find it?
15:18:22 From Nicholas Guttenberg to Everyone : in information theoretic terms, it would be that you have some x(t), transform it to a set of variables z_i(t), then you want to maximize I(z(t), z(t+tau)) while minimizing I(z_i(t+tau), z_j(t+tau)). Or to minimize redundancy.
15:19:40 From Dennis Wilson to Everyone : found it "Necessary Conditions for Open-Ended Evolution" by Lisa Soros
15:19:43 From Emily Dolson (she/her) to Everyone : @Dennis: here’s a link "Necessary Conditions for Open-Ended Evolution" by Lisa Soros
15:19:54 From Emily Dolson (she/her) to Everyone : hah, you beat me to it
15:20:21 From Kevin Frans to Everyone : Another interesting point re: info theory is what signal you are trying to predict/recreate. Like predicting the grains in an oatmeal bowl gives a very high mutual information but is kind of uninteresting on a higher level
15:21:22 From Nicholas Guttenberg to Everyone : Yeah this kind of breaks down if you aren’t talking about a chaotic system
15:22:28 From Nicholas Guttenberg to Everyone : That gets you to more elaborate schemes, like trying to maximize the contrast between models rather than maximizing MI. E.g. you want an abstraction where a model that sees a broader picture can predict it, but a model that sees a narrower picture cannot. But then you’re imposing human biases
15:23:01 From Mark Bedau to Everyone : Lisa: A condition that is necessary for OEE in one system might not be necessary in another system. Your thoughts?
15:26:23 From Olaf Witkowski to Everyone : Impressively efficient talk Lisa:)
15:26:40 From Jan Kim to Everyone : I wonder whether Lisa’s three conditions are necessarily straightforward to establish? They sure are where the criteria are explicitly incorporated in the rules of the system, but they can emergently / implicitly result from the axiomatic rules of a system and that may lead to room for contention.
15:27:02 From Carlos Gershenson to Everyone : Information/entropic measures are statistical, so by definition they are simplifying/generalizing an actual system. This has advantages and disadvantages. Still, for any arbitrary measure, I’m sure one can find a system that constantly maximizes it while it varies endlessly. Whether we consider this OEE or even find it interesting is another matter…
15:28:49 From Mark Bedau to Everyone : Lisa: It looks like original system might be stabilizing by the end of the run you showed (which took 1 month wall clock time!)
15:29:13 From Olaf Witkowski to Everyone : (Mark and Lisa’s talk pointed to the question I really like about the universality of math in any system, which I’ll ask on Slack not to clog this chat)
15:29:35 From Sharon Minsuk to Everyone : Would just like to point out that “survival of the fit enough” is what Darwinism actually is. “Survival of the fittest” is an oversimplified popularization.
15:29:52 From Nicholas Guttenberg to Everyone : @Carlos in the end, human levels of interest are also a statistical measure of sorts. So we might iteratively create more and more stringent standards that exceed our own ability to enjoy the distinctions…
15:31:43 From Nicholas Guttenberg to Everyone : I think there’s something here about whether we can create an agent which is better at being bored than humans
15:31:49 From Charles Ofria to Everyone : @Sharon: Agreed; different meaningful fitness measures all boil down to number of offspring produced. I love the idea of a minimal criterion, but ultimately individuals still have more or less of a chance of meeting that criterion, so there is a range of fitness values.
15:34:12 From Stefano Tiso (RUG) to Everyone : @Sharon I agree. To everyone, isn’t the explicit definition of an explixicit cost/fitness function from which then we deduce the number of produced offspring a problem when looking for OE? Shouldn’t we just let individuals reproduce and leave up to them to find the solutions that will lead to produce more offspring than others?
15:35:17 From Carlos Gershenson to Everyone : @Sharon: reminds me of this book by Klaus Jaffe “EVOLUTIONARY BIO-DYNAMICS:
From ‘the survival of the luckiest’ to an evolutionary economics” Ethan Frome
15:35:20 From Luc Caspar to Everyone : @Stefano Is there any work related to this idea already? It is something I am really intereted in further developping.
15:35:51 From Lisa Soros to Everyone : @Sharon yep also agree, just in the context of evolutionary computation the “survival of the fittest” narrative seems to be particularly strong
15:35:54 From Yuri Lavinas to Everyone : Wow, they look so nice!
15:38:06 From Kevin Frans to Everyone : @Stefano @Luc @Lisa IMO fitness we use in evolutionary algorithms isnt really a true fitness, it’s more like an arbitrary value we use for selection — true fitness should be more along lines of “how much does my genome survive over generations”
15:39:28 From Stefano Tiso (RUG) to Everyone : @Luc, mmm I cannot think of some seimnal paper off the top of my head(also trying to follow the presentaiton). We are trying to follow this philosophy in my lab (MARM lab from the university of Groningen) when creating models for theoritcal evolutionary biology. But to be completely honest, one of the reasons I came to this conference and workshop is because I feel that this kind of approach is most developed in Alife. I will let you know if some paper comes to mind.
15:39:29 From Luc Caspar to Everyone : @Kevin should the selection be based on the genome or rather on the phenotype/behavior? Which one better reflects a better adaptadability to the particular ecosystem?
15:39:35 From Emily Dolson (she/her) to Everyone : @kevin 100% agree. But we can convert evolutionary algorithm style fitness to something closer to biological fitness by looking at expected number of offspring
15:39:38 From Charles Ofria to Everyone : @Kevin: This is a very important point. In biology fitness is measured, which in evolutionary algorithms fitness is assigned and then not always used as the only factor to determine the number of offspring.
15:40:00 From Charles Ofria to Everyone : *while
15:40:29 From Luc Caspar to Everyone : @Stefano Thank you very much. If you want to talk in more details about that I’d be happy to do that on slack.
15:40:51 From Stefano Tiso (RUG) to Everyone : @Luc, sure!
15:42:08 From Olaf Witkowski to Everyone : @Kevin on fitness: completely agree, and our work on meta learning vs. evolvability might be relevant here
15:43:16 From Jan Kim to Everyone : What is the sequence complexity measure? (Sorry, may have missed it)
15:44:44 From Emily Dolson (she/her) to Everyone : he didn’t go into a ton of detail, but I think they’re basically looking at the number of sites in the genome that actually affect fitness
15:45:19 From Charles Ofria to Everyone : We use a few different complexity metrics, but indeed that one we do knockouts and see if fitness decreases.
15:46:40 From Dennis Wilson to Everyone : @Matthew, thanks for the talk. could you elaborate more on the plans for neuromorphic hardware?
15:46:41 From Stefano Tiso (RUG) to Everyone : @Kevin, to my knowledge arbitrary fitness functions are also the norm in theorethical evolutionary biology, and I would push myself to say that there are also very few empirical systems where “true” fitness (and not fitness components) can be measured. That is why I think using AL models could be a good starting point to look at more emergent fitness measures.
15:48:40 From Nicholas Guttenberg to Everyone : From my point of view, the ‘utility’ of the fitness concept is that it predicts what will be around at asymptotically long times. But the interesting thing about minimal criterion, multi-niche systems, etc, is not that a particular fitness measure is exactly flat, but that those things cause the traditional fitness to become non-predictive at long times
15:48:50 From Roberto Gallotta to Everyone : @Kevin if we were to use true fitness for selection, however, would we have any push for diversity? Or is it a wrong assumption on my end?
15:49:57 From Kevin Frans to Everyone : I guess one viewpoint is that real-fitness is a function of arbitrary-fitness + selection-function. So a genome with high arbitrary-fitness might actually have the same real-fitness as a fellow genome, if the selection function used for reproduction is following minimal criterion and they both qualify
15:50:14 From Norman Packard to Everyone : usually biologists think of ‘measuring fitness’ in terms of measuring reproduction rates.
15:52:13 From Emily Dolson (she/her) to Everyone : Agree with Norman - we can’t “use true fitness for selection” because true fitness is an effect of the selection process, not something we can know ahead of time
15:52:51 From Kevin Frans to Everyone : ^ agree
15:53:22 From Stefano Tiso (RUG) to Everyone : But isn’t what happens in chromaria some sort of true fitness?
15:53:31 From Yuri Lavinas to Everyone : Can’t we evolve fitness?
15:53:35 From Luc Caspar to Everyone : If spreading ones genome as much as possible is the goal of the game, would reproductive success be a better measure than fitness or simple reproduction rate?
15:54:30 From Jan Kim to Everyone : @Charles @Emily @Matthew thanks for clarifying this. I have questions about this approach – in addition to the redundancy issue which you mentioned at the end of your talk, there’s also an issue of fitness effects that are conditional on environment (e.g. genes for metabolising other sugars may appear neutral in an environment rich in glucose), and effects that are conditional on gene regulation (e.g. genes requiring a trans-activating factor may appear neutral if that factor is never expressed, either due to mutation or environmental conditions).
15:54:43 From Stefano Tiso (RUG) to Everyone : @Norman I would say biologist think of “measuring fitness” in terms of measuring components of fitness, such as reproduction rates.
15:55:39 From Norman Packard to Everyone : @Stefano… ok…
15:56:03 From Emily Dolson (she/her) to Everyone : @Stefano I don’t think that what happens in chromaria is “true fitness” in any sense that doesn’t also exist in other systems. All ALife systems have true fitness, it’s just that some also have a property called “fitness” that is different from fitness in the biological sense
15:56:05 From Anya Vostinar to Everyone : @Luc biologists do generally consider reproductive success what is relevant, though obviously rate is part of that, but survival of offspring, and survival of offsprings’ offspring are also important
15:58:23 From Hiroki Sayama to Everyone : @Anya 's point is critical. “Fitness” heavily depends on time scales. Here is a tale from almost two decades ago…Phys. Rev. Lett. 88, 228101 (2002) - Relationship between Measures of Fitness and Time Scale in Evolution
15:58:51 From Anya Vostinar to Everyone : And then runs into a halting problem, so you have to draw the line somewhere and accept that you’ll never know the true fitness until the world ends
15:59:09 From Anya Vostinar to Everyone : Or extinction
15:59:31 From Stefano Tiso (RUG) to Everyone : @Anya, exactly what I meant.
15:59:50 From Kevin Frans to Everyone : Yes, I think it’s the same parallel to measuring learning ability or evolvability; you need to pick a time horizon for any comparison to make sense
16:00:55 From Luc Caspar to Everyone : So as @Nicholas suggested, the fitness used in evolutionary computation could be interepreted as an approximation of this unobtainable measure?
16:01:27 From Nicholas Guttenberg to Everyone : If you use the Price equation as a definition of fitness, you can generalize to a multi-generational Price equation, in which case the fitness at horizon tau becomes something like the partial derivative of the population carrying a trait at time tau with respect to the population carrying that trait at time 0.
16:01:49 From Nicholas Guttenberg to Everyone : (there’s a normalization I’m forgetting here)
16:01:49 From Anya Vostinar to Everyone : I usually think of it as explicitly defined (evolutionary comp) versus implicit
16:01:54 From Emily Dolson (she/her) to Everyone : @Jan yes, those are absolutely important issues! A simple and imperfect solution is to assess this metric in whatever environment the individual experienced during its lifetime. but that’s not ideal, because this sucks environmental complexity into individual complexity
16:01:55 From Nicholas Guttenberg to Everyone : Nathaniel Virgo did some stuff with this
16:03:00 From Olaf Witkowski to Everyone : OpenOpenAI
16:03:21 From Luc Caspar to Everyone : talk about meta-meta-naming
16:03:33 From Anya Vostinar to Everyone : I like the ‘cacophony’ part, has interesting implications haha
16:11:56 From Jan Kim to Everyone : @Emily yes, these make matters complicated. It’s scarily easy to get all these elements into a rather simple model of evolution – I’ve seen them all in my (now rather ancient) LindEvol models, and grappling with these issues therefore seems familiar to me. Maybe now there are some new solutions (or approaches)?
16:15:09 From Jan Kim to Everyone : About Ken’s robot gait example: I think one issue is the focus on the individual robot (with a fixed morphology). That sure results in a finite space of possible gaits. If robots can interact (and thereby create niches for each other, as Lisa might put it), there might be much more scope for open-endedness.
16:15:59 From Kevin Frans to Everyone : I like this idea of building divergent objectives out of many convergent objectives; i.e. optimize to solve a bunch of tasks but those tasks keep changing
16:16:44 From Yuri Lavinas to Everyone : Here is where you start evolving the simuations too 
16:16:50 From Yuri Lavinas to Everyone : simulations*
16:17:08 From Luc Caspar to Everyone : And everything within the simulation should be evolved as well
16:17:13 From Olaf Witkowski to Everyone : @Yuri #spoilers
16:17:33 From Luc Caspar to Everyone : If anything is left to the designer it could be seen as a limiting factor
16:18:14 From Yuri Lavinas to Everyone : Ops, but the presentation makes you think that already. It drives you there, not my fault
16:18:21 From Yuri Lavinas to Everyone : hahah
16:18:53 From Olaf Witkowski to Everyone : (@Yuri I’m half-joking, but I guess POET is close enough to your description)
16:19:17 From Olaf Witkowski to Everyone : (also, there should be no suspense in science)
16:19:30 From margaretasegerstahl to Everyone : Re Ken just now: Using scales and ranks when it is difficult to find more precise metrics for what one wants to describe or study, is a good working strategy in the study of complex systems.
16:21:09 From Yuri Lavinas to Everyone : @olaf I got the half-joke :D. and… yeah, I remembered POET too.
16:21:39 From Yuri Lavinas to Everyone : Ah, there it is
16:22:03 From Yuri Lavinas to Everyone : But it makes a lot of sense to evolve the environment, as Earth keeps changing as life on it.
16:24:33 From Mark Bedau to Everyone : Ken: In the MCC framework (mazes and solvers) there seems to be no qualitative novelty; seems blocked by the framework. But (as Susan Stepney emphasized earlier) this qualitative novelty is important. Your thoughts?
16:24:48 From Emily Dolson (she/her) to Everyone : if you think about the “environment” as a combination of the biotic and abiotic environment, having it coevolve with the agent isn’t as un-biological as it might initially sound
16:24:55 From Sina K. to Everyone : serendipity is such a great word
16:25:16 From Anya Vostinar to Everyone : It seems like the environment/agent relationship is fairly similar to a host/parasite relationship
16:25:41 From David King to Everyone : The environment gets a vote.
16:25:58 From Kevin Frans to Everyone : @Ken, I like the POET philosophy. One thought is that the environment mutation creates a ton of tasks but they are all somewhat similar (bipedal walker), do you have ideas for how to make those tasks more diverse ala tasks on Earth?
16:26:01 From Yuri Lavinas to Everyone : @Emily, agreed
16:26:52 From Norman Packard to Everyone : @Emily: agree
16:27:08 From Dragana Laketic to Everyone : @Emily Dolson: imho, coevolution seems as the most biological and natural approach.
16:28:02 From Carlos Gershenson to Everyone : It becomes very relevant where to draw the boundary between system and environment, as discussed in the Illusions of Self session… a relevant aspect is that one can draw different boundaries at different scales. Search processes sometimes get “stuck” (not open-ended) because they cannot jump scales (up or down).
16:28:04 From Roberto Gallotta to Everyone : @Emily: agreed, static environments are a simplification / depend on the frame of reference
16:28:14 From Luc Caspar to Everyone : @Dragana agreed
16:28:34 From margaretasegerstahl to Everyone : In the spirit of autopoiesis - the coupling between system and its environment (Structural Coupling), how they affect each other and change with time, must be acknowledged in the study of living systems if we are to really understand them.
16:28:47 From Emily Dolson (she/her) to Everyone : @Carlos: great point, totally agree
16:29:15 From Olaf Witkowski to Everyone : @Mark this adhocness-avoidance might address part of your question
16:29:15 From Anya Vostinar to Everyone : There also seems to be parallels to how parasites have been shown to increase host diversity by indirectly selecting for novelty, which then also led to higher host complexity
16:29:39 From Emily Dolson (she/her) to Everyone : @Anya yes!!!
16:30:03 From Dragana Laketic to Everyone : Static environments can be considered only if the rate of change of the environment happens at much larger time scale - which is in most cases not the case and especially if “accidents” are considered which may take the whole process to some new direction.
16:30:17 From Alex Lalejini (he/him) to Everyone : @Anya, agreed! Very good point!
16:33:23 From Tim Taylor to Everyone : ANNECS - a nice idea
16:33:39 From Hiroki Sayama to Everyone : Question to Ken: Does this coevolution framework imply that our “general” intelligence is not general at all, but rather, it works simply because it co-evolved together with our environment? Maybe the concept of AGI is an illusion…
16:33:54 From Anya Vostinar to Everyone : I wonder if ANNECS could be applied to niches…
16:34:07 From Dennis Wilson to Everyone : how is novelty measured for ANNECS? is any difference counted as “novel”?
16:34:16 From Carlos Gershenson to Everyone : Question for Ken: Could one generalize your results about the failure of curricula to academic education? (I guess it depends on what purpose we assume education has)
16:34:29 From Emily Dolson (she/her) to Everyone : @Anya ooo, cool idea! I think the biggest challenge there is just automatically identifying niches
16:35:05 From Anya Vostinar to Everyone : Well, yes, it wouldn’t be easy, but could be fun
16:35:49 From Jordan Pollack to Norman Packard(Direct Message) : hi Norman, I planned to do something like you but never got around to it. but I did buy a domain deepevolution.net which now points to your site. Deep evolution I meant to be systems which keep learning and growing thru millions of generations without converging.
16:36:07 From Susan Stepney to Everyone : maybe biology/earth look so open is because they ar so big – no chance to explore it all, but is still actually bounded
16:36:12 From Tanner Lund to Everyone : @Hiroki I tend to agree
16:36:24 From Olaf Witkowski to Everyone : Great talk, Ken, thanks!
16:36:25 From Lisa Soros to Everyone : agree with @Susan for the most part, but also I think that’s ok!
16:36:59 From margaretasegerstahl to Everyone : Ken, Thank you for the very interesting talk!!
16:37:03 From Alastair Channon to Everyone : Thanks Ken. Great talk, showing how to apply and push forward OEE ideas
16:37:19 From Olaf Witkowski to Everyone : I think @Kevin and @Mark had questions?
16:37:44 From Olaf Witkowski to Everyone : And @Hiroki
16:37:49 From Steen Rasmussen to Everyone : @Susan: Agree
16:37:50 From Kevin Frans to Everyone : Yes mine is along the lines of, What do you think are ways to evolve more diverse environments without degenerating into something unrecognizable?
16:37:58 From Emily Dolson (she/her) to Everyone : Yeah, that’s fair!
16:38:00 From Rory Greig to Everyone : Question for Ken: it’s fantastic to have PATA-EC as a domain general novelty criteria, but it seems quite tied to a setup with separate agents and environments. What would a domain general novelty criteria look like where the environment is composed of other agents?
16:38:37 From Stefano Tiso (RUG) to Everyone : @Rory
16:38:46 From margaretasegerstahl to Everyone : Any Newson whether this OEE4 chat can be saved/posted somewhere?
16:38:52 From Jan Kim to Everyone : @Carlos I also was wondering about how Ken’s comments on the POET curricula seemed to apply to real-world university curriculum design too.
16:38:59 From margaretasegerstahl to Everyone : news
16:39:57 From Luc Caspar to Everyone : @margareta the first session is already on slack in the #workshop-oee4 channel.
16:40:19 From Alastair Channon to Everyone : Question for Ken: Isn’t the obvious answer to include agents in the environment, to give a selection feedback loop?
16:40:47 From Norman Packard to Everyone : Could Agent-Environment dichotomy be replaced by sufficiently complex population of agents that collectively create environment?
16:41:38 From Carlos Gershenson to Everyone : Keith Downing generalizes evolution and other AI tasks as search problems. It seems open-endedness is related to the halting problem, i.e. if you halt, then you are not open ended… See Undecidability and Irreducibility Conditions for Open-Ended Evolution and Emergence | Artificial Life | MIT Press
16:41:40 From Anya Vostinar to Everyone : It seems like the walking task is part of the problem, it’s not obvious how they would interact, but if you switched to other tasks, it would be easier
16:41:50 From margaretasegerstahl to Everyone : We functioning types of organisms introduced through the “big” Major transitions in biology may be seen in the context of attractors and stable states…not very familiar with this myself, but came across Karl Friston’s Markov Blanket last winter…
16:41:59 From David King to Everyone : @ Susan, fully agree. I think the universe is bounded, it’s just a really, really, big space to evolve into. So maybe you are right, there could be an eventual end to OEE.
16:42:33 From Takashi Ikegami to Everyone : Thanks KEN!
16:42:37 From Lisa Soros to Everyone : @ Anya if you want to look at a similar system that involves more tasks (but is still fundamentally limited in some ways like POET), my student made a POET-like system that runs on games from the General Video Game AI suite: [2007.08497] Co-generation of game levels and game-playing agents
16:43:00 From Dragana Laketic to Everyone : If there are more agents of the same kind around, the environment of one individual agent with be different (and more complex) than otherwise which will lead to different behaviour in an agent.
16:49:04 From Kevin Frans to Everyone : This is like meta-learning for science
16:49:44 From Norman Packard to Everyone : phase transitions: shared driving force was measurement of critical exponents. NOT exemplars in terms of particular systems.
16:50:06 From Tanner Lund to Everyone : @Kevin you may like this metascience study about clusters of researchers: https://www.metascience2019.org/presentations/james-evans/
16:51:47 From Dennis Wilson to Everyone : shared focus on some problems (or experimental protocols) has been somewhat halting progress in deep learning (ie, MNIST/ImageNet). how do we avoid this?
16:52:00 From Hiroki Sayama to Everyone : I see values of this idea of creating shared exemplars, and Norman’s web portal is already a great initiative for this. Meanwhile, I also see risks (and self-contradiction for us studying open-endedness). We should definitely discuss
16:52:45 From Tanner Lund to Everyone : It’s not just shared goals/exemplars/etc., but the amount of connectivity/collaboration between researchers or labs.Diversity of ideas vs combined might focused on one idea
16:55:51 From Emily Dolson (she/her) to Everyone : I think these are all fantastic ideas!
16:56:35 From Hiroki Sayama to Everyone : Maybe “Open-Endedness Explained”? > Carlos Gershenson
16:56:44 From Emily Dolson (she/her) to Everyone : (although I do understand Hiroki’s concerns about approaching them too single-mindedly)
16:56:57 From margaretasegerstahl to Everyone : If we workshop participants were being observed right now, regarding how we are acting in this chat and what we write and the talks that are going on at the same time - do you think there is some kind of OEE instance happening here, occurring in real time?
16:58:06 From margaretasegerstahl to Everyone : Thank you Mark, extremely important talk!
16:58:07 From Susan Stepney to Everyone : I want to go, and to stay!! Can I clone?
16:58:14 From Dennis Wilson to Everyone : is there a channel on slack to continue discussion?
16:58:30 From Lisa Soros to Everyone : @ Dennis #workshop-oee4
16:58:45 From Dennis Wilson to Everyone : thanks @lisa!
16:58:50 From Sina K. to Everyone : @margaret I think so, we’re certainly making up sets of new problems and trying to find agents that can solve them!
16:58:54 From Carlos Gershenson to Everyone : Please share the highlights of this discussion on slack. See you later!
17:02:25 From Susan Stepney to Everyone : Gravitationl waves as the physical analogue of colour?
17:02:39 From Nicholas Guttenberg to Everyone : So there’s another side of it, which is that if you don’t have strict constraints then open-ended algorithms that fill up from the bottom get stuck
17:02:56 From Nicholas Guttenberg to Everyone : like, POET on environments which are encoded by RNNs and which are played by RNN agents
17:03:04 From Nicholas Guttenberg to Everyone : You get billions of password-guessing games
17:03:46 From Nicholas Guttenberg to Everyone : thats why I think its important to have criteria that aggressively get bored once they can generalize a pattern
17:04:07 From Susan Stepney to Everyone : artificial boredom algorithm?
17:04:34 From Nicholas Guttenberg to Everyone : Yes exactly
17:04:54 From Nicholas Guttenberg to Everyone : There are deep connections between curiosity algorithms in ML and novelty search in ALife
17:05:35 From Nicholas Guttenberg to Everyone : and the sort of ‘adversarial’ curiosity methods have an agent that basically gets to challenge the other one to say ‘I’ve seen it already, show me something new’. But there has to be grounding so that the agent who gets bored has to prove it somehow
17:05:44 From Mark Bedau to Everyone : Norman: re phase transitions, at the focus on critical exponents in an example of a Kuhnian exemplar.
17:05:44 From Susan Stepney to Everyone : environment goes from geosphere, biosphere, technosphere…
17:06:41 From Olaf Witkowski to Everyone : @Nicholas Is there a way to describe such boredom in an implicit way, in the way @Ken described it earlier in his talk?
17:07:13 From Nicholas Guttenberg to Everyone : @Olaf that’s the challenge. Things like Schmidhuber’s compression progress or learning progress sort of get at it, but they’re hard to make concrete
17:07:21 From Dragana Laketic to Everyone : Could we look into changes in the behaviour of agents / entities at certain level which can consistently be observed if some “novelty” / innovation happens at a higher level? In Susan’s example, if it is observed that certain entities react / interact / reproduce with each other and than it is detected to be due to this novel property - the colour? Some repeated patterns at he level of agents / entities at their level due to existence of the novelty / innovation? (Is this along the lines of constraints imposed by novel entities?)
17:07:52 From Susan Stepney to Everyone : that’s why we made innovation relative to current model, rather than original – boredom is baked into the definition
17:08:13 From Sharon Minsuk to Everyone : My intuition (hopefully to be acted on in the future) is that the problem is not the algorithms, but the substrates (worlds, individuals, genomes) on which the algorithms act. Those substrates are inherently “boring”. We have a, dare I say “paradigm”, that says complexity will emerge from the simplest possible units. I’m not sure that’s true. We may have taken that too far.
17:09:31 From Olaf Witkowski to Everyone : @Nich exactly, and hopefully something implicit that comes out of a distributed agent system
17:09:44 From Susan Stepney to Everyone : This q about environment seems to be “how do we make the environment OE?”
17:09:59 From Nicholas Guttenberg to Everyone : The multi-level thing I think is kind of important I think. It’s almost two different kinds of OE
17:10:24 From Stefano Tiso (RUG) to Everyone : @Lisa, so would it be a good idea to do something similar to pre-biotic evolution? Where individuals emerge from “inanimate” entities that are already present in the environment?
17:10:26 From Nicholas Guttenberg to Everyone : there’s one idea which is sort of like ‘really complex programs are still programs’, but another is the major transitions point of view where the OE-ness is because a new level of agent comes into existence constantly
17:10:31 From Luc Caspar to Everyone : @Lisa I think that if you want to go to that extreme, it means that everything has to be evolvable and you’d have to forget the deliniation between environment and agent
17:10:34 From Susan Stepney to Everyone : Even with the agent gaits – they’ll never learn to fly
17:10:44 From Nicholas Guttenberg to Everyone : or even with the agent gaits, they’ll never form a society…
17:10:53 From Lisa Soros to Everyone : @ Luc I think that’s ok, you just have to find the right representation (which is a huge “just”)
17:10:53 From Susan Stepney to Everyone : true!
17:11:25 From Tanner Lund to Everyone : To play devil’s advocate: you could argue that environments are just collections of other agents that we’re not currently focused on (and at various scales).
17:11:34 From Penny Faulkner Rainford to Everyone : but if you added water to the environment they might learn to swim
17:11:37 From Luc Caspar to Everyone : @Lisa that would be true. And you would also “just” to find a powerful enough computer
17:11:47 From Nicholas Guttenberg to Everyone : @Tanner I think the thing is, you need to be able to play with those boundaries over the course of a single run
17:11:55 From Nicholas Guttenberg to Everyone : E.g. the agent-env separation has to change
17:11:55 From Susan Stepney to Everyone : Energy/entropy/fields etc are environment, and don’t easily fit as agents
17:12:07 From Tanner Lund to Everyone : @Nicholas yeah that makes sense
17:12:27 From margaretasegerstahl to Everyone : Re what Steen is saying right now: We need a Manhattan project, and not come out but with a holistic view (metatheory) that integrates the main pieces we are repeatedly tossing around.
17:13:27 From Hiroki Sayama to Everyone : I fully agree Ken
17:13:47 From Susan Stepney to Everyone : @Ken completely agree
17:14:03 From Penny Faulkner Rainford to Everyone : Agree, not all of us work with evolution or even learning processes
17:14:08 From Emily Dolson (she/her) to Everyone : that Watson and Szathmary paper suggests that the probably are indeed the same problem
17:14:11 From Steen Rasmussen to Everyone : @Ken: Agree, we need to invite more areas into the discussion/exploration
17:14:29 From Hiroki Sayama to Everyone : (I never programmed GA or NN even once in my whole life)
17:14:42 From David King to Everyone : I see a lot of parallels to research on emergence. A lot of similar questions and problems.
17:14:56 From Mark Bedau to Everyone : By the way, evolution does not mean only Darwinian evolution.
17:15:24 From Tim Taylor to Everyone : So maybe the next workshop will be called OEE for Open-Endedness Engineering??
17:16:33 From Olaf Witkowski to Everyone : @Hiroki, are you saying we should attend more AI conferences?
17:16:56 From Hiroki Sayama to Everyone : @olaf not necessarily…
17:16:58 From Tim Taylor to Norman Packard(Direct Message) : Hi Norman. I have something to add but I see how to raise my hand (maybe because I am host?)
17:17:10 From margaretasegerstahl to Everyone : I think we should instead expand, and somehow try to bring Artificial/engineering adn biological OEE under same umbrella regrading modelling and theoretical study of OEE.
17:17:11 From Nicholas Guttenberg to Everyone : The message I heard is that we all need to work on our Twitter follower count
17:17:28 From Hiroki Sayama to Everyone : @Olaf rather, if we party hard and have fun with ALife/OE, then they will come to our party
17:17:33 From Tanner Lund to Everyone : Blogs, videos, newsletters…
17:17:48 From Sina K. to Everyone : If you build it, they will come…
17:18:03 From Alastair Channon to Everyone : I like the way Hiroki put it
17:18:33 From Emily Dolson (she/her) to Everyone : is this when I should put in a plug for contributing to the new ALife Encyclopedia/wiki?
17:18:40 From Charles Ofria to Everyone : I’m a bit iffy about removing “evolution” from OEE. Specifically, I worry that we are in danger of diluting our discussions and having them go in very non-artificial life directions.
17:19:02 From Luc Caspar to Everyone : @Emily yes, I would know more about that.
17:19:07 From Olaf Witkowski to Everyone : @Hiroki I think so too but like @Nich just said: maybe we need some followership to make our party as open as possible (and my twitter followership is not something I cultivated much)
17:19:15 From Luc Caspar to Everyone : *would like to
17:19:35 From margaretasegerstahl to Everyone : Rephrasing: …Open ended systems: their behaviour and evolution?
17:19:40 From Lisa Soros to Everyone : time to start twitch streaming alife simulations
17:19:41 From Olaf Witkowski to Everyone : @Sina I like to believe so
17:19:56 From Kevin Frans to Everyone : @Lisa XD
17:19:57 From Alastair Channon to Everyone : I think there is something foundational in OEE among OE processes. Are there OE processes that are not the product of OEE?
17:19:58 From Nicholas Guttenberg to Everyone : I think streaming ALife development would be great. You already have people livecoding shader art
17:20:01 From Susan Stepney to Everyone : @Takashi – what was the name of that philosopher?
17:20:25 From Nicholas Guttenberg to Everyone : @Alastair I’d argue that you see things like that with GANs in machine learning. They’re basically coevolutionary systems without the evolution.
17:20:27 From Takashi Ikegami to Everyone : Quentin Meillassoux
17:20:27 From Sina K. to Everyone : Twitch streaming some science might actually be an interesting idea
17:20:31 From Olaf Witkowski to Everyone : Personally, I’d propose working workshops in OEE
17:20:36 From Emily Dolson (she/her) to Everyone : @Luc awesome! Come check out the #workshop-web channel on the conference slack! or just send me a DM (trying not to spam this chat too much)
17:21:09 From Lisa Soros to Everyone : to the point of @charles, I think there still needs to be a special place for OE_E_ in particular even as we’re trying to work on open-endedness in general
17:21:36 From Alastair Channon to Everyone : @Lisa @Charles Agree with that
17:21:41 From Sharon Minsuk to Everyone : @Lisa that sounds about right.
17:22:07 From Tanner Lund to Everyone : I’m a newcomer to ALife. Experiment on me.
17:22:12 From Luc Caspar to Everyone : @Emily Will do for sure.
17:22:34 From Richard Löffler to Everyone : Can all this knowledge about alife, agents and evolution not be used to directly socially engineer a followership? :)))
17:22:37 From Charles Ofria to Everyone : @Lisa: exactly. I’m not saying that we should ignore open-endedness as a general phenomena (there are a lot of interesting question there!), but we also need to prevent too much dilution from core ALife.
17:22:53 From Penny Faulkner Rainford to Everyone : @Alistair I think there are certainly OE aspects of chemical reations even with out biochem or evolution level chemistry. I think evolution is the result of OE not the cause of it.
17:23:49 From Alastair Channon to Everyone : @Penny good point, it all started with sparse clouds of hydrogen and helium
17:23:52 From Susan Stepney to Everyone : move from evo-devo to get structure, to NS-devo?
17:24:08 From Sharon Minsuk to Everyone : To me as a person interested in evolution, my question is whether I can learn something conceptually useful about open-endedness from other processes, like learning or immune systems. And then, since I’m interested in ARTIFICIAL evolution, I want to know is there any other artificial system that has achieved the open endedness that artificial evolution has not?
17:25:26 From Kevin Frans to Everyone : I think using OE to solve AI optimization problems is pretty feasible, especially the idea of using serendipity to bypass certain hard tasks. It’s less on the alife side but it’s still a very useful way to use OE
17:25:30 From Emily Dolson (she/her) to Everyone : @Penny agreed! especially when you think about David Baum’s idea that chemical ecosystems gave rise to evolution as we traditionally think about it. I’ve been chatting with him and Alyssa Adams about OEE as a paradigm for understanding the origin of life
17:26:42 From Tim Taylor to Everyone : I agree with @Charles - I’m a little wary of calling everything open-endedness - I don’t think we fully understand the relationship between OE and OEE at this stage
17:26:47 From Lisa Soros to Everyone : @emily if you have any good resources could you send them my way? I’ve been interested in that stuff too lately, curious what you’ve been found relevant from more of a biology background
17:27:24 From Olaf Witkowski to Everyone : @Charles @Lisa I’m not sure that there is a big risk of dilution, I think we would connect with more topics in ALife by (sometimes, not always) dropping the last E in OEE
17:27:26 From Susan Stepney to Everyone : like model systems in biology – weed, mouse, frog?
17:27:49 From Penny Faulkner Rainford to Everyone : @Emily, absolutely! This is what interests me is the creating of evolution from open endedness I think that for me is the really interesting emergent feature of at least some open-ended systems.
17:27:52 From Lisa Soros to Everyone : @Olaf I agree, I just think the “sometimes, not always” part is important
17:28:26 From Olaf Witkowski to Everyone : Sure, I don’t see it going anywhere
17:28:28 From Emily Dolson (she/her) to Everyone : @Lisa will do!
17:28:30 From Thomas Willkens to Everyone : Q: What might a “brute force” algorithm look like for traversing the space of all possible state spaces in a given universe? Imagine or construct a universe small enough where such a traversal is possible. Could/should this be done from an information-theoretic point of view?
17:28:54 From Nicholas Guttenberg to Everyone : Embedded or external?
17:29:36 From Nicholas Guttenberg to Everyone : If the universe is an ergodic system, thermodynamics is the algorithm. But I’m not sure that the universe is actually ergodic
17:29:39 From margaretasegerstahl to Everyone : Is there anybody else here, focusing on biological OEE and how it can be connected with study of OEE in ALife? Pls contact me e.g. via slack, it would be nice to know!
17:30:50 From Emily Dolson (she/her) to Everyone : @Margaretasegerstahl Mike Wiser has done some work on that
17:31:13 From Susan Stepney to Everyone : @Nicholas – universe not ergodic, because expanding?
17:31:37 From Tanner Lund to Everyone : Because of absorbing barrier(s), I assume
17:31:39 From margaretasegerstahl to Everyone : @Emily Thanks
17:31:57 From Charles Ofria to Everyone : @Margaret I was also about to point you to Mike Wiser; he’s a graduate from Rich Lenski’s lab and is interested in open-endedness in experimental evolution.
17:31:59 From Alastair Channon to Everyone : @Margaretasegerstahl you might find this paper interesting: https://people.reed.edu/~mab/publications/papers/ecal97.pdf
17:32:36 From Jan Kim to Everyone : The examplar idea somehow reminds me of reading a kind of plea I read (in the preface of one of the ALife proceedings from the 90s, I think), that the stream of papers reporting yet another neural network (YANN), or yet another complex system (YACC) should stop.
17:32:37 From Nicholas Guttenberg to Everyone : @Susan I agree
17:33:21 From margaretasegerstahl to Everyone : @Charles Ofria, @Alastair Channon, thanks
17:33:49 From Charles Ofria to Everyone : @Olaf: My biggest concern is having AI folks dive into this area and take approaches very different from ours. We can always re-think if that happens, though.
17:35:26 From Penny Faulkner Rainford to Everyone : @Charles then perhaps the answer is to wait a little while and put together this sort of exemplar set up so we have a stronger idea of this is what it is and how we work
17:35:34 From Olaf Witkowski to Everyone : @Charles sure, I understand the fear there, but I’m also fine with AI people “stealing” from us, since the nature of who we are is this (scruffy, quoting Inman ) scientists, and a big pool of rich new ideas
17:35:40 From Susan Stepney to Everyone : need low level stability to allow higher level diversity?
17:35:41 From margaretasegerstahl to Everyone : @Charles Ofria That is partly why I am also very anxious for reaching some new stage in respect to more holistic understanding of OEE
17:37:21 From Luc Caspar to Everyone : @Charles it might also be a good opportunity to firmly define our identity and goals.
17:37:48 From Sharon Minsuk to Everyone : Re the comments about all cells being alike… they do have commonalities, fundamental ones, but there is also such a divergence of cell types built upon those commonalities. Both within an multicellular individual, and across the tree of life. I don’t agree that a cell is a cell is a cell. It depends where you put your attention.
17:38:21 From Hiroki Sayama to Everyone : @margareta Fully agreed. We need ways to recognize and interpret multiscale patterns even before applying any measurements
17:38:34 From Jan Kim to Everyone : My impression is that if there’s any level of organisation at which a system is open-ended, the system, as a whole, is open-ended. E.g. the facts that the genetic code is frozen (mostly, anyway), major metabolic pathways are more or less frozen, and human geneticst may be frozen (or evolving at a much slower speed than culture), do not conflict with stating that carbon-based life on Earth is evolving open-endedly.
17:39:18 From Susan Stepney to Everyone : once you get sufficiently complex, every instance is individual at the detailed level – a key q is what level of abstraction to work at/model without losing anything critical
17:39:21 From Steen Rasmussen to Norman Packard(Direct Message) : Hi Norman, are you still in Italy?
17:39:34 From Norman Packard to Steen Rasmussen(Direct Message) : yes!
17:40:10 From Alastair Channon to Everyone : @Jan Kim This is an excellent and important point.
17:41:21 From Tanner Lund to Everyone : @Olaf agree
17:41:56 From Susan Stepney to Everyone : like the Humies – Bio-competitive open-endedness?
17:42:57 From Jair Pereira to Everyone : I think biology is coupled together with the psychical world. I mean, the atoms are door opening to molecules, those are door opening to amino acid, then protein. It just happen that the structure of cells converged to what we know today
17:43:27 From Charles Ofria to Everyone : Regarding AI folks getting into open-endedness, I’m fine with that in general (and happy for them to steal any ideas from our field), but I mostly want to keep our papers and workshops more focused on the Alife side of things. As I said, this may not be an issue, but it is something we should keep in mind.
17:43:44 From Alastair Channon to Everyone : Thanks Norman
17:43:52 From Steen Rasmussen to Everyone : Wonderful workshop!
17:43:54 From Lisa Soros to Everyone : Thanks, everyone!
17:43:58 From Tim Taylor to Everyone : Yes, thanks everyone for a great workshop!
17:44:02 From Jan Kim to Everyone : Many thanks all
17:44:06 From Luc Caspar to Everyone : thanks for this wonderful workshop
17:44:07 From Penny Faulkner Rainford to Everyone : Thanks all, good chat!
17:44:07 From Hiroki Sayama to Everyone : Thank you very much organizers of this fantastic workshop!!
17:44:09 From Kevin Frans to Everyone : Thank you all, good to see eveyronye!
17:44:14 From Christen Patrik to Everyone : Thanks!
17:44:15 From Emily Dolson (she/her) to Everyone : thanks everyone!
17:44:17 From Olaf Witkowski to Everyone : Thanks for organizing this!
17:44:19 From Stefano Tiso (RUG) to Everyone : Thank you everyone!
17:44:19 From Olaf Witkowski to Everyone : Bye all!