Random Notes from the Artificial Life XI conference, Aug. 4-8, 2008.  <http://alifexi.org/>
Biased point of view by Larry Yaeger

First a couple comments on the conference organization.  This was the first time the main ALife conference was held in Europe, and it was nice to see such a strong, vibrant community of researchers doing excellent work.  Despite being sponsored by the U. of Southampton, the organizers decided to hold the conference in Winchester, UK.  It was a good choice, as the facilities and housing were good, and the town was quaint and beautiful.  Some surprisingly good food was to be had as well, not to mention a goodly number of pubs, including one right across from the main accommodations.

This year's organizers also decided to try an experiment with the structure of the conference.  Instead of a single track of oral presentations and a bunch of posters, they had six parallel oral sessions and no posters.  They also permitted submission of abstracts in addition to full papers.  I was initially aghast at the parallel tracks.  We'd worked hard and made some tough choices to keep it to a single track for ALife X, and I hate missing talks I really want to see because of scheduling conflicts.  However, I must say, I generally much prefer hearing the author present a nice, succinct description of her work, more than I do trying to look over other people's heads at posters I want to read or feeling guilty when I have to walk past a poster I don't.  I still am disappointed I had to miss a number of talks I would have really liked to have seen, and it was a bit troubling when a lot of conversations ran, pretty much in their entirety, "So, did you see X?"  "No."  "Ah."  But in the end, to my amazement, I think I like the parallel sessions better.  I hope that in the future the organization of topics can be improved to reduce obvious conflicts of interest.  And there has been talk of videotaping the talks and putting them on the web, which would pretty much take the sting out of scheduling conflicts, because you could, at least eventually, watch both talks.

The organizers also decided to work with MIT Press to publish the papers online, and not produce a paper volume.  These are fully citable works, with page numbers and an ISBN, just no paper.  And they will be FREE online.  Personally, I'd like to have both on-line and paper versions, but if it's down to one, I'd rather have the papers freely available to everyone online than produce an expensive book.  And since the material is available for free online, it is unlikely MIT Press would sell many paper copies.  So this is probably the wave of the future.  The papers should have been on MIT Press's web site by the conference, but weren't.  They were given out to all attendees on a 256 MB flash drive (nice touch!), and as of this writing are available in a "pre-print" version (though these are actually the final papers) here:  <http://www.alifexi.org/proceedings/>.

The general consensus was that all these conference revisions were mostly for the better.  Regardless, you have to have variation for selection to work.  Kudos to Seth Bullock, Jason Noble, Richard Watson, and Mark Bedau for putting together a noble experiment and a great conference.

Some general trends noticed this year...  Wet ALife (aka synthetic biology) had a strong presence, and can probably be expected to grow.  The role of information theory and complexity continues to grow, with four separate sessions on Information in Complex Systems & ALife, plus one on Entropy, and at least one talk on information-theoretic complexity in one of the Open-Ended Evolution sections (mine).  Network science is being utilized a bit more in ALife as well.

Now a few random comments about specific content.  I did NOT take careful notes, just a few tidbits here and there.  So there are probably talks I liked a lot and have left out because I didn't happen to write anything down at the time.  And, no matter what, I could only see 1/6 of the conference.  (Okay, a smidgeon more than that, since the keynote talks were single-session, but that low fraction does still bother me a bit.)

TU 9:00am (Keynote)
Takashi Ikegami - "ALife is Dead" (should have ended with "Long live ALife") - Great talk.  Some really insightful points...  (paraphrasing)  We should be naturalists for a while before we try to be Darwin and come up with a grand theory.  A singular result may reveal a new general rule.  Said amateur science should be embraced, and pointed to some good, working examples--this is kind of the Web 2.0 version of science, and it makes a lot of sense to engage and benefit from interested individuals world wide.  Production should precede and may ultimately provide reproduction--suggests an interesting priority for ALife research.  Also showed some lovely work with oil drops and chemicals specifically selected to help form and maintain these drops and give them movement.  (A paper on this with Martin Hanczyc was also given and won the best presentation award.)

TU 10:30am (Evolutionary Theory 1)
Powers et al - Group selection increased by coexistence dynamics - Discussed known issues about how variation in altruist/parasite population ratios amongst multiple groups can overcome competitive exclusion.  Presented new results that show how a non-zero altruist/parasite ratio--a minimum level of coexistence--can overcome competitive exclusion.  He only worked with fixed population sizes, and had to assume coexistence, rather than obtaining it as an outcome, so it wasn't surprising that it would overcome competitive exclusion (which is kind of the definition of coexistence).  It would be genuinely nice to extend this work to include variable total population size and see if there aren't some possible evolutionary pressures to maintain coexistence conditions, instead of simply positing their existence.  But I think it's actually important to call attention to this alternative to competitive exclusion.

TU 1:00pm (Neural Systems)
Adami et al - Topological properties... - An interesting method for genetically encoding neural growth characteristics, though results using it so far seem a bit limited.

Soltoggio et al - Evolutionary advantages of neuromodulated plasticity... - Interesting talk.  Makes a strong case for adding neuro-modulation via specific neuro-modulator neurons.  They receive input like any neuron, but their output provides a multiplier on the learning at the regular synapses.  Neuromodulation allowed much more effective learning of four-arm mazes, while they aren't needed for simple two-arm T mazes.  I'm not 100% convinced, but this makes a good case for the value of modulator neurons in a learning system.

TU 3:30pm (Evolutionary Theory 3)
Powers & Watson - The group selection debate and ALife... - Nice, almost tutorial overview of ideas on altruism.  I liked the identification of the equivalence between kin selection and group selection, and between ideas based on genetic similarity and variance between groups.  Mentioned briefly the lack of conflict between selfish genes and group selection, which I've long advocated.  Called for routine identification of type 1 (benefits self, but benefits conspecifics more) vs. type 2 (negative benefit to self) altruism by authors reporting work on altruism.  I'll be sure to do that.

Groom et al. - How epigenetic evolution can guide genetic evolution - In a simple model of methylation, a quickly mutating "methome" degrades (only) a slowly mutating genome.  This pulls the population off of a high-valued, low-fitness peak towards a low-valued, high-fitness peak.  The current implementation has problems, in that it cannot evolve up from the low-valued range.  Richard Watson said they tried a more general approach and had problems with it.  After being inspired by Eva Jablonka, the next morning's keynote speaker, I suggested to Richard that the methome's mutation rate be varied according to how much "stress" an individual experiences, where "stress" is reasonably measured as distance away from the desired state.  I don't know it'll do what they want, but it seems like it might, and Richard said he likes the idea.  I hope they try it.

WE 9:00am (Keynote)
Eva Jablonka on epigenesis.  Epigenetic phenomena are very widespread.  Some seemed to be long-lived (over many generations) as well, but it wasn't clear just how long-lived.  Unless they are essentially immortal (well, as long as the lineage is around), they seem more a curiousity than a true inheritance mechanism.  But it's very clear that there are many paths to at least limited inheritance besides genes.  And there's a lot of cytoplasmic "stuff" along for the ride in gametes that can easily influence evolution.

WE 10:30am (Open-Ended Evolution 1)
Yaeger et al... - Passive and driven trends in the evolution of complexity - Some crazy guy gave a talk about evolving complexity in Polyworld and teasing out natural selection from random genetic drift.

WE 1:30pm (Open-Ended Evolution 2)
Lehman & Stanley - Exploiting open-endedness... - They replaced the usual Fitness function in a Genetic Algorithm with a behavioral Novelty function, and it does significantly better on a task with deception (a maze in which the agent must go the wrong way to ultimately go the right way).  Very interesting.  They claim it is different from an exhaustive search, because it starts at low complexity and gradually adds complexity.  That does distinguish it from a random search, but I don't think it necessarily distinguishes them from an enumerative search that starts from a simple system and adds complexity.  Still, the GA never converged, and they did, for the hardest problem.  I'd like to see the technique applied to some more standard test problems.

Channon (Alistair) - Origins and maintenance of organismal complexity - His Geb system had already demonstrated Class 3b evolutionary activity.  He demonstrates it is now Class 3c.  "Complexity" is defined to be the number of active genes actually contributing to the neural network development.  

WE 3:30pm (Open-Ended Evolution 3)
Kelly et al. - Enrichment of interaction rules in a string-based artificial chemistry - 4-slot truth table defines all outcomes of the interaction of two chemicals--either copying the first of the pair or not.  "Folded" four bits into two and performed matching with the original.

Bedau - The arrow of complexity hypothesis - Really good talk.  Generally proposed the hypothesis that there is an arrow of complexity, but doesn't care whether it is driven or passive, only whether it would be seen with high probability on "replaying the tape".  Said that Gould (in Wonderful Life) discounted even the likelihood of a random variance type of increase in complexity in general, and Mark disagrees with this, and is now seeking evidence.  He said that McShea's simple models fail to serve as evidence because they don't actually address complexity, only a scalar value, that could correspond to anything, and offer no hint of where the complexity might come from, what form it might take, etc.  Said Tierra failed to show any consistent increase in complexity.  Said no ALife system showed increasing evolutionary activity (not sure if he was aware of Channon's claims about Geb).  Wants a single working example, and then wants to see how robust the result is, and wants to confirm that it is complex "for the right reasons".

TH 9am (Keynote)
Andy Ellington (work done with Zack Booth Simpson, his software guy)
Great quote about life:  It doesn't exist.  For poets, not scientists.  There are only replicators of various degrees of complexity.  (Couldn't agree more, and have been saying much the same for well over a decade, though I usually have said life is just a convenient definition for things over a certain degree of complexity, rather than cutting to the chase and saying it doesn't exist at all.)

TH 10:30am (Information in Complex Systems & ALife 1)
Anil Seth - Measuring emergence via nonlinear Granger causality - Really good talk.  Defined G-emergence in terms of a combination of Granger-autonomy (from Seth's 2007 paper at ECAL) and G-causality, and examined flocking boids with different degrees of randomness.  His measure seemed to capture the intuitive sense of emergence, related in an inverted-U fashion with flock coherence.  Someone asked a question that raised a minor concern for me, which is that the way Anil treated his boid-position time-series, converting (x,y,orientation) to a single distance-from-center metric, may induce a lack of causality that shows up as autonomy, hence emergence.  But given that his metric captures the difference between the varying degrees of randomness, I don't think this is a problem, at least as long as you're looking at relative levels of emergence (rather than attempting to put things on some absolute scale, which Anil wasn't trying to do).

TH 3:30pm (Information in Complex Systems & ALife 3)
Anthony et al. (w/Polani) - How global structure is reflected in local structure... - "Empowerment" vs. network global closeness centrality

I really liked this quote from his talk:  "Nature uses only the longest threads to weave her patterns, so that each small piece of her fabric reveals the organization of the entire tapestry." - Richard Feynman.  It precisely echoes Takashi's comment about the singular revealing the general.

Polani et al. - Modelling stigmergic gene transfer - Mostly about demonstrating evolution of an encoding scheme to improve copying accuracy in the presence of noise.  Nice idea, nice talk.  Suggests that DNA came about as a means of providing copying fidelity.

Miconi - Fitness transmission: a genealogic signature of adaptive evolution - Excellent talk.  He didn't refer to McShea (though afterwards he said he knew of his work), but he basically extended McShea's ancestor-descendant test to measuring number of offspring (actually grandchildren) instead of some scalar measure, and demonstrated this measure calculated for his EvoSphere (and maybe Channon's Geb, can't remember), suggesting that it captures the presence of adaptive evolution in a system.  Also demonstrated that Bedau & Packard's evolutionary activity measure can be fooled into giving a false positive by certain specially constructed random phenomena, and made the point that ensuring no such effects are present is difficult in any moderately complex simulation.  I'm not sure how likely such false positives really are, but it would be a good idea to test both evolutionary activity and Miconi's adaptive evolution measure, I think, especially since Miconi's measure also comes with a caveat that it will respond to any means of enhancing reproduction, not just biological mechanisms (including geographical, niche, and societal variations).

FR 9:00am (Keynote)
Peter Schuster - RNA secondary structure and neutral networks

FR 10:30am (Entropy)
Clark & Williams - An adaptive model of marine biogeochemistry in the Archaean - Interesting attempt to model whole of the Earth and run the simulation for millions or billions of years.  Necessarily simple model (well, each individual component is relatively simple--there's still a lot going on).  Unsurprising results (which is a good thing).  Dry, funny delivery.

Dyke - Entropy production in an energy balanced Daisyworld model - Shows that in this simple model, maximum homeostatis is achieved by maximum entropy production.  Simple, but nice.  Hilarious delivery.

FR 1:30pm (Keynote)
George Attard - Synthetic Life: How Close Are We? - Overview of wet ALife and some detail on his synthetic biology experiments.