Hacker News new | past | comments | ask | show | jobs | submit login
The Computer Maverick Who Modeled the Evolution of Life (2014) (nautil.us)
48 points by wheresvic3 53 days ago | hide | past | web | favorite | 21 comments



I urge anyone who's interested in that kind of stuff to read "Allure of machinic life" by John Johnston [1]. A fascinating read that touches upon all strands of natural computing (soft / hard ALife, wetware, AI, evolutionary computation, swarm intelligence, behavior-based robotics, 1st and 2nd order cybernetics, theory of self-reproducing automata and whatnot) with a pondering touch of Deleuzian philosophy.

There's no mention of Barricelli though, but another equally interesting book [2] recognizes his pioneering work on artifical symbiogenesis (and that was back in 1953!), and speculates on its importance in the genealogy of computer worms and viruses.

[1]: https://mitpress.mit.edu/books/allure-machinic-life

[2]: https://www.peterlang.com/view/title/58766



Steven Levy's 1992 book "Artificial Life" is well worth a read.


If you find this interesting, I recommend also looking into genetic algorithms, artificial life (alife), cellular automata, Conway's game of Life, falling sand games, and Karl Sims virtual creatures.


Advanced reading search terms:

- Cellular Potts Models

- Evolution of evolvability

- Replicators and the Error treshold


Nanopond more closely matches what you might envision an artificial petri dish looks like: https://www.youtube.com/watch?v=J5yE8Si8rMM


As a counter perspective check out "Introduction to Evolutionary Informatics"

https://www.amazon.com/Introduction-Evolutionary-Informatics...


Intelligent design is neither.


The book is very readable, and has quite a lot of artificial life simulation analysis and mathematical proof. While it is written by well known members of the ID crowd, the book does not advance any claims about intelligent design. The book analyzes why these simulations work, and clearly presents all its claims, so it is quite accessible for the reader to pick apart and check their validity.


Ah, so that's the other part of your misconceptions.

As far as I can see, the point made in the (first part of) the book is that informed algorithms (on average) always outperform random searches and that genetic algorithms are akin to random searches.

The first part is true, the second part isn't. Evolution isn't random, mutation is random. Selection is very much not random, it is informed by the environment. This drastically reduces the size so-called "search space".

Using the term "search" for the outcomes of evolution is a curious choice anyway. Evolution doesn't search for anything in particular, it produces adaptation to an environment. It's a process without end.

As for computer simulations, check out the simulated biped that evolved hopping like a kangaroo:

https://www.youtube.com/watch?v=kQ2bqz3HPJE


The book is not about biological evolution. They are analyzing artificial life simulations like Avida and Terra. Different topic than what we are discussing in the other thread.

The fundamental point is that information is conserved, so if artificial life sims produce information, it must have been inserted into the code at some point, and it is possible to track down these particular insertion points, which the authors do. In information theory the conservation of information theorem goes by various names, such as the data processing inequality and independence conservation.

Anyways, I highly recommend the book. Everything is clearly laid out for your own analysis and application.


> The fundamental point is that information is conserved, so if artificial life sims produce information, it must have been inserted into the code at some point, and it is possible to track down these particular insertion points, which the authors do.

Fair enough. Let's apply this to the video I linked. The authors certainly didn't add "information" about what walking or hopping looks like into the code explicitly. It just emerged. Following your logic, the hopping behavior can't be new information, it must've been contained in the rules of the simulation.

The simulation itself is made up mostly of an off-the-shelf physics library, a couple of muscle-skeleton arrangements (authored by hand), and a fitness function measuring the distance traveled by activating those muscles through weights (the learning variables). Given this environment, walking, hopping and running behaviors invariably emerge. Fascinating!

Adapting this to our world, the laws of physics and the initial arrangement of matter contain all the "information" required for life to eventually emerge in the universe. This leaves you with a "god of the gaps" opportunity: An intelligent designer must've designed those laws and made that arrangement of matter so that it all works out. This is however entirely compatible with Darwinian models.

I don't have an issue with that kind of hypothesis, except that it is unclear what the existence of such a god/designer would imply for our particular lives here on earth.


Have you worked much with evolutionary algorithms? It also seems you think that evolutionary algorithms are somehow analogous to biological evolution. To what degree do you think that is true?


> Have you worked much with evolutionary algorithms?

Not really, like I said, there's usually better algorithms than genetic algorithms. They're fun to play around with, though.

> It also seems you think that evolutionary algorithms are somehow analogous to biological evolution. To what degree do you think that is true?

The general principle - selection and random mutation - is the same.


What gives you so much confidence in GAs?

Also, I am surprised you've followed this discussion for so long. Most people fire off some canned phrases and then disappear.


> What gives you so much confidence in GAs?

Do I sound confident about them? I've said multiple times that they usually aren't very good and that usually you have a better algorithm. Usually, you can apply intelligent design.

The fundamental difference between us is probably that you'd look at life in nature and say something like: "This species is so remarkable, it must've been designed!".

I wouldn't jump to that conclusion. It's only remarkable if it emerged naturally. If it was designed, then I would have to point out all sorts of obvious flaws in the design.


I haven't said anything about design. You are making assumptions. I said Darwinism appears to be completely useless when it comes to practical application, and has been replaced by other mechanisms that are identified in the biological data. In this new thread I said there is a conservation of information law that the book authors apply to artificial life simulations. I am talking about what we know through mathematics and quantitative empirical analysis.

Regarding GAs, the conservation of information applies to how much information we can expect from a single random sample. If we can sample from the entire space of possible solutions, then there is a chance of sampling the optimal solution. The conservation of information applies to the necessary input to the algorithm in order to bring the chance to a certain level.


> I haven't said anything about design. You are making assumptions.

Of course I am, just like you assume "materialism" on my part. I have trouble believing that you're not an ID proponent, even though you're not upfront about it. Otherwise, why bring up books by people with a clear ID agenda?

I'm honestly not sure what your goal is here, anyway. You rarely address any questions that I ask. Making assumptions is all I can do.

> I said Darwinism appears to be completely useless when it comes to practical application, and has been replaced by other mechanisms that are identified in the biological data.

You actually called Darwinism "quackery" and said that it is "dead". This is the least strong version of your position so far.

I don't disagree that Darwinism, in the sense that it describes the origin of species, has no direct practical applications. Nobody is in the business of simulating entire ecosystems for thousands or even millions of years. Using that fact as some sort of criticism of the theory itself seems rather hamfisted.

> Regarding GAs, the conservation of information applies to how much information we can expect from a single random sample. If we can sample from the entire space of possible solutions, then there is a chance of sampling the optimal solution.

Fair enough, but GAs are awful for finding the optimal solution. However, nature doesn't require the optimal solution. It just looks for the next best thing, over and over. GAs tend to converge to some level of fitness and stay there.

In fact, it's likely that a random algorithm outperforms those GAs that can get stuck in local maxima. Again, this has no bearing on nature. Nature is fine with local maxima.


My goal is to discuss the topic of the thread. In this thread the book I mentioned, which is just about artificial life simulations, not biological evolution. I think it will confuse the discussion to try and bring in biological evolution and intelligent design. I've researched GAs and information theory to a fair extent, and am happy to answer questions and discuss on that topic.


> I think it will confuse the discussion to try and bring in biological evolution and intelligent design.

Fair enough, I was looking to continue where we left off in the other thread.

> I think it will confuse the discussion to try and bring in biological evolution and intelligent design. I've researched GAs and information theory to a fair extent, and am happy to answer questions and discuss on that topic.

Again, that's fair enough. However, if we're leaving out biological evolution, I don't think there's much of an argument to be had.


I'd be happy to continue discussion over email. I don't like the HN comments overflowing the side of the screen. My email is eric.holloway at google email setvice.




Guidelines | FAQ | Support | API | Security | Lists | Bookmarklet | Legal | Apply to YC | Contact

Search: