Creation: Life And How to Make It

Sue Wilcox has been writing and speaking for years about ALife (Artifical Life), virtual worlds, and other technologies that define (and question) the fringes of life. The seemingly intractable question of whether there can be artificial life doesn't seem to bother Wilcox one whit: she asserts that there can be, and is. Here she reviews a book on making life, but not as we know it. "The author's challenge to himself," she says, "was to make life within a computer, not just unchanging, low-level life, but intelligent life. In this book he describes how to do it from first principles." (Read on for more.)

Creation: Life and How to Make it
author	Steve Grand
pages	224
publisher	Weidenfeld and Nicolson
rating	8.5
reviewer	Sue Wilcox
ISBN	0-297-64391-6
summary	The ideas behind artificial life, explained for a non-expert audience.

If God wrote a book about the way he put the universe together, why the laws of physics were the way they were, how he came to design humans and all the other life forms on Earth, and why they are interdependent with each other and with the planet, it would be a lot like Steve Grand's Creation: Life and how to make it.

Steve is a self confessed digital god -- and he can prove it: there are over a million lifeforms created by him running around in computers all over the world. They live in their own world of Albia within the computer game Creatures. These are not the run-of-the-mill scripted non-player-characters common in computer games. These little creatures aren't simply programmed to behave: their behavior emerges from the way they are. They are artificial life -- ALife.

This is a lightly written but mind-bendingly deep book. When you realize you have been smooth talked into abandoning the last fifty years of AI research and development along with the majority of current thinking on ALife you know the Grand philosophy has gotten into your blood.

Creation isn't just about the inhabitants of a game; it's about existence, the nature of life, and perhaps more important to humans, the nature of intelligence. What is a conscious mind, and can machines have one too?

This is not a book about exactly how to write the code behind Alife; instead it's about how to think about both simulations and actual living organisms, so that there's some point to writing the code.

Explaining how to think about the world, starting with understanding subatomic particles, then moving onto items of greater complexity -- atoms, then molecules, then autocatalytic networks, self-reproducing systems, adaptive systems, intelligence and mind -- is something Steve is very good at. It must come from all the thinking he does. He says that sitting in a darkened silent room and just thinking is one of his favorite occupations. It's left him with an almost Buddhist sense of detachment from reality as most people conceive of the world.

He's pushing for a paradigm shift in our view of reality, and like others before him who've tried that -- Copernicus, Gallileo, Newton, and Einstein -- he's finding it hard work standing the world on its head. But as with his predecessors once the ground has moved under your feet the new place you're standing seems completely right and obvious. It's a new way of seeing that is vital to continued progress.

If there has to be a God, I wouldn't mind letting Steve have a go at the job -- as long as he isn't answerable to another marketing department controlling what his creatures look like. Those cutesy Norns -- ugghh!

Re:this is completely the wrong approach!

I have read the book and this is NOT the approach he takes. In fact he strongly argues against this exact approach for the reasons you've given (and others).

As I remember it, the main reason he got into subatomic particles and such was in the area of the book where he was talking about life itself and how it works. He never once suggested that we should emulate that in order to create artificial life. In fact, one of the key arguments was that artificial life shouldn't be the same as physical life since the enviroments are different. Life tends to form and grow according to the enviroment it lives in and it wouldn't make sense to have normal biological life in a computer simluation.

Instead, the approach he takes is that he starts with some very basic principals and builds from there adding things like desire, attention, needs, wants, etc. to the creature. The end result is that it looks and acts a whole lot like real life. And with the addition of a digial DNA concept, it even reproduces like real life in that genes are inherited and even a little bit of teaching between parents and children take place (from what I can gather this is just one of the millions of behaviors that have been observed that were not directly programmed!). It even goes so far that the genes describe the color of the "fur" on the creatures and so sometimes common traits will pop up where you can recognize familes by physical attributes. It is quite amazing, really (I have played the game as well).

At any rate, my point is that he does *not* suggest starting at the bottom and building a complete life/universe simulator up. He just started there so as to better understand the process of life itself and its various drives and motivations. Then he went about similating those drives and NOT the actual details that lead us bilogical beings up to them (because, after all, an artificial life form does not live in our world!).

It really is an amazing book and IMHO the review does very little justice to it. You just have to read it!

Re:Slashdot Alife

Here's an idea - go through the keywords of the given story and look them up on everything2 [everything2.com]! Find relevant nodes and perhaps follow those to other nodes until you have enough text to put a comment together. Then you need some kind of grammar algorithm so that it sounds reasonably coherent.

How about a post generated by a neural network? You could train it by feeding it stories and resulting replies for each story - and then the scores for each of those replies. The idea would be to train it for +5 posts. Of course, every time it posted something itself, it would recieve feedback on what score its own post recieved.

--

Re:A Matter Of Definitions

Aside from the fact that your thermostat example is stolen from Chalmers, the real problem is that you both seem to be claiming to have said something deep, but then admitting it all hinges on the definition.

If we define life in such a way as to include thermostats then:

a) that's probably doesn't capture the essence of what we want to use the word for, and

b) it's then not very deep to say that thermostats are alive!

OTOH, if we define life in such a way as to exclude thermostats, then you are wrong.

I'm going for the latter.

Historical book?

I haven't read the book mentioned in the article, but I have read (and played with the programs contained in it) a much earlier book, published in 1981 by TAB Books, called "Robot Intelligence (with experiments)" by David L. Heiserman (TAB Books, 1981, ISBN 0-8306-9685-7).

In this book, the author explores ideas and meanings behind a type of life he calls "Evolutionay Adaptive Machine Intelligence" or EAMI for short. He explores this through a number of BASIC code programs written in stages, from simple "Alpha-class" systems, to much more complex "Gamma-class" systems.

What makes this book all the more interesting is that in theory (and I believe this is explored somewhat in the book) you can apply all of this back to real-world machines: This book is simply the culmination of two earlier robotics project books by the same author: "Build Your Own Working Robot" (TAB Books, 1976, ISBN 0-8306-6841-1) and "How to Build Your Own Self-Programming Robot" (TAB Books, 1979, ISBN 0-8306-9760-8). This last book actually started to explore the concepts outlined in "Robot Intelligence", but stopped just short of it. The point is, this series of books showed the hobbiest of a couple of decades ago (thereabouts) how to build real ALife, long before it was very popular (not to mention cheap).

I encourage anyone with interest in this subject to pick these three books up. As far as I know, they are long out of print, so happy hunting.

In a side note - does anyone know what happened to Mr. Heiserman and his robots?

Worldcom [worldcom.com] - Generation Duh!

Norn torture

Obviously some people think Norns are alive judging by the amount of hatemail AntiNorn got for his Norn Torture [geocities.com] page. Personally I think it's hilarious.

this is completely the wrong approach!

Explaining how to think about the world, starting with understanding subatomic particles, then moving onto items of greater complexity -- atoms, then molecules, then autocatalytic networks, self-reproducing systems, adaptive systems, intelligence and mind -- is something Steve is very good at. It must come from all the thinking he does.

This is just my opinion, but I think this is completely the wrong way to go about it. I think an analogy will help explain what I mean. Suppose I show you a picture, then ask you to write a program which generates pictures. It would be stupid for you to start analysing the subatomic particles so that you could get a better idea of the pigments, so you could understand the use of colour...

A much better approach in the case of the picture (and I would assert in the case of life) would be to give up the reductive approach. Instead, one should focus on the function the picture serves and try to replicate that. I.e. try to do the same job rather than trying to do the job in the same way.

I think this whole obsession with reductive solutions stems from the success of physics. In physics looking at subatomic particles has been fruitful, but this is because of the particular nature of the problems physics tries to solve. Problems involving life and consciousness, for example, are very different: we should rethink our methods rather than blindly trying to apply those that worked in the past.

Re:AI life is overhyped..

Emotions are just things that animals/humans have evolved to enhance survival of genes.

Jealousy? That's someone have sex with someone you want.

Anger? That's someone who has just taken your food.

Love? Someone you want to perpetuate your genes with.

The point is that AI can have just the same emotions and for the same reasons. The digital genes (and all genes are digital in fact) behave very much the same.

There's one difference between artificial and real

AI creatures don't climb artificial mountains.

Of course not;

Because it isn't there.

Life is not a formal system

Ah, yet another book which uses cute language and twisting analogies to make us believe we're seeing something that isn't there. If I were to tell you that I created a potato in a computer, and then invited you to eat it, you'd laugh at me. When dealing with solid examples of real things, computer simulations of them seem laughably easy to separate from the real mccoy, however, when you come to abstract concepts for which nobody has a clear definition ( Life, Intelligence, etc. ) its actually much easier to fool people ( including yourself ). It becomes especially easy when you take the time to spin yarns about the structure of the universe, from subatomics upwards, in order that it all fit your end goal.

Formula: Uncertainty + sophisticated language = plausible story.

Now don't get me wrong here, I'm sure the book is a great and valuable work in the field of ALife, which is a very intriguing field in computing today. But to make claims that one has actually created life in a computer that is equivalent to even the simplest forms we see around us in the natural world is just plain ridiculous.

I'm not going to waste time debating the particulars of the systems involved here and whether they do or don't meet the criteria for living things as abstracted by whomever, or whether those criteria are a sufficient definition for life. I just want to point out something that all too often gets forgotten when dealing with informational systems - the gulf between a simulation and the simulated.

Firstly it is a necessary element of all simulations that they are a reduced set of the properties of the system being simulated. You cannot pefectly simulate a real system - the only perfect simulation is the original system itself. But this is a minor point. Attached to this, and more importantly is the fact that informational simulations HAVE NO CAUSAL POWERS! Having no physical existance, they cannot cause or affect anything except what is formally defined for them in the simulation. If I say that I have simulated a tornado in my computer, nobody is going to worry that it might destroy my city. If you put a lovely roaring log fire screensaver on your computer it will never keep you awake at night worrying if the fire will melt your monitor. A computer will never be crushed by the weight of the eifel tower if it has a 3d model of it. All of these are informational simulations of NON-FORMAL systems and therefore do not entail the be-all and end-all of the systems they simulate, and therefore ( finally... ) are not equivalent to the real thing. Only formal systems ( ie like a game of chess ) can be said to exist entirely in any medium in which they are rendered. Non-formal systems can only exist as a simulation when rendered in other media. Life is not a formal system.

Why is it that stories like these are so easy to fool us? Well probably because humans are informational creatures - we are the makers and consumers of our own information. Therefore we sometimes find it difficult to see the line between simulations and the simulated - because both get represented to our minds by roughly equivalent information. Add to that the above mentioned confusion and elaborate talk and its a marvelous act of intellectual slight-of-hand that makes us see life where it doesn't exist.

Nice try guys, keep it up - its a fun show =)

Work of his more interesting than Creatures.

I ran into a few people from his company at the Alife VI Conference in LA in 1998. The Creatures game was part of what they were discussing, but not really. They were really excited about a new contract they had with DERA. (British Defence Evaluation Research Agency, public/private defence contractor org, and home of the Harrier jet)

They had contracted to build more adaptive and intelligent combat flight enemies for the simulations. The pilots were able to predict how the existing rule based systems worked, and were becoming rigid in their own reactions. So they contacted these guys, and they built a system (way less complex than the characters in Creatures) pretty quickly.

The first version that they came out with was incredibly effective, but you'd be unlikely to come across this strategy in a human pilot: barrel roll incessantly, pull up if the enemy is above you, and fire when they're in your sights. Very simple rules, works no matter what the position of the enemy, and would pulp a human pilot. After some tweaking, they ended up with something that more resembled human behavior.

But the first round got them thinking. In a dogfight, maneuverability is key. A plane can handle maybe 15 Gs, a human pilot 8-10 tops. If a fighter plane weren't dependent on the limitations on the human pilot, it would win against a plane having such limitations, *every time*. or nearly. be able to pull sharper turns, more extreme maneuvers, etc.

Based on this, and the way the flight sim was coded (the neural net flying the plane got its inputs from the data that would be available from the actual instruments), they were proceeding with a proposal to put this puppy in a live plane. Haven't heard anything more about it, but I still get the willies when I see the Creatures box in stores.

--Shameless SelfPlug Check out the papers I published on social environments and language origination using multiagent sims.
http://www.cs.rochester.edu/u/www/u/stoness/ [rochester.edu]
(about halfway down the page

Life as a Game.

I suppose that everyone is going to post up their favorite theory of what life is.

The one that gets my vote is the idea that life is a game because the definition of a game supplies the metarules. Things that you need in things like things you can do, things you cannot do, loopholes to exploit, limitations due to the nature of the game itself, multiple levels of games, games within games, games you do not know about, being someone's pawn, etc.

Then you get into the philosophy of game design. It was an illuminating thought that most people would not like to live in real life that world that is their favorite gaming world. And looking at the games people play, and sometimes trap themselves inside.

The philosophical payoff is knowing what is the price you have to pay to get out of the game or change it if you want, and knowing better the games you are really playing in the first place.

So many problems, so little time

1) This is not a book review. We heard nothing about the contents of the book except that they were "mind-bending". So bend our minds a little with some excerpts or paraphrases or something.

2) You twice compare the author to God (including one comparison that compares the book to something God would write)...yet you only give it an 8.5. Surely it would be newsworthy to explain how "God went wrong" and lost 1.5 points.
--

A Matter Of Definitions

Well, I cant say I've read the book, but I can tell you this much -- I once wrote a paper for a Philosophy class I was in that argued that ordinary household thermostats can technically be considered alive, if you agree that the fundemental definition of life is an object that both consumes and produces energy, responds to its environment..The ability to reproduce isn't necessarrilly required -- Life itself could be a dead-end.

If you make the definition intentionally vague, you can fit pretty much any dynamic system under the flag of being "alive"...So be careful when someone tells you that they have a formula for it. Chances are it hangs heavilly on the definition of "alive" to make it work.


Bowie J. Poag

Slashdot Alife

We've all heard of the Turing test (and if you haven't, you're reading the wrong website). I wonder how feasible it would be to pass a "Slashdot test". Write a robot that posts replies to articles on Slashdot. The winner would be the first unaided bot to hit the karma cap.

Anyone know if this has been attempted before? (I'm aware of the First Post scripts; they obviously aren't going to get karma.) If I had some free time (oh I wish) this would definitely be a cool project to undertake.

I wonder if a simple script that ripped unusual words from the article, Google searched for sites with those words, then posted an 'informative' link would work...
--

Desparation is the English Way...

Life does not depend on unpredictability. If something's really alive, then it would still be alive whether or not you could completely predict its behavior. The ability to predict the behavior of an organism does not by definition kill it. Life doesn't necessarily depend on randomness, either.

You can totally predict the evolution of Conway's game of Life, and other deterministic cellular automata, given the initial configuration. It's not necessary to solve the halting problem in order to predict the state in the future -- you just execute the completely deterministic rules. Simple. Conway's Life is awkwardly Turing complete, but it's inefficient for the purpose of general computation (much less efficient that a Turing machine). But at least it means that theoretically you could implement a higher level of Conway's Life (or any other computable function) in terms of a lower level Conways' Life implementation, but it would take a whole lot of time and space.

Andy Weunsche at the Santa Fe Institute has come up with a beautiful way to plot out the deterministic state map of any cellular automata rule: it's a colorful branching graphical fish-eye tree representation of the topology of every possible state and transition of a cellular automata rule (the basin of attraction fields).

You can see for yourself how a given cellular automata rule is completely deterministic, by viewing all the possible interconnected states at once. "Garden of Eden" states (that there was no possible way to arrive at through the rule, so they must be original conditions) are drawn at the extreme tips of the branches, that converge into cycles of the basins of attraction (repeating dead-ends where there is now way to break out). This is really wonderful stuff, well worth scrolling through the whole gallery:

http://www.santafe.edu/~wuensch/gallery/ddlab_gall ery.html [santafe.edu]

On the other hand, the halting problem has to do with one program's ability to predict if another program will halt (not to just simply simulate the program's execution at a higher level: because if the other program doesn't halt, the simulator will never halt either, therefore failing to give the result). It means that there are undecidable questions that a deterministic Turing complete program can't answer: even if the answers are out there somewhere, they just can't be reached by a Turing machine. It also depends on being able to represent any program as data (a number), that can be given to another program as input, which is essential to the Universal Turing Machine in "On Computable Numbers".

The paradox can be demonstrated by asking such a hypothetical program (called "HaltingProblem") to predict whether another subtly (yet insideously) modified version of itself, called "HaltingProblemNot", will halt.

Given a program "HaltingProblem" that attempts to predict if another program halts (taking as input data that program and its inputs), you can always construct another program "HaltingProblemNot" to give it as input, for which it will never be able to give you a correct answer.

"HaltingProblemNot" just has to call the first program "HaltingProblem" as a subroutine, and then it inverts the return value (not just logically, but by halting if it says it won't halt, and infinitely looping if it says it doesn't halt). An obnoxious trick (called diagonalization), but it's proven to work every time. The fatal Achilles' heel of logic -- Godel strikes again.

No matter how cleverly written, the original program "HaltingProblem" is doomed to fail given "HaltingProblemNot" and another program as input, by either looping infinitely or returning the incorrect result.

This does not mean the mind is any more powerful than a Turing machine, nor unpredictable. Nobody really knows for sure. The only thing we know for sure is that there are many things we'll never know.

Gilda Radner summed it up: