Blog all dog-eared unpages: Philosophy & Simulation: The Emergence of Synthetic Reason by Manuel DeLanda

Philosophy & Simulation: The Emergence of Synthetic Reason by Manuel DeLanda

This was incredibly hard-work as a read for this bear of little brain, but worth it. Very rewarding and definitely in resonance with earlier non-fiction reads this year (The Information, What Technology Wants, The Nature of Technology)

I’ve put the things that really gave me pause in bold below.

an unmanifested tendency and an unexercised capacity are not just possible but define a concrete space of possibilities with a definite structure.

a mathematical model can capture the behavior of a material process because the space of possible solutions overlaps the possibility space associated with the material process.

Gliders and other spaceships provide the clearest example of emergence in cellular automata: while the automata themselves remain fixed in their cells a coherent pattern of states moving across them is clearly a new entity that is easily distinguishable from them.

This is an important capacity of simulations not shared by mathematical equations: the ability to stage a process and track it as it unfolds.

In other words, each run of a simulation is like an experiment conducted in a laboratory except that it uses numbers and formal operators as its raw materials. For these and other reasons computer simulations may be thought as occupying an intermediate position between that of formal theory and laboratory experiment.

Let’s summarize what has been said so far. The problem of the emergence of living creatures in an inorganic world has a well-defined causal structure.

The results of the metadynamic simulations that have actually been performed show that the spontaneous emergence of a proto-metabolism is indeed a likely outcome, one that could have occurred in prebiotic conditions.

Because recursive function languages have the computational capacity of the most sophisticated automata, and because of the random character of the collisions, this artificial chemistry is referred to as a Turing gas.

An evolving population may, for example, be trapped in a local optimum if the path to a singularity with greater fitness passes through points of much lesser fitness.

Roughly, the earliest bacteria appeared on this planet three and a half billion years ago scavenging the products of non-biological chemical processes; a billion years later they evolved the capacity to tap into the solar gradient, producing oxygen as a toxic byproduct; and one billion years after that they evolved the capacity to use oxygen to greatly increase the efficiency of energy and material consumption. By contrast, the great diversity of multicellular organisms that populate the planet today was generated in about six hundred million years.

The distribution of singularities (fitness optima) in this space defines the complexity of the survival problem that has to be solved: a space with a single global optimum surrounded by areas of minimum fitness is a tough problem (a needle in a haystack) while one with many local optima grouped together defines a relatively easy problem.

from the beginning of life the internal models mediating the interaction between a primitive sensory system and a motor apparatus evolved in relation to what was directly relevant or significant to living beings.

with the availability of neurons the capacity to distinguish the relevant from the irrelevant, the ability to foreground only the opportunities and risks pushing everything else into an undifferentiated background, was vastly increased.

Finally, unlike the conventional link between a symbol and what the symbol stands for, distributed representations are connected to the world in a non-arbitrary way because the process through which they emerge is a direct accommodation or adaptation to the demands of an external reality.

This simulation provides a powerful insight into how an objective category can be captured without using any linguistic resources. The secret is the mapping of relations of similarity into relations of proximity in the possibility space of activation patterns of the hidden layer.

Both manual skills and the complex procedures to which they gave rise are certainly older than spoken language suggesting that the hand may have taught the mouth to speak, that is, that ordered series of manual operations may have formed the background against which ordered series of vocalizations first emerged.

When humans first began to shape flows of air with their tongues and palates the acoustic matter they created introduced yet another layer of complexity into the world.

Says(Tradition, Causes(Full Moon, Low Tide)) Says(My Teacher, Causes(Full Moon, Low Tide))

A mechanism to transform habit into convention is an important component of theories of non-biological linguistic evolution at the level of both syntax and semantics.

a concentration of the capacity to command justified by a religious tradition linking elite members to supernatural forces or, in some cases, justified by the successful practical reasoning of specialized bureaucracies.

Needless to say, the pyramid’s internal mechanism did not allow it to actually transmute a king into a god but it nevertheless functioned like a machine for the production of legitimacy.

social simulations as enacted thought experiments can greatly contribute to develop insight into the workings of the most complex emergent wholes on this planet.

abandon the idea of “society as a whole” and replace it with a set of more concrete entities (communities, organizations, cities) that lend themselves to partial modeling in a way that vague totalities do not.

London Games Festival: The Future of AI in games

Imperial

Just been to a talk at Imperial College London, put on as part of the London Games Festival, presenting viewpoints form the games industry (Peter Molyneux and someone from Eidos) and from AI Academia. Very accessible and interesting.

I’ve tried my best to do an Alice, but I’ve not quite got the knack – so far from verbatim notes below:

The future of AI in games
London Games Festival

4.10.06

peter molyneux, prof. mark cavazza., dr. simon colton

intro
john cass, icl

article in the economist from the summer (CF)

next challenge is to develop believable characters and intelligences in game worlds

bring together two communities: the game devlopers from industry and artificial intelligence research community from academia

take industry to a new level

—-

peter molyneux

this is the most interesting area of game design to him

sorry – on behalf of games industry for grabbing the term AI and totally abusing it.

there is very little real AI in games

AI is mistaken for
– navigation
– avoidance
– crude simulations
– scripted behaviour

this is where we are, where do we want to be?

we need a whole raft of REAL AI and we’re starting to get the processing power to do it. next gen consoles could be the key.

– agent AI: need for convincing characters, recognizing what you are doing as a player. we are doing so much more as players – more freedom, more emotion. fable2: friendship, family – relationships… how do this convincingly?

– cloning AI: online is here to stay and this creates big problems… what about having a clone of yourself to remain in a persistent world so you can stay ‘present’ when you should go to sleep (UK vs. australia)

– learning AI – adapting to players and play.

– balancing AI: we’ve failed because we are not mass market – we only appeal to a very small audience… biggest game = 20m should be 200m… one of the reasons we have not got the reach is that we have no way to balance the difficulty of the game – looking at how the player plays and balance the game play accordingly (cf. czymihalyi flow, robin hunicke’s work)

AI future – will change the way that games are designed, create new types of game, create unique experiences… my game experience will be different from yours. far more realistic worlds can be created… visually we are getting close, but need great AI to back this up otherwise they will feel flawed. i will be able to stand up in 5yrs time and say look at how games have changed due to AI.

—-
DR. MARK CAVAZZA, UNIVERSITY OF TEESIDE

AI for interactive storytelling

‘long term endeavor to reconcile linear story and interaction’

reincorporate aesthetic qualities of linear media

character-based storytelling: Hierarchical Task Network Planning (AI technique – look up?) to describe characters roles.

AI maintains consistency of the story, while allowing adaptation… but often driving towards satisfying conclusion (interactive storytelling is not just changing the ending!)

sitcom generator: each characters role is described as a HTN plan. (modelled on ‘Friends’)

dynamic interactions between characters contribute to generating multiple situation not encoded in the original roles.

sitcom chosen to test the theory – as they are essentially/generally simple story forms (not shakespeare!)

we are generating a lot of stories and a lot of them are rubbish… need to filter these… and we can only generate about 6mins…

what’s the diff between this and The Sims? Sims have no narrative drive, they react (narrative is in the eye of the beholder)

every time these characters act.. they have a plan.

silent movies atm, but next step is dialogue.
this is very processing power intensive, but making progress with small scaling demonstrations. (shows one) Scalability is not really there atm.

real challenge is to develop true interactive storytelling capabilities.

The world is an actor: worlds behaviour drives narrative events. blurring the boundaries of physics and AI – the world is ‘plotting against the character’… inspired by the ‘final destination’ movies!

the whole environment ‘has a plan’

its easy to look clever in AI in small exmaples, the real challenge is scability… but we think the principles here are sound.

(doing research project with DTI/Eidos)

Dr. Simon Colton
AI and Games – Do’s and Don’ts

(games industry)unhealthy obsession #1: the modeling of opponents

(AI academia) unhealthy obsession #2: playing board games
From the machine learning journal: ‘learning to bid in bridge’ is a 30 yr project and it’s still going!

multiple mismatches in these two worlds
– what AI in games have low ram, low cycles, low time
– AI agents really want lots of ram, time, cycles

– ‘An AI’ that is referred to in games does not exist as termed by academia… a ‘complete AI’ would have emotional intelligence, reasoning, etc…

we’re developing AI the wrong way round – higher reasoning rather than basic instincts (cf. rodney brooks)

– ‘playing chess is a doddle compared to avoiding a tiger’

– AI researchers think it’s about BEATING the player, whereas games industry want AIs to help engage the player further in the game world.

so, what else can we do

– data mining game-play data
— changing how the game plays
– affective computing (HCI)
— how to tell from a players face what their emotional response is and changing game-play
– automatic avatars (to step in your place for sleep and toilet breaks!)
– but could be most useful in the design stage

comparison to the biotech industry
is designing a game more difficult than designing a drug? maybe? do drug companies have more funds? more IP issues? maybe?
BUT – drug companies absolutely make more use of AI in their design process than the games industry…

picks and shovels (where the money is) – getting the computer to program itself (misused phrase,but.. )
– machine learning
– genetic programming
— combining gives more than the sum of parts

one possible approach

evolutionary approach enables you to generate new entities for games – NPCs, cars, object… program AIs to use middle-ware to create these things

AI makes 100 bad models of a football – choose best 10 then breed… 1000s of generations later get valuable assets…

machine learns your aesthetic as a designer…

AI for game environment design

possible human-computer interaction in the design phase of games

designer creates a few building in his/her style
AI takes over and creates rest of city, designer refines the process…

great at design stage, but possibilities at run-time…

now the hard part: it’s still not easy to use AI/machine learning techniques in the off the shelf manners
– the best techniques come with a human (expert)

majority of AI academics don’t know how games are designed – start of a conversation?

summary: good AI opponents still a way off

AI people should think about engaging rather than conquering opponents

games people should think more about using AI tools in the design phase.

google: “AI bite”