Universal laws and architectures. John Doyle John G Braun Professor Control and Dynamical Systems, EE, BE Caltech

Transcription

1 Universal laws and architectures John Doyle John G Braun Professor Control and Dynamical Systems, EE, BE Caltech

2 Turing on layering The 'skin of an onion' analogy is also helpful. In considering the functions of the mind or the brain we find certain operations which we can explain in purely mechanical terms. This we say does not correspond to the real mind: it is a sort of skin which we must strip off if we are to find the real mind. But then in what remains we find a further skin to be stripped off, and so on. Proceeding in this way do we ever come to the 'real' mind, or do we eventually come to the skin which has nothing in it? In the latter case the whole mind is mechanical. 1950, Computing Machinery and Intelligence, Mind

3 Universal laws and architectures? Universal conservation laws (constraints) Universal architectures (constraints that deconstrain) Mention recent papers* Focus on broader context not in papers Lots of case studies for motivation *try to get you to read them? A rant

4 This paper aims to bridge progress in neuroscience involving sophisticated quantitative analysis of behavior, including the use of robust control, with other relevant conceptual and theoretical frameworks from systems engineering, systems biology, and mathematics. Very accessible No math Doyle, Csete, Proc Nat Acad Sci USA, JULY

5 Lots from cell biology my case glycolytic oscillations for hard limits studies bacterial layering for architecture Networking and clean slate architectures wireless end systems info or content centric application layer integrate routing, control, scheduling, coding, caching control of cyber-physical PC, OS, VLSI, antennas, etc (IT components) Neuroscience brains neuroendocrine control Medical and exercise physiology

6 Cell biology Networking & clean slate architectures Neuroscience Medical physiology Smartgrid, cyber-phys Wildfire ecology Earthquakes Lots of aerospace Physics: turbulence, stat mech (QM?) Toy : Lego, clothing, buildings, Synesthesia my case studies

7 Existing design frameworks Sophisticated components Poor integration Limited theoretical framework Fix?

8 Happy families are all alike; every unhappy family is unhappy in its own way. Leo Tolstoy, Anna Karenina, Chapter 1, first line What could this mean? Given incredible diversity of people and environments? It has to be a statement about organization, and specifically architecture. Happy = empathy + cooperation + simple rules? Constraints on components and architecture

9 accessible accountable accurate adaptable administrable affordable auditable autonomy available credible process capable compatible composable configurable correctness customizable debugable degradable determinable demonstrable dependable deployable discoverable distributable durable effective efficient evolvable extensible failure transparent fault-tolerant fidelity flexible inspectable installable Integrity interchangeable interoperable learnable maintainable manageable mobile modifiable modular nomadic operable orthogonality portable precision predictable producible provable recoverable relevant reliable repeatable reproducible resilient responsive reusable robust safety scalable seamless self-sustainable serviceable supportable securable simplicity stable standards compliant survivable sustainable tailorable testable timely traceable ubiquitous understandable upgradable usable Requirements on systems and architectures

10 accessible accountable accurate adaptable administrable affordable auditable autonomy available credible process capable compatible composable configurable correctness customizable debugable degradable determinable demonstrable dependable deployable discoverable distributable durable effective efficient evolvable extensible failure transparent fault-tolerant fidelity flexible inspectable installable Integrity interchangeable interoperable learnable maintainable manageable mobile modifiable modular nomadic operable orthogonality portable precision predictable producible provable recoverable relevant reliable repeatable reproducible resilient responsive reusable robust safety scalable seamless self-sustainable serviceable supportable securable simple stable standards compliant survivable sustainable tailorable testable timely traceable ubiquitous understandable upgradable usable Simplified, minimal requirements

11 accessible accountable accurate adaptable administrable affordable auditable autonomy available credible process capable compatible composable configurable correctness customizable debugable degradable determinable demonstrable dependable deployable discoverable distributable durable effective evolvable extensible failure transparent fault-tolerant fidelity flexible inspectable installable Integrity interchangeable interoperable learnable maintainable manageable mobile modifiable modular nomadic operable orthogonality portable precision predictable producible provable recoverable relevant reliable repeatable reproducible resilient responsive reusable safety scalable seamless self-sustainable serviceable supportable securable stable standards compliant survivable tailorable testable timely traceable ubiquitous understandable upgradable usable Requirements on systems and architectures efficient robust simple sustainable

12 Requirements on systems and architectures efficient simple sustainable resilient robust

13 Requirements on systems and architectures efficient simple sustainable robust

14 Requirements on systems and architectures fragile sustainable resilient robust simple efficient wasteful complex

15 Requirements on systems and architectures sustainable fragile complex robust simple efficient wasteful

16 Want to understand the space of systems/architectures fragile Case studies? Strategies? robust Want robust and efficient systems and architectures efficient wasteful Architectures?

17 Resilient architectures are all alike; every brittle system is brittle in its own way. Apologies to Tolstoy Resilience includes robustness, efficiency, sustainability, scalability, etc etc Effective architectures provide flexible tradeoffs across all these dimensions Subject to laws which are hard constraints on what is achievable Defer resolving terminology, focus on Theorems and concrete case studies

18 Resilient architectures are all alike; every brittle system is brittle in its own way. Good architecture = constrains that deconstrain (Gerhart and Kirschner)

19 The dangers of naïve biomemetics Feathers and flapping? Or lift, drag, propulsion, and control?

20 Getting it (W)right, 1901 We know how to construct airplanes... (lift and drag) also know how to build engines. (propulsion) When... balance and steer[ing]... has been worked out, the age of flying will have arrived, for all other difficulties are of minor importance. (control) Wilbur Wright on Control, 1901 (First powered flight, 1903)

21 Universals? Feathers and flapping? Lift, drag, propulsion, and control?

22 Universals? Complexity control, robust/fragile tradeoffs Fragility Hijacking, side effects, unintended Of mechanisms evolved for robustness Math: robust/fragile constraints ( conservation laws ) Both Accident or necessity?

23 Very accessible No math I m interested in fire

24 Accessible ecology UG math

25 Wildfire ecosystem as ideal example Cycles on years to decades timescale Regime shifts: grass vs shrub vs tree Fire= keystone specie Metabolism: consumes vegetation Doesn t (co-)evolve Simplifies co-evolution spirals and metabolisms 4 ecosystems globally with convergent evo So Cal, Australia, S Africa, E Mediterranean Similar vegetation mix Invasive species

26 Current Technology? fragile At best we get one robust efficient wasteful

27 ??? fragile Often neither robust efficient wasteful

28 fragile??? Bad architectures?? gap? robust Bad theory?? efficient wasteful

29 fragile Universal law? robust efficient wasteful

30 Exponential improvement in efficiency F 100% 10% 1% log F F Efficiency.1%

31 When will lamps be 200% efficient? Exponential improvement 100% 10% 1% Solving all energy problems? log F F Efficiency.1% Note: this is real data!

32 When will lamps be 200% efficient? Oops never. 50% 10% 1% F F 1 F Efficiency.1% Note: need to plot it right.

33 Doyle s law? Exponential improvement 100% 10% log F 1% F Efficiency.1%

34 Universal law 100% 10% 1% F Efficiency.1%

35 Universal law? efficient wasteful 100%

36 fragile robust Some features robust to some perturbations Other features or other perturbations

37 laws and architectures? fragile Sharpen hard bounds Case studies robust efficient wasteful

38 Control, OR Bode Pontryagin Kalman Shannon Comms Nash Von Neumann Theory? Deep, but fragmented, incoherent, incomplete Carnot Turing Compute Godel Heisenberg Einstein Boltzmann Physics

39 fragile??? Bad architectures?? gap? robust Bad theory?? efficient wasteful

40 fragile Find and fix bugs bad Bad architectures? Sharpen hard bounds Case studies wasteful

41 Compute Turing ( ) Turing 100 th birthday in 2012 Turing machine (math, CS) test (AI, neuroscience) pattern (biology) Arguably greatest* all time math/engineering combination WW2 hero invented software *Also world-class runner.

42 Key papers/results Theory (1936): Turing machine (TM), computability, (un)decidability, universal machine (UTM) Practical design (early 1940s): code-breaking, including the design of code-breaking machines Practical design (late 1940s): general purpose digital computers and software, layered architecture Theory (1950): Turing test for machine intelligence Theory (1952): Reaction diffusion model of morphogenesis, plus practical use of digital computers to simulate biochemical reactions

43 Fast and flexible Slow Solve problems Make decisions Take actions Fast Flexible Inflexible

44 Laws and architectures Slow Architecture (constraints that deconstrain) Fast Flexible Inflexible

45 Compute Turing Godel Shannon Comms slow?? fragile? inflexible? Each theory one dimension Tradeoffs across dimensions Assume architectures a priori Progress is encouraging, but Stovepipes are an obstacle Carnot Control Bode Heisenberg Einstein Boltzmann Physics

46 Compute Turing Communicate Shannon Delay is most important Delay is least important Bode Control, OR Heisenberg Einstein Carnot Boltzmann Physics

47 Compute Turing Delay is most important Bode Control, OR

48 Compute Turing Delay is most important Bode Control, OR

49 Turing as new starting point? Software Hardware Compute Turing Delay is most important Digital Analog Bode Control, OR

50 Turing as new starting point? Essentials: 0. Model 1. Universal laws 2. Universal architecture 3. Practical implementation Software Hardware Digital Analog Turing s 3 step research: 0. Virtual (TM) machines 1. hard limits, (un)decidability using standard model (TM) 2. Universal architecture achieving hard limits (UTM) 3. Practical implementation in digital electronics (biology?)

51 Who/what

52 Slow Flexible Turing architecture Slow Software Hardware Fast Hard limits? Flexible Digital Analog Fast Inflexible Inflexible

53 Flexible General purpose Large uncertainties Diverse problems Fast Solve problems Make decisions Take actions Flexible Low latency/delay Fast

54 Slow Flexible Slow Hopelessly Fragile Fast Potentially Robust Flexible Fast Inflexible Inflexible

55 fragile robust Some features robust to some perturbations Other features or other perturbations

56 Increased complexity? fragile robust Some features robust to some perturbations Other features or other perturbations

57 Robust Modular Simple Plastic Evolvable and xor Fragile Distributed Complex Frozen Frozen tradeoffs

58 Slow Flexible Turing architecture Slow Software Hardware Digital Analog Fast Flexible Fast Inflexible Inflexible

59 Slow Flexible Slow Horizontal App Transfer Software Hardware Digital Horizontal HW Transfer Analog Fast Flexible Inflexible

60 Slow execution Flexible reprogramming Software Faster execution Less flexible Software Hardware Modern technology gives lots of intermediate alternatives. Digital Analog

61 Applications Control, share, virtualize, and manage resources Operating System Processing Memory I/O Want to emphasize the differences between these two types of layering. Software Hardware Digital Analog

62 Horizontal App Transfer What matters is the OS. Applications Operating System Software Hardware Horizontal HW Transfer Digital Analog

63 Horizontal Applications App Transfer Some people write apps and build hardware But most software and hardware is acquired by horizontal transfer from others Similarly, most new ideas (humans) and new genes (bacteria) are acquired horizontally Operating System Software Hardware Horizontal HW Transfer Digital Analog

64 Compute Turing Why Necessity

65 Essentials: 0. Model 1. Universal laws 2. Universal architecture 3. Practical implementation TM Hardware Digital Turing s 3 step research: 0. Virtual (TM) machines 1. hard limits, (un)decidability using standard model (TM) 2. Universal architecture achieving hard limits (UTM) 3. Practical implementation in digital electronics (biology?)

66 Hardware Digital Analog being digital should be of greater interest than that of being electronic. That it is electronic is certainly important because these machines owe their high speed to this But this is virtually all that there is to be said on that subject. That the machine is digital however has more subtle significance. One can therefore work to any desired degree of accuracy Lecture to LMS

67 Hardware Digital Analog digital of greater interest than that of being electronic any desired degree of accuracy This accuracy is not obtained by more careful machining of parts, control of temperature variations, and such means, but by a slight increase in the amount of equipment in the machine Lecture to LMS

68 Summarizing Turing: Digital more important than electronic Robustness: accuracy and repeatability. Achieved more by internal hidden complexity than precise components or environments. TM Hardware Digital Analog Turing Machine (TM) Digital Symbolic Logical Repeatable

69 avalanche The butterfly effect quite small errors in the initial conditions can have an overwhelming effect at a later time. The displacement of a single electron by a billionth of a centimetre at one moment might make the difference between a man being killed by an avalanche a year later, or escaping. 1950, Computing Machinery and Intelligence, Mind

70 quite small errors in the initial conditions can have an overwhelming effect at a later time. It is an essential property of the mechanical systems which we have called 'discrete state machines' that this phenomenon does not occur. Even when we consider the actual physical machines instead of the idealised machines, reasonably accurate knowledge of the state at one moment yields reasonably accurate knowledge any number of steps later. 1950, Computing Machinery and Intelligence, Mind

71 memory Logic TM Hardware Turing s 3 step research: 0. Virtual (TM) machines 1. hard limits, (un)decidability using standard model (TM) 2. Universal architecture achieving hard limits (UTM) 3. Practical implementation in digital electronics (biology?)

72 memory Logic slow time fast TM has memory large space

73 memory Logic time slow fast space is free TM has memory large space

74 memory Logic time? Decidable problem = algorithm that solves it Most naively posed problems are undecidable.

75 data program (TM) UTM Turing s 3 step research: 0. Virtual (TM) machines 1. hard limits, (un)decidability using standard model (TM) 2. Universal architecture achieving hard limits (UTM) 3. Practical implementation in digital electronics (biology?)

76 Software Hardware data program (TM) UTM 2. Universal architecture achieving hard limits (UTM) Software: A Turing machine (TM) can be data for another Turing machine A Universal Turing Machine can run any TM A UTM is a virtual machine. There are lots of UTMs, differ only (but greatly) in speed and programmability (space assumed free)

77 TM program HALT UTM The halting problem Given a TM (i.e. a computer program) Does it halt (or run forever)? Or do more or less anything in particular. Undecidable! There does not exist a special TM that can tell if any other TM halts. i.e. the program HALT does not exist.

78 Thm: TM H=HALT does not exist. That is, there does not exist a program like this: H( TM, input) 1 if TM input halts 0 otherwise Proof is by contradiction. Sorry, don t know any alternative. And Turing is a god.

79 Thm: No such H exists. H( TM, input) 1 if TM input halts 0 otherwise Proof: Suppose it does. Then define 2 more programs: 1 if H ( TM, input) 0 H '( TM, input) loop forever otherwise H *( TM ) H '( TM, TM ) Run H *( H*) H '( H*, H*) halt if H*( H*) loops forever loop forever otherwise Contradiction!

80 data TM UTM Implications Large, thin, nonconvex everywhere TMs and UTMs are perfectly repeatable But perfectly unpredictable Undecidable: Will a TM halt? Is a TM a UTM? Does a TM do X (for almost any X)? Easy to make UTMs, but hard to recognize them. Is anything decidable? Yes, questions NOT about TMs.

81 Really slow Computational complexity Slow Fast Undecidable Decidable NP P Flexible/ General Inflexible/ Specific

82 PSPACE NP P NL PSPACE NL Computational complexity Space is powerful and/or cheap. Decidable PSPACE NP P NL Flexible/ General Inflexible/ Specific

83 These are hard limits on the intrinsic computational complexity of problems. Really slow Slow Must still seek algorithms that achieve the limits, and architectures that support this process. Fast Undecidable Decidable NP P Flexible/ General Inflexible/ Specific

84 Compute Delay is even more important in control Computational complexity of Designing control algorithms Implementing control algorithms Software Hardware Digital Analog Control Control Sense Plant Act

85 Slow Flexible Most UTMs here Slow Fast Unachievable robustness Flexible Fast Inflexible Inflexible

86 Slow Flexible Most UTMs here Slow Fast Impossible Flexible Fast Inflexible Inflexible

87 Issues for engineering Turing remarkably relevant for 76 years UTMs are implementable Differ only (but greatly) in speed and programmability Time/speed/delay is most critical resource Space (memory) almost free for most purposes Read/write random access memory hierarchies Further gradations of decidable (P/NP/coNP) Most crucial: UTMs differ vastly in speed, usability, and programmability You can fix bugs but it is hard to automate finding/avoiding them

88 Issues for engineering Turing remarkably relevant for 76 years UTMs are implementable Differ only (but greatly) in speed and programmability Time/speed/delay is most critical resource Space (memory) almost free for most purposes Read/write random access memory hierarchies Further gradations of decidable (P/NP/coNP) Most crucial: UTMs differ vastly in speed, usability, and programmability You can fix bugs but it is hard to automate finding/avoiding them

89 Conjectures, biology Memory potential Examples Insects Scrub jays Autistic Savants Gallistel and King But why so rare and/or accidental? Large memory, computation of limited value? Selection favors fast robust action? Brains are distributed (not studied by Gallistel)

90 Compute Turing Communicate Shannon Delay is most important Delay is least important Bode Control, OR Heisenberg Einstein Carnot Boltzmann Physics

91 data TM UTM time Suppose we only care about space? And time is free Bad news: compression undecidable. Shannon: change the problem! space

92 Shannon s brilliant insight Forget time Forget files, use infinite random ensembles Good news Laws and architecture! Info theory most popular and accessible topic in systems engineering Fantastic for some engineering problems Communications Shannon

93 Shannon s brilliant insight Forget time Forget files, use infinite random ensembles Bad news Laws and architecture very brittle Communications Shannon Less than zero impact on internet architecture Almost useless for biology (But see Lestas et al, 2010) Misled, distracted generations of biologists (and neuroscientists)

94 Compute Turing Lowering the barrier Communicate Shannon Delay is most important New progress! Delay is least important Carnot Bode Control, OR Heisenberg Einstein Boltzmann Physics

95 System Emergent : Nontrivial consequences of other constraints Architecture =Constraints data TM UTM Protocols Components

96 Systems requirements: Algorithm to solve problem (Un)decidable Constraints data TM UTM UTM Components and materials: Automata + memory

97 Universal architectures What can go wrong?

98 Applications Control, share, virtualize, and manage resources Operating System Processing Memory I/O Want to emphasize the differences between these two types of layering. Software Hardware Digital Analog

99 Networking Software Hardware Virtual Software Hardware Digital Analog Physical Digital Analog

100 Diverse applications (HMT) TCP IP Diverse Physical

101 App caltech.edu? IPC IP addresses interfaces (not nodes) DNS App Global and direct access to physical address!

102 App Robust? Secure Scalable Verifiable Evolvable Maintainable Designable IPC IP addresses interfaces (not nodes) DNS App Global and direct access to physical address!

103 Naming and addressing need to have scope and resolved within layer translated between layers not exposed outside of layer Related issues VPNs NATS Firewalls Multihoming Mobility Routing table size Overlays Application TCP IP Physical

104 Until late 1980s, no congestion control, which led to congestion collapse TCP IP Diverse Physical

105 Original design challenge? Deconstrained (Applications) Networked OS Constrained TCP/ IP Deconstrained (Hardware) Facilitated wild evolution Created whole new ecosystem completely opposite Expensive mainframes Trusted end systems Homogeneous Sender centric Unreliable comms

106 Outside Inside Transmitter Receiver Ligands & Receptors Responses Reactions control control Protein Assembly control Receiver Coming later: contrast with cells DNA/RNA Responses DNA Cross-layer control Highly organized Naming and addressing

107 Next layered architectures Deconstrained (Applications) Few global variables Constrained? Don t cross layers Control, share, virtualize, and manage resources Deconstrained (Hardware) Comms Memory, storage Latency Processing Cyber-physical

108 Persistent errors and confusion ( network science ) Architecture is least graph topology. Every layer has different diverse graphs. Application TCP IP Architecture facilitates arbitrary graphs. Physical

109

110 Notices of the AMS, 2009

111 Who and what

112 Sequence ~100 E Coli (not chosen randomly) ~ 4K genes per cell ~20K different genes in total ~ 1K universally shared genes ~ 300 essential (minimal) genes DNAp Gene Repl. DNA RNAp Ribosome xrna transc. RN A transl. AA Proteins Horizontal Gene Transfer ATP ATP Catabolism Precursors See slides on bacterial biosphere

113 Neuro motivation Reflex Fast Inflexible

114 Slow Flexible Motor Prefrontal Sensory Learning Ashby & Crossley Striatum Acquire Translate/ integrate Automate Thanks to Bassett & Grafton

115 Slow Flexible Motor Prefrontal Sensory Ashby & Crossley Acquire Translate/ integrate Automate Striatum Fast Inflexible

116 Slow Flexible Striatum Fast Inflexible

117 Slow Flexible Build on Turing to show what is necessary to make this work. Prefrontal Motor Sensory Acquire Translate/ integrate Automate Striatum Reflex Fast Inflexible

118 Prefrontal Motor Sensory Striatum Reflex

119 Slow Flexible Turing architecture Slow Software Hardware Fast Hard limits? Flexible Digital Analog Fast Inflexible Inflexible

120 Flexible General purpose Large uncertainties Diverse problems Fast Solve problems Make decisions Take actions Flexible Low latency/delay Fast

121 Slow Flexible Slow Hopelessly Fragile Fast Potentially Robust Flexible Fast Inflexible Inflexible

122 Robust Metabolism Regeneration & repair Healing wound /infect Human complexity Fragile Obesity, diabetes Cancer AutoImmune/Inflame Start with physiology Lots of triage

123 Robust Metabolism Regeneration & repair Healing wound /infect Benefits Efficient Mobility Survive uncertain food supply Recover from moderate trauma and infection

124 Robust Metabolism Regeneration & repair Healing wound /infect Fat accumulation Insulin resistance Proliferation Inflammation Mechanism? Fragile Obesity, diabetes Cancer AutoImmune/Inflame Fat accumulation Insulin resistance Proliferation Inflammation

125 What s the difference? Robust Metabolism Regeneration & repair Healing wound /infect Fragile Obesity, diabetes Cancer AutoImmune/Inflame Controlled Dynamic Fat accumulation Insulin resistance Proliferation Inflammation Uncontrolled Chronic

126 Controlled Dynamic Low mean High variability Fat accumulation Insulin resistance Proliferation Inflammation

127 Death Controlled Dynamic Fat accumulation Insulin resistance Proliferation Inflammation Uncontrolled Chronic Low mean High variability High mean Low variability

128 Restoring robustness? Robust Metabolism Regeneration & repair Healing wound /infect Fat accumulation Insulin resistance Proliferation Inflammation Controlled Dynamic Low mean High variability Fragile Obesity, diabetes Cancer AutoImmune/Inflame Fat accumulation Insulin resistance Proliferation Inflammation Uncontrolled Chronic High mean Low variability

129 Robust Human complexity Metabolism Regeneration & repair Immune/inflammation Microbe symbionts Neuro-endocrine Complex societies Advanced technologies Risk management Yet Fragile Obesity, diabetes Cancer AutoImmune/Inflame Parasites, infection Addiction, psychosis, Epidemics, war, Disasters, global &!%$# Obfuscate, amplify, Accident or necessity?

130 Robust Fragile Metabolism Obesity, diabetes Regeneration & repair Fat accumulation Cancer Healing wound /infect Insulin resistance AutoImmune/Inflame Proliferation Inflammation Fragility Hijacking, side effects, unintended Of mechanisms evolved for robustness Complexity control, robust/fragile tradeoffs Math: robust/fragile constraints ( conservation laws ) Both Accident or necessity?

131 fragile robust Some features robust to some perturbations Other features or other perturbations

132 Increased complexity? fragile robust Some features robust to some perturbations Other features or other perturbations

133 Robust Modular Simple Plastic Evolvable and xor Fragile Distributed Complex Frozen Frozen tradeoffs

134 Slow Flexible Turing architecture Slow Software Hardware Digital Analog Fast Flexible Fast Inflexible Inflexible

135 Slow Flexible Slow Horizontal App Transfer Software Hardware Digital Horizontal HW Transfer Analog Fast Flexible Inflexible

136 Cyber-physical: decentralized control with internal delays.

137 Decision-making components Decentralized, but initially assume computation is fast and memory is abundant.

138 Plant is also distributed with its own component dynamics

139 Internal delays between components, and their sensor and actuators, and also externally between plant components

140 Huge theory progress in last decade, year, mo., Going beyond black box: control is decentralized with internal delays.

141 The best case study so far DNAp Gene Repl. DNA Layered architecture of the bacterial biosphere RNAp Ribosome xrna transc. RN A transl. AA Proteins ATP ATP Catabolism Precursors Not done here in detail, see slides elsewhere

142 How? Universal architectures

143 Slow execution Flexible reprogramming Software Faster execution Less flexible Software Hardware Modern technology gives lots of intermediate alternatives. Digital Analog

144 Applications Control, share, virtualize, and manage resources Operating System Processing Memory I/O Want to emphasize the differences between these two types of layering. Software Hardware Digital Analog

145 Horizontal App Transfer What matters is the OS. Applications Operating System Software Hardware Horizontal HW Transfer Digital Analog

146 Horizontal Applications App Transfer Some people write apps and build hardware But most software and hardware is acquired by horizontal transfer from others Similarly, most new ideas (humans) and new genes (bacteria) are acquired horizontally Operating System Software Hardware Horizontal HW Transfer Digital Analog

147 Slow Flexible Very Slow Process Slow Horizontal App Transfer Software Hardware Digital Analog Fast result Fast Flexible Fast Fast Inflexible Inflexible Inflexible

148 Slow Flexible Slow Software Hardware Technology Evolution Digital Analog Fast Flexible Fast Fast Inflexible Inflexible Inflexible

149 Slow Flexible Acquire Translate/ integrate Automate Horizontal Meme Transfer Motor Prefrontal Striatum Reflex Very Slow Process Sensory Fast Inflexible

150 Slow Flexible Motor Prefrontal Sensory Slow Fast Horizontal App Transfer Flexible Software Hardware Horizontal HW Transfer Universal Architecture Striatum Reflex Digital Analog Fast Inflexible Inflexible

151 Slow Flexible DNAp RNAp Ribosome Gene Repl. DNA Software Hardware xrna transc. RN A transl. AA Proteins ATP ATP Catabolism Precursors Motor Digital Analog Prefrontal (bacterial) biosphere Technosphere Cognisphere Striatum Reflex Sensory Fast Inflexible

152 Flexible/ Adaptable/ Evolvable DNAp RNAp Ribosome Gene Repl. DNA xrna transc. RN A transl. AA Proteins Horizontal Gene Transfer Software Hardware Horizontal App Transfer ATP ATP Catabolism Precursors Horizontal Meme Motor Transfer Digital Analog Prefrontal Striatum Reflex Sensory Depends crucially on layered architecture

153 Horizontal App Transfer Horizontal Meme Transfer Horizontal Gene Transfer Most software and hardware new ideas (humans) new genes (bacteria) is acquired by horizontal transfer, though sometimes it is evolved locally

154 Exploiting layered architecture Horizontal Bad Meme Transfer Virus Horizontal Bad App Transfer Fragility? Horizontal Bad Gene Transfer Virus Parasites & Hijacking

155 Depends crucially on layered architecture Build on Turing to show what is necessary to make this work. Acquire Translate/ integrate Automate Horizontal Gene Transfer Horizontal App Transfer Horizontal Meme Transfer Amazingly Flexible/ Adaptable

156 Compute Turing Slow Flexible Software Universal laws and architectures Delay is even more important Hardware Digital Analog Fast Inflexible Control Bode Control Sense Plant Act

157 Compute Turing Why Necessity

158 delay a w u K P x x px w u t 1 t t t a p 1 delay a

159 delay a K No delay or no uncertainty w u x u px w t a t t x 0 u w P x px w u t 1 t t t a p 1

160 delay a u K x No delay or no uncertainty u px w t a t t x 0 u w w P With delay and uncertainty x px w u t 1 t t t a p 1 a x u p w

161 Linearized pendulum on a cart m q m L u x M M

162 2 q cosq q sinq M m x ml u xcosq lq g sinq 0 y x l sinq linearize l m M m x mlq u x lq gq 0 y x q l M

163 Robust =agile and balancing

164 Robust =agile and balancing

165 m l l M Efficient=length of pendulum (artificial)

166 2 x 1 ls g u D s s Mls M m g q D s s 2 ls g y x lq 1 D s M m x mlq u x lq gq y x q l 0 g m g p 1 r r z l M l transform+ algebra 1 Control Low l l m

167 noise error p 1 l T j E N l Delay

168 1 ln T j d 0 0 Easy, even with eyes closed No matter what the length Proof: Standard UG control theory: Easy calculus, easier contour integral, easiest Poisson Integral formula

169 Harder if delayed or short l Short l Delay M

170 Also harder if sensed low r m M Low l l m l Control M

171 noise error p 1 l T j E N l Delay

172 noise error p 1 l T j E N l Delay Delay is hard 1 2p 1 ln T j d ln T ( ) 2 2 mp p p p l 0

173 noise error Any control p This holds for any controller so is an intrinsic constraint on the difficulty 1 l l of the problem. T j E N Delay Delay is hard 1 2p 1 ln T j d ln T ( ) 2 2 mp p p p l 0

174 1 2p 1 ln T j d p 2 2 p l 0 Fragility 1 l Too fragile up For fixed length L down large small 1/ small large 1/ 1/delay

175 1 2p 1 ln T j d p 2 2 p l 0 We would like to tolerate large delays (and small lengths), but large delays severely constrain the achievable robustness. large small 1/ small large 1/

176 p 1 l M Short is hard l Delay 1 2p 1 ln T j d ln T ( ) 2 2 mp p p p l 0

177 length L 1 2p 1 ln T j d ln T ( ) 2 2 mp p p p l Fragility p 0 1 l Too fragile up For fixed delay Why oscillations? Side effects of hard tradeoffs L down

178 length L 1 2p 1 ln T j d ln T ( ) 2 2 mp p p p l 0 Fragility 1 p l Too fragile up The ratio of delay between people is proportional to the lengths they can stabilize. L down

179 Eyes moved down is harder (RHP zero) Similar to delay m l l M

180 m Suppose r 1 M Units M g 1 r m M l l m l 2 ls g y x lq s ls g M p g g 1 z p 1 z l l z p 1

181 Compare g 1 p 1 r p0 p0 1 l Move eyes g m g 1 p 1 r r z l M l 1 1 p z 1 r 1 r 2 1 p g z l 3 3 p r p 1 p 1 3 l l m m r M M l l

182 Fragility ln z z 1 2z z p ln S j d ln 2 2 z z p 0 1 2p z p ln T j d ln p z p p p change length 1 1 r z p 1 down This is a cartoon, but can be made precise. z p 1 change sense l L l

183 Fragility Hard limits on the intrinsic robustness of control problems. Must (and do) have algorithms that achieve the limits, and architectures that support this process. ln z z p p 1 2 ln z S j d ln z p 2 2 z z p 0 down This is a cartoon, but can be made precise. L

184 Fragility Really slow high delay low delay Slow Fast How do these two constraints (laws) relate? ln z z p p Undecidable Flexible/ General Decidable NP Inflexible/ Specific down 1 2 ln z S j d ln z p 2 2 z z p 0

185 Delay comes from sensing, communications, computing, and actuation. Delay limits robust performance. control Delay noise l 1 2p 1 ln T j d ln T ( ) 2 2 mp p p p l 0

186 noise How do these two constraints (laws) relate? control l Computation delay adds to total delay. Computation is a component in control. large delay small delay Delay Flexible This is about speed and flexibility of computation. Inflexible

187 Fragility 1 2p 1 ln T j d p 2 2 p l 0 Delay makes control hard. Computation delay adds to total delay. Computation is a component in control. 1 l down large large delay small delay Flexible up L small computation. Inflexible

188 How general is this picture? fragile simple tech Implications for human evolution? Cognition? Technology? Basic sciences? robust complex tech efficient wasteful

189 I recently found this paper, a rare example of exploring an explicit tradeoff between robustness and efficiency. This seems like an important paper but it is rarely cited.

190 1 m Phage Bacteria

191 Phage lifecycle Survive Lyse Infect Multiply

192 fragile thin small Capsid Genome Survive robust thick big fast Multiply slow

193 UG biochem, math, control theory Chandra, Buzi, and Doyle Most important paper so far.

194 Theorem! Fragility 1 ln z z S j d p 2 2 ln z z p 0 z and p functions of enzyme complexity and amount ln z z p p simple enzyme complex enzyme Savageaumics Enzyme amount

195 fragile Hard tradeoff in glycolysis is robustness vs efficiency absent without autocatalysis too fragile with simple control plausibly robust with complex control z z 10 1 Simple, but p too fragile p complex 1 ln 0 z S j d 2 2 z z p ln z p No tradeoff k expensive

196 System Emergent : Nontrivial consequences of other constraints Architecture =Constraints Protocols Components

197 Systems requirements: Survive in hostile environments Constraints Components and materials: Chemistry

198 Constraints Constrained ( conserved ): Moieties 1. NAD 2. Adenylate 3. Carbon 4. phosphate 5. oxygen 6. Oxidized state of metabolites 7. Reduced state of metabolites 8. High energy potential release Components and materials: Chemistry

199 Bacterial biosphere carriers: ATP, NADH, etc Precursors, Enzymes Translation Transcription Replication Protocols Architecture = protocols = constraints that deconstrain

200 Catabolism Precursors Systems requirements: functional, efficient, robust, evolvable Co-factors Carriers Hard constraints: Thermo (Carnot) Info (Shannon) Control (Bode) Compute (Turing) Genes Trans* DNA replication Proteins Diverse Universal Control Protocols Constraints Diverse Components and materials: Energy, moieties

201 Systems requirements: functional, efficient, robust, evolvable 1 ln 0 z S j d 2 2 z ln z p z p Constraints Constrained ( conserved ): Moieties 1. NAD 2. Adenylate 3. Carbon 4. phosphate 5. oxygen 6. Oxidized state of metabolites 7. Reduced state of metabolites 8. High energy potential release Protocols Components and materials: Energy, moieties

202 Catabolism Precursors ATP Inside every cell almost Enzymes Crosslayer autocatalysis Building Blocks Macro-layers AA ATP RNA DNA transl. transc. Repl. Proteins Ribosome xrna RNAp Gene DNAp

203 Catabolism Precursors ATP Energy Enzymes Crosslayer autocatalysis Building Blocks Macro-layers AA ATP RNA DNA transl. transc. Repl. Proteins Protein biosyn xrna Gene Ribosome RNAp DNAp

204 Catabolism Precursors ATP Energy energy ATP Rest of cell Yeast anaerobic glycolysis Minimal model

205 Catabolism Precursors ATP intermediate metabolite Energy Reaction 1 ( PFK ) x Autocatalytic feedback energy ATP Rest of cell Yeast anaerobic glycolysis Reaction 2 ( PK ) Minimal model

206 control feedback Reaction 1 ( PFK ) x Reaction 2 ( PK ) h control g ATP energy Tight control creates weak linkage between power supply and demand disturbance Rest of cell Robust = Maintain energy (ATP concentration) despite demand fluctuation

207 Constrained ( conserved ): Moieties 1. NAD 2. Adenylate 3. Carbon 4. phosphate 5. oxygen 6. Oxidized state of metabolites 7. Reduced state of metabolites 8. High energy potential release 1 ln z z S j d p 2 2 ln z z p 0

208 Hard tradeoff in glycolysis Fragile disturbance energy ATP Rest of cell Robust Robust = Maintain energy (ATP concentration) despite demand fluctuation

209 Fragile disturbance Accurate vs sloppy What makes this hard? 1. Instability (autocatalysis) 2. Delay (enzyme amount) Robust Robust Disturbance rejection Accurate

210 Fragile What makes this hard? 1. Instability 2. Delay Robust The CNS must cope with both Today s important point

211 enzymes catalyze reactions Reaction 1 ( PFK ) energy ATP Rest of cell Reaction 2 ( PK )

212 enzymes Reaction 1 ( PFK ) enzymes catalyze reactions, another source of autocatalysis energy ATP Rest of cell Reaction 2 ( PK ) Protein biosyn enzymes Efficient = low metabolic overhead low enzyme amount

213 enzymes reaction rates enzyme amount Can t make too many enzymes here, need to supply rest of the cell. enzymes energy ATP enzymes catalyze reactions, another source of autocatalysis Rest of cell Protein biosyn Efficient = low metabolic overhead low enzyme amount ( slow reactions)

214 Fragile? Robust = Maintain ATP ATP Robust? Efficient Wasteful Efficient = low enzyme amount ( slow reactions)

215 fragile Hard tradeoff in glycolysis is robustness vs efficiency absent without autocatalysis too fragile with simple control plausibly robust with complex control z z 10 1 Simple, but p too fragile p complex 1 ln 0 z S j d 2 2 z z p ln z p No tradeoff k expensive

216 What (some) reviewers say to establish universality for all biological and physiological systems is simply wrong. It cannot be done a mathematical scheme without any real connections to biological or medical universality is well justified in physics for biological and physiological systems a dream that will never be realized, due to the vast diversity in such systems. does not seem to understand or appreciate the vast diversity of biological and physiological systems a high degree of abstraction, which make[s] the model useless

217 This picture is very general fragile simple tech Implications for human evolution? Cognition? Technology? Basic sciences? robust complex tech cheap large delay fast multiply metabolic expensive small slow

218 This picture is very general Domain specific costs/tradeoffs metabolic overhead cheap metabolic expensive CNS reaction time (delay) large small phage multiplication rate fast multiply slow

219 This picture is very general fragile simple tech robust complex tech metabolic cost cheap expensive reaction time large phage x rate fast Domain specific costs/tradeoffs small slow

220 thin small Capsid thickness Genome size fragile Survive robust thick big fast multiply slow

221 1 2p 1 ln T j d p 2 2 p l 0 Fragility 1 l Too fragile up For fixed length L down large small 1/ small large 1/ 1/delay

222 Fragility 1 2 ln z S j d ln z p 2 2 z z p 0 r m m M ln z z p p change length 1 1 r down This is a cartoon, but can be made precise. change sense L l M l

223 fragile z z 10 1 Simple, but p too fragile p complex Hard tradeoff in glycolysis is robustness vs efficiency absent without autocatalysis too fragile with simple control plausibly robust with complex control No tradeoff k 1 ln 0 metabolic cost z S j d 2 2 z ln z z p p

224 Really slow Slow Computational complexity Turing has the original universal law Fast Decidable NP P Flexible/ General Inflexible/ Specific

225 Fragility 1 2p 1 ln T j d p 2 2 p l 0 Delay makes control hard. Computation delay adds to total delay. Computation is a component in control. 1 l down large high delay low delay Flexible up L small computation. Inflexible

226 Fragility 1 l 1 2p 1 ln T j d p 2 2 p l 0 up This needs formalization: What flexibility makes control hard? Large, structured uncertainty? down large small

227 What about: Cyber-physical: decentralized control with internal delays?

228 delay a u K x No delay or no uncertainty u px w t a t t x 0 u w w P With delay and uncertainty x px w u t 1 t t t a p 1 a x u p w

229 Focus on delays delay a K u x w P x px w u t 1 t t t a p 1 Actuator delay a=act

230 Focus on delays Actuator delay a=act

231 0 Decentralized control t=transmission 0 l=internal s=sense r=internal a=act p=plant

232 l p a s r l=internal s=sense r=internal a=act total remote + plant delay p=plant

233 0 Communications delay t=transmission 0

234 t l p a s r Then decentralized control design can be made convex 0 l t t l p a s r r 0 s p a Rotkowitz, Lall, and a cast of thousands including Lamperski, Parrilo (Friday)

235 A primary driver of human brain evolution? t 0 Want t p a s 0 s a s small a p

236 Wolpert, Grafton, etc robust Brain as optimal controller Acquire Translate/ integrate Automate Reflex

237 Huge theory progress in last decade, year, mo., Going beyond black box: control is decentralized with internal delays.

238 Decision-making components in the brain Decentralized, but initially assume computation is fast and memory is abundant.

239 Plant is also distributed with its own component dynamics

240 Internal delays between brain components, and their sensor and actuators, and also externally between plant components

241 Internal delays involve both computation and communication latencies

242 Compute Turing Delay is most important This progress is important. New progress! Communicate Shannon Delay is least important Carnot Bode Control, OR Heisenberg Einstein Boltzmann Physics

243 Huge theory progress in last decade, year, mo., Going beyond black box: control is decentralized with internal delays.

244 Going beyond black box: control is decentralized with internal delays. Slow Flexible Motor Prefrontal Sensory Mammal NS seems organized to reduce delays in motor control Striatum Reflex Fast Inflexible

245 Universal architectures Implications (Layered architectures discussed elsewhere)

246 Turing as new starting point? Essentials: 0. Model 1. Universal laws 2. Universal architecture 3. Practical implementation Software Hardware Digital Analog Turing s 3 step research: 0. Virtual (TM) machines 1. hard limits, (un)decidability using standard model (TM) 2. Universal architecture achieving hard limits (UTM) 3. Practical implementation in digital electronics (biology?)

247 Horizontal App Transfer What matters is the OS. Applications Operating System Software Hardware Horizontal HW Transfer Digital Analog

248 Flexible/ Adaptable/ Evolvable DNAp RNAp Ribosome Gene Repl. DNA xrna transc. RN A transl. AA Proteins Horizontal Gene Transfer Software Hardware Horizontal App Transfer ATP ATP Catabolism Precursors Horizontal Meme Motor Transfer Digital Analog Prefrontal Striatum Reflex Sensory Depends crucially on layered architecture

249 Sequence ~100 E Coli (not chosen randomly) ~ 4K genes per cell ~20K different genes in total ~ 1K universally shared genes DNAp Gene Repl. DNA RNAp Ribosome xrna transc. RN A transl. AA Proteins Horizontal Gene Transfer ATP ATP Catabolism Precursors See slides on microbial biosphere laws and architectures.

250 selection + drift + mutation + gene flow + facilitated variation Horizontal Gene Transfer large functional changes in genomes

251 natural selection + genetic drift + mutation + gene flow + facilitated variation Genome can have large changes Genes

252 natural selection + genetic drift + mutation + gene flow + facilitated variation Small gene change can have large but functional phenotype change Architecture Phenotype Genotype

253 natural selection + genetic drift + mutation + gene flow + facilitated variation Only possible because of shared, layered, network architecture Phenotype Architecture Genotype

254 Standard theory: natural selection + genetic drift + mutation + gene flow Greatly abridged cartoon here Gene alleles Selection Shapiro explains well what this is and why it s incomplete (but Koonin is more mainstream)

255 Standard theory: selection + drift + mutation + gene flow Pheno- type Selection Gene alleles

256 Standard theory: selection + drift + mutation + gene flow Phenotype Gene Selection No new laws. No architecture. No biology.

257 selection + drift + mutation + gene flow Phenotype Gene alleles No gap. Selection All complexity is emergent from random ensembles with minimal tuning. No new laws. No architecture.

258 The battleground Phenotype Phenotype Gene alleles No gap. Just physics. Huge gap. Need supernatural Genes?

259 Phenotype Gene alleles What they agree on No new laws. No architecture. No biology. No gap. Huge gap. Phenotype Genes

260 Depends crucially on layered architecture Analog Digital Horizontal Gene Transfer Hardware Software Motor Horizontal App Transfer Prefrontal Striatum Reflex Sensory Horizontal Meme Transfer Amazingly Flexible/ Adaptable

261 Putting biology back into evolution

262 Universal architectures What can go wrong?

263 Unfortunately, not intelligent design Ouch.

264 Why? left recurrent laryngeal nerve

265 Why? Building humans from fish parts. Fish parts

266 It could be worse.

267 Slow Flexible DNAp RNAp Ribosome Gene Repl. DNA Software Hardware xrna transc. RN A transl. AA Proteins ATP ATP Catabolism Precursors Motor Digital Analog Prefrontal (bacterial) biosphere Technosphere Cognisphere Striatum Reflex Sensory Fast Inflexible

268 delay=death sense move Spine

269

270

271

272 Vestibuloocular reflex

273 Fast Slow

274 Ac Move head Sense Fast Act Fast? Slow Sense Same actuators Delay is limiting Move hand Slow Act

275 Ac Versus standing on one leg Eyes open vs closed Contrast young surfers old football players Move head Sense Fast Act Same actuators Delay is limiting Fast? Slow Sense Move hand Slow Act

276 delay=death sense move Spine

277 Reflect Reflex sense move Spine

278 Reflect Reflex sense move Spine

279 Reflect Reflex sense move Spine

280 Reflect move

281 Reflect Reflex sense move Spine

282 Layered architectures (cartoon) Physiology Organs Cells

283 sense Reflect Reflex move Spine Prediction Goals Actions Actions errors

284 Cortex Meta-layers cartoon Prediction Goals Actions Physiology Actions Organs errors

285 Visual Cortex Why? Visual Thalamus? 10x

286 Prediction Goals Conscious perception Prediction Goals Actions What are the consequences? Physiology Actions Organs Why? Visual Cortex 10x errors Visual Thalamus There are 10x feedback neurons?

287 Seeing is dreaming Conscious perception 3D +time Simulation Zzzzzz.. Conscious perception

288 Same size?

289

290 Same size

291 Same size

292 Same size Toggle between this slide and the ones before and after Even when you know they are the same, they appear different

293 Same size? Vision: evolved for complex simulation and control, not 2d static pictures Even when you know they are the same, they appear different

294 Seeing is dreaming Conscious perception 3D +time Simulation + complex models ( priors ) Zzzzzz.. Conscious perception

295 Conscious perception Seeing is dreaming 3D +time Simulation + complex models ( priors ) Conscious perception Prediction errors

296 Inferring shape from shading

297

298 Which blue line is longer?

299 Which blue line is longer?

300 Which blue line is longer?

301 Which blue line is longer?

302 Which blue line is longer?

303 Which blue line is longer?

304 Which blue line is longer? With social pressure, this one. Standard social psychology experiment.

305

306 Chess experts can reconstruct entire chessboard with < ~ 5s inspection can recognize 1e5 distinct patterns can play multiple games blindfolded and simultaneous are no better on random boards (Simon and Gilmartin, de Groot)

307 When needed, even wasps can do it.

308 Polistes fuscatus can differentiate among normal wasp face images more rapidly and accurately than nonface images or manipulated faces. Polistes metricus is a close relative lacking facial recognition and specialized face learning. Similar specializations for face learning are found in primates and other mammals, although P. fuscatus represents an independent evolution of specialization. Convergence toward face specialization in distant taxa as well as divergence among closely related taxa with different recognition behavior suggests that specialized cognition is surprisingly labile and may be adaptively shaped by species-specific selective pressures such as face recognition.

309 Fig. 1 Images used for training wasps. Published by AAAS M J Sheehan, E A Tibbetts Science 2011;334:

310 weak fragile slow Human evolution hands feet All very skeleton different. muscle skin gut long helpless childhood strong robust fast Apes How is this progress?

311 weak fragile Homo Erectus? This much seems pretty consistent among experts regarding circa 1.5-2Mya hands feet skeleton muscle skin gut Roughly modern Very fragile strong robust So how did H. Erectus survive and expand globally? efficient (slow) inefficient wasteful

312 endurance weak fragile speed & strength Apes strong robust (fast) efficient (slow) inefficient wasteful

313 weak fragile (slow) Human evolution hands feet skeleton muscle skin gut Apes strong robust (fast) efficient (slow) inefficient wasteful

314 weak fragile Architecture? Evolvable? Hard tradeoffs? Apes strong robust efficient (slow) inefficient wasteful

315 endurance weak fragile + sticks stones fire teams From weak prey to invincible predator? strong robust efficient (slow) Speculation? There is only evidence for crude stone tools. But sticks, fire, teams might not leave a record?

316 weak fragile Speculation? With only evidence for crude stone tools. But sticks and fire might not leave a record? strong robust + sticks stones fire teams efficient (slow) From weak prey to invincible predator Before much brain expansion? Plausible but speculation?

317 Cranial capacity Before much brain expansion? Gap? Today 2Mya

318 weak fragile hands feet skeleton muscle skin gut Key point: Our physiology, technology, and brains have coevolved strong robust + sticks stones fire efficient (slow) From weak prey to invincible predator Before much brain expansion? Probably true no matter what Huge implications.

319 Cranial capacity Today 2Mya

320 weak fragile strong robust + sticks stones fire efficient (slow) hands feet skeleton muscle skin gut From weak prey to invincible predator Before much brain expansion? Key point needing more discussion: The evolutionary challenge of big brains is homeostasis, not basal metabolic load. Huge implications.

321 weak fragile Architecture? strong robust + sticks stones fire efficient (slow) inefficient wasteful

322 weak fragile strong robust + sticks stones fire efficient (slow) hands feet skeleton muscle skin gut From weak prey to invincible predator Before much brain expansion?

323 weak fragile Architecture? strong robust + sticks stones fire efficient (slow) inefficient wasteful

324 Human complexity? fragile Consequences of our evolutionary history? robust sticks stones fire efficient wasteful

325 fragile Constraints (that deconstrain) robust Hard tradeoffs? efficient wasteful

326 Catabolism Precursors ATP Environment Enzymes Action Building Blocks AA transl. Proteins ATP Ribosome RNA transc. xrna RNAp DNA Repl. Gene DNAp Control >90% of most bacterial genomes

327 Catabolism Precursors expensive Fast Inflexible expensive inflexible ATP Enzymes Building Blocks AA ATP transl. Proteins Ribosom cheap flexible fast RNA DNA slow transc. Repl. cheap xrna Gene Slow Flexible RNA DNA

328 Slow, Broad scope Meta-layers Unfortunately, we re not sure how this all works. Fast, Limited scope

329 Flexible/ Adaptable/ Evolvable DNAp RNAp Ribosome Gene Repl. DNA xrna transc. RN A transl. AA Proteins Horizontal Gene Transfer Software Hardware Horizontal App Transfer ATP ATP Catabolism Precursors Horizontal Meme Motor Transfer Digital Analog Prefrontal Striatum Reflex Sensory Depends crucially on layered architecture

330 New sciences of complexity and networks? worse Science as Pure fashion Ideology Political Evangelical Nontech trumps tech Edge of chaos Self-organized criticality Scale-free networks Creation science Intelligent design Financial engineering Risk management Merchants of doubt

331 Theorem! Fragility 1 ln z z S j d p 2 2 ln z z p 0 z and p functions of enzyme complexity and amount ln z z p p simple enzyme complex enzyme Savageaumics Enzyme amount

332 Fragility hard limits General Rigorous First principle complex simple Overhead, waste? Plugging in domain details Domain specific Ad hoc Phenomenological

333 Control Wiener Comms Kalman Bode robust control Shannon General Rigorous First principle? Fundamental multiscale physics Foundations, origins of noise dissipation amplification catalysis Heisenberg Boltzmann Carnot Physics

334 What I m not going to talk much about It s true that most really smart scientists think almost everything in these talks is nonsense Why they think this Why they are wrong Time (not space) is our problem, as usual Don t have enough time for what is true, so have to limit discussion of what isn t No one ever changes a made up mind (almost) But here s the overall landscape

335 Control Comms Complex networks Wildly successful Compute New sciences of complexity and networks edge of chaos, self-organized criticality, scale-free, Stat physics Heisenberg Carnot Boltzmann Physics

336 Popular but wrong D. Alderson, NPS 336

337 Complex systems? Fragile Even small amounts can create bewildering complexity Scale Dynamics Nonlinearity Nonequlibrium Open Feedback Adaptation Intractability Emergence

338 Complex systems? Robust Scale Dynamics Nonlinearity Nonequlibrium Open Feedback Adaptation Intractability Emergence Fragile Scale Dynamics Nonlinearity Nonequlibrium Open Feedback Adaptation Intractability Emergence

339 Complex systems? Robust complexity Scale Dynamics Nonlinearity Nonequlibrium Open Feedback Adaptation Intractability Emergence Resources Controlled Organized Structured Extreme Architected

340 These words have lost much of their original meaning, and have become essentially meaningless synonyms e.g. nonlinear not linear Can we recover these words? Idea: make up a new word to mean I m confused but don t want to say that Then hopefully we can take these words back (e.g. nonlinear = not linear) Fragile complexity Scale Dynamics Nonlinearity Nonequlibrium Open Feedback Adaptation Intractability Emergence

341 New words Emergulent Emergulence at the edge of chaocritiplexity Fragile complexity Scale Dynamics Nonlinearity Nonequlibrium Open Feedback Adaptation Intractability Emergence

342 Complex networks New sciences of complexity and networks edge of chaos, self-organized criticality, scale-free, doesn t work Alderson & Doyle, Contrasting Views of Complexity and Their Implications for Network-Centric Infrastructure, IEEE TRANS ON SMC, JULY 2010 Stat physics

343 Complex systems? Control Comms Complex networks Compute Stat physics Carnot Jean Carlson, UCSB Physics Heisenberg Boltzmann Physics

344 Complex networks Alderson &Doyle, Contrasting Views of Complexity and Their Implications for Network-Centric Infrastructure, IEEE TRANS ON SMC, JULY 2010 Control Sandberg, Delvenne, & Doyle, On Lossless Approximations, the Fluctuation- Dissipation Theorem, and Limitations of Measurement, IEEE TRANS ON AC, FEBRUARY, 2011 Stat physics Heisenberg Carnot Boltzmann Physics

345 The last 70 years of the 20 th century will be viewed as the dark ages of theoretical physics. (Carver Mead) orthophysics Complex networks From prediction to mechanism to control Sandberg, Delvenne, & Doyle, On Lossless Approximations, the Fluctuation- Dissipation Theorem, and Limitations of Measurement, IEEE TRANS ON AC, FEBRUARY, 2011 Stat physics, fluids, QM Heisenberg Boltzmann Carnot Physics

346 J. Fluid Mech (2010) Streamlined Laminar Flow Turbulence and drag? Turbulent Flow

347 Physics of Fluids (2011) Dennice Gayme, Beverley McKeon, Bassam Bamieh (UCSB ME), Antonis Papachristodoulou, John Doyle

348 Physics of Fluids (2011) z y x high-speed region Flow downflow u upflow low speed streak y z x U w Coherent structures and turbulent drag 3D coupling Blunted turbulent velocity profile Turbulent Laminar U w

349 U w fragile Laminar Turbulent Laminar robust? Turbulent efficient wasteful

350 Disturba nce Existing design frameworks Sophisticated components Poor integration Limited theoretical framework Fix? Slow, Broad scope Interface Comms Remote Sensor Fast, Limited scope Sensor Actuator Plant