Fifty years ago, we put humans on the moon. They stayed a short time and left. Now, the goal is more ambitious and considerably more complicated — far too complicated for any person or even group of people to visualize completely.

For your interest, here is a brief thought on how that visualization might be accomplished. We call it:

THE UNIVERSAL ENGINEER
An Artificial Intelligence System in Which Everything Communicates With Everything Else

The Question: How can separate intelligences communicate well enough that their collective activity becomes capable of solving problems no individual intelligence could solve?

The Assignment: Establish 50 humans on the Moon and create a self-sustaining civilization capable of persisting for thousands of years without essential support from Earth.

I. THE PROBLEM

Human beings solve complicated problems by breaking them down. Medicine is divided into specialties. Engineering is divided into specialties. Science is divided into disciplines. Governments are divided into departments.

A lunar settlement similarly would have specialists in propulsion, structures, radiation, agriculture, medicine, psychology, energy, communications, manufacturing, genetics, education, and thousands of other subjects.

Division makes complicated problems intellectually manageable. It also creates a fundamental weakness. Specialization divides a reality that is not divided. The food specialist may discover the best way to grow food without knowing that the solution creates a problem for the water specialist. The water specialist solves that problem but increases power requirements, which creates huge problems for the power specialist. Etc.

Today, AI data centers are creating huge power, labor, construction, political and other economic problems, with each sector vying for dominance. But what is best for the future of America? How will we determine that?

For the Moon mission, each response changes cost, reliability, energy consumption, manufacturing requirements, and available cargo capacity. And somewhere far down the chain, a solution to a food problem may have changed the psychological environment of a child who will be born on the Moon 40 years later.

Every solution becomes a new system stimulus. Every stimulus produces responses. Those responses become new stimuli. The consequences propagate. Human brains can follow only a tiny fraction of those interactions.

We suggest that a system we call “The Universal Engineer” would be designed specifically to track and solve such complex problems.

II. NATURE ALREADY HAS SOLVED A SIMILAR PROBLEM

A human body contains trillions of cells performing an extraordinary variety of specialized functions. No central intelligence consciously directs all of them. The kidneys regulate their domain. The lungs regulate theirs. So do the heart, liver, digestive system, skin, endocrine system, nervous system, immune system, muscles, bones, and blood vessels.

They are specialized without being isolated. They continually communicate. The brain does not calculate every heartbeat or decide how much sodium every nephron should reclaim. In effect, the architecture says: “Do what you do. Sense what is happening. Respond locally when appropriate. Tell the rest of the organism what has changed. Escalate what requires broader attention.”

That is the model for the Universal Engineer.

It would not be one gigantic artificial intelligence attempting consciously to think about everything. It would be a society of specialized intelligences joined by a universal communication system.

The intelligence of the Universal Engineer would reside not merely in its individual specialists, but in each of their ability to weigh options and to communicate with all other specialists and ultimately with a decision central.

III. THE FUNDAMENTAL OBJECTIVE

Everything begins with one objective: Maximize the probability that an initial population of 50 humans can establish a self-sustaining lunar civilization capable of surviving for thousands of years without essential support from Earth.

That is the trunk of the decision tree. Everything else is subordinate. Cost, safety, comfort, food production, energy, genetics, psychological stability, and political stability all matter. But none is independently the objective. Each matters because of its effect upon the probability of long-term survival.

The Universal Engineer therefore does not ask merely, “Which habitat is safest?” It asks, “How does this degree of safety, including everything required to produce it and every consequence produced by it, affect the overall probability of long-term survival?

IV. THE ORGANS OF THE UNIVERSAL ENGINEER

The lunar-survival problem initially might be divided into major domains such as:

1. Transportation and logistics;
2. Habitat and physical protection;
3. Atmosphere and environmental control;
4. Water, food and biological cycling; e
5. Energy and physical resources;
6. Human biology and medicine;
7. Reproduction, population and genetics;
8. Psychology, society and governance;
9. Knowledge, education and communication;
10. Manufacturing, maintenance and technological adaptation.

These divisions are provisional. Reality does not contain departments. Humans invented departments because human minds require them. The Universal Engineer should be free to redraw those boundaries whenever another organization better represents reality.

More importantly, the domains must never become informational islands. Every part of the Universal Engineer (UE) must be capable of receiving information generated by every other part. That is how nature created humans, with every organ communicating to and from the entire body

For a living creature, the above might correspond to kidneys, heart, intestines, bones, etc. each of which communicates with all the others, to a greater or lesser degree, via nerves, blood, etc. For example, a pain in your toe might be felt in your knee via what is called “referred” pain.” No part of the human body is isolated from the rest.

Thus, though, for instance, the kidneys can be seen as a blood cleansing computer system, the results of its work are communicated to every part of the body. That is the system used by the Universal Engineer.

V. THE UE  COMMUNICATION SYSTEM

The UE should be designed around a principle that living organisms demonstrate extraordinarily well: Specialize locally. Communicate globally.

Information should travel through many physical channels because different physical phenomena reveal different aspects of reality. The UE therefore should possess the technological equivalents of multiple biological communication systems.

Optical “bloodstream.” High-capacity optical networks would carry enormous quantities of general information throughout the system. This is the informational circulation system: observations, measurements, models, hypotheses, predictions, warnings, failures, uncertainties, decisions, and changes of state continually flowing through the UE.

Electronic nervous system. Electronic communication would provide extremely rapid local sensing, signaling, and control. Where milliseconds or microseconds matter, information should not wait for higher-level deliberation. Local systems should respond locally while simultaneously reporting what occurred.

Radio system. Wireless communication would connect mobile machinery, astronauts, sensors, vehicles, satellites, robots, remote installations, and components for which physical connections are impractical.

Acoustic and vibration system. Structures and machinery constantly communicate their physical condition through vibration. Bearings, pumps, pressure vessels, habitats, excavation equipment, rotating machinery, and structural members all possess physical signatures. Changes in those signatures may reveal deterioration long before conventional failure alarms occur. The UE should listen.

Thermal system. Temperature itself carries information. Every machine, habitat, organism, battery, chemical process, agricultural system, and electronic component produces a changing thermal signature. The UE should perceive the lunar settlement not merely as objects, but as a continuously changing three-dimensional thermal field.

Mechanical system. Pressure, strain, acceleration, load, torque, displacement, vibration, and movement continually reveal the physical state of structures and machines. These are not merely engineering measurements. They are communications from physical reality.

Chemical system. The atmosphere, water supply, soil or growth media, waste streams, human bodies, plants, microorganisms, industrial processes, and medical systems continually communicate through chemistry. Oxygen, carbon dioxide, trace gases, hormones, metabolites, nutrients, contaminants, pathogens, salts, proteins, and thousands of other substances are information. The chemistry itself is part of the communication network.

Radiation system. The radiation environment must become another continuously sensed information field. Solar activity, cosmic radiation, shielding performance, accumulated biological exposure, electronic degradation, material changes, and local variations would become part of the Universal Engineer’s continuously updated understanding of the colony.

And these categories should not be considered exhaustive. The UE should continually ask, “What physical phenomenon exists from which useful information can be extracted?”

Magnetic fields, electromagnetic emissions, gravity, electrical resistance, capacitance, optical changes, chemical gradients, microscopic deformation, sound, radiation, heat, pressure, and phenomena not initially recognized as useful communication channels all become candidates.

Nothing should be dismissed merely because humans traditionally have not called it a communications system.

VI. EVERYONE HEARS; EACH LISTENER DECIDES WHAT MATTERS

This may be the most important architectural principle of the Universal Engineer. Ordinary communication systems often require the sender to determine who should receive a message. That creates the dangerous assumption that the sender must know who needs to know.

But in an enormously complicated system, the sender frequently cannot know. A microscopic change in atmospheric chemistry might matter to medicine, agriculture, corrosion, manufacturing, psychology, fire protection, water processing, or some specialty whose relevance nobody anticipated.

Therefore, information should not be restricted only to recipients whom the sender believes will need it. All relevant information will enter the common circulation. Every function can receive it. This is a corollary the human bloodstream, where every function receives blood, and each function uses the information it needs. The kidneys use different information from the blood vs. the lungs, bones, nose, etc.

The intelligence lies in the receiving filters. Each specialist AI continuously asks, “Does this matter to me?” At each level, most information will be ignored. Some will cause minor adjustments. Some will be retained because they might become significant when combined with later information. A tiny fraction will trigger immediate attention.

The filters themselves will learn. An agricultural AI that previously ignored a particular vibration frequency might discover that the frequency predicts a pump condition affecting nutrient delivery. From then on, its filter changes.

Thus communication becomes not only Universal, but evolutionary. The system, as a whole, learns not only what information means, but who needs to hear it.

VII. NO SINGLE COMMUNICATION CHANNEL IS AUTHORITATIVE

This redundancy is essential. A pump might report electronically that it is operating normally, but its vibration signature might say otherwise. Its electrical consumption might have changed, or its temperature might be increasing. The UE should not merely ask what the pump’s electronic controller reports, but what does everything we can observe about this pump collectively tell us?

The same principle applies to humans, plants, buildings, machines, ecosystems, and eventually the entire lunar civilization. Different channels provide partially independent descriptions of the same reality. Disagreement among them is information.

VIII. THE UNIVERSAL MODEL

At the center of the UE system would be a continuously changing computational representation of the entire lunar community. It would be a dynamic network containing entities, relationships, histories, constraints, probabilities, assumptions, uncertainties, and competing explanations.

Every significant communication changes that model. Every proposed action enters it as a new stimulus. The UE then propagates the consequences.

A proposed agricultural change affects water which affects energy, which affects heat in an ongoing cascade of information.

The process continues until additional consequences become sufficiently improbable or sufficiently insignificant that computational resources are better used elsewhere. The Universal Engineer continually decides what deserves further thought.

IX. WEIGHTING

The UE must weight possibilities.

Every significant predicted consequence would be evaluated according to such considerations as probability, magnitude, reversibility, uncertainty, time horizon, detectability, repairability, resource requirement, and effect upon the fundamental survival objective.

A one-in-a-thousand possibility capable of destroying the colony may deserve more attention than a near certainty of minor inconvenience. The weights themselves remain provisional. The system must be capable of saying, “Our recommendation depends heavily upon Assumption X. If X is wrong, Recommendation B becomes preferable to Recommendation A.”

X. HISTORY IS PART OF THE MACHINE

Every observation, prediction, experiment, failure, successful adaptation, mistaken assumption, and unexpected interaction changes what the Universal Engineer knows. The system therefore must possess persistent, shared history.

Today’s response becomes part of tomorrow’s starting condition. A failure is not merely a failure, but information that changes future possibilities.

The UE consequently becomes more capable through use because every significant interaction changes the filters, models, probabilities, relationships, and assumptions with which the next problem is approached. This information is promulgated throughout the system.

XI. THE SURPRISE SYSTEM

One major part of the Universal Engineer should have no responsibility for producing food, generating energy, maintaining atmosphere, protecting health, or accomplishing any other conventional mission. Its assignment should be to learn what else has been missed.

It continually asks, “What if?” “What assumptions are being shared?” “What systems thought to be independent actually interact?” “What low-probability events would become catastrophic when they occur together?” “What measurements disagree?” “What information is everyone filtering out?”

This is the UE’s intellectual “immune system.”

XII. THE HUMAN ROLE

Humans retain authority over major decisions, but the Universal Engineer should provide more than recommendations. It should explain:

1. We considered these alternatives.
2. These are the most probable outcomes.
3. These are the consequences we traced.
4. These are the weights we used.
5. These are the major uncertainties.
6. These assumptions most strongly affect our conclusion.
7. Change this assumption or this weight, and the preferred solution changes.

The comparison is artificial intelligence with incomplete information versus humans possessing considerably less information and considerably less ability to integrate it. Ignorance and difficulty are not show-stoppers.; they are show starters.

XIII. WHAT CAN BE BUILT NOW

Much of the foundation already exists.

We already possess distributed computing, enormous data storage, fiber-optic communication, wireless networking, specialized AI processors, scientific simulation, machine learning, optimization, digital twins, knowledge graphs, sensor networks, model-based systems engineering, and multimodal artificial intelligence.

We already can measure vibration, chemistry, temperature, radiation, strain, movement, electrical activity, electromagnetic fields, biological activity, and thousands of other physical variables.

The missing invention is not any one of these components. The missing invention is their integration into a system in which information communicated everywhere can become useful anywhere.

XIV. WHAT MUST BE DEVELOPED

The first research program should concentrate upon communication.

It must develop a common informational language capable of translating among physics, chemistry, medicine, biology, psychology, agriculture, engineering, economics, and disciplines that do not yet exist.

It must develop persistent shared memory so that discoveries, failures, decisions, assumptions, and uncertainties remain available throughout the system.

It must develop universal information circulation in which relevant state changes enter common streams accessible throughout the Universal Engineer.

It must develop intelligent receiving filters capable of deciding which tiny fraction of that information matters to each specialist.

It must develop cross-channel correlation so that electronic, optical, thermal, mechanical, chemical, acoustic, magnetic, biological, and radiation observations can be recognized as different descriptions of the same event.

It must develop dynamic weighting so that computational attention flows toward consequences whose probability, uncertainty, magnitude, or irreversibility justify additional thought.

And it must develop the ability to discover new communication channels.

XV. LET THE UNIVERSAL ENGINEER DESIGN THE UNIVERSAL ENGINEER

Humans should not attempt to specify the final architecture. First, build the best preliminary Universal Engineer present technology permits. Then give assign it to design a better UE. Version 0.1 proposes Version 0.2.

The proposed architecture is simulated and tested. Version 0.2 inherits what worked and changes what did not. Then Version 0.2 helps design Version 0.3. Every generation inherits the accumulated history of the generations preceding it.

The system evolves — not through random biological mutation, but through deliberate proposal, testing, selection, and retention. Eventually its architecture may bear little resemblance to what its human designers originally imagined.

XVII. THE FIRST DEMONSTRATION

Give the Universal Engineer the lunar-survival problem from its first day, not because it immediately can solve it, but every failure exposes a weakness.

The lunar civilization therefore serves two purposes simultaneously.

It is the first great problem the UE attempts to solve.

And it is the instrument used to create Universal Engineering itself.

THE LARGER PURPOSE

Climate systems, human disease, cities, economies, ecosystems, spacecraft, energy systems, biological organisms, and civilizations share a fundamental characteristic: Everything affects everything else.

Human civilization has accumulated specialized knowledge while retaining a limited ability to combine that knowledge. The Universal Engineer is an attempt to overcome that limitation.

Its fundamental principle is borrowed from life itself. Filter locally. Integrate everywhere. Specialize locally. Communicate globally.

UE’s purpose is not to know everything. Its purpose is to make it increasingly difficult for something important to remain unknown merely because nobody thought to tell the right specialist.

The UE requires no new laws of physics in order to begin. The computers exist. The sensors exist. The communication technologies exist. The artificial intelligences are beginning to exist.

What does not yet exist is the architecture that allows all of them to function as parts of one communicating organism. We should build that architecture first.

Complex problems are not difficult merely because they contain many parts. They are difficult because the parts affect one another. And its solution: Divide the work, not the information.

Consider budgeting, i.e.  the question: On what should we best spend our dollars? That is a UE question.

The introductory question was: How can separate intelligences communicate well enough that their collective activity becomes capable of solving problems no individual intelligence could solve?

The moon is but one of many objectives that Universal Engineering can help us reach

Rodger Malcolm Mitchell

 

 

Humans seem to present a peculiar evolutionary irony. We are the animal capable of growing extraordinarily long facial hair, but we also are the animal that invented shaving. That raises an intriguing possibility: the ability to remove facial hair may have helped remove any disadvantage associated with excessive facial-hair growth.

Shaving didn’t cause genes for long beards. Rather, once humans could cut their beards, excessive growth became less of an evolutionary liability. A cultural invention potentially changed the environment in which biological characteristics were selected.

That suggested a much broader phenomenon. The solution changes the problem.

Humans repeatedly invent ways to overcome natural constraints. But once we overcome a constraint, the solution itself becomes part of the environment. We then change—biologically, culturally, behaviorally, or structurally—in response to the new environment we ourselves created.

Clothing and housing allow us to survive with relatively little protective body hair. So those born with scant hair no longer perished. They produced children without much hair.

Eyeglasses allow people with poor eyesight to function and reproduce successfully. Medicine allows people to survive and reproduce under conditions that otherwise might have killed them.

Tools reduce the survival advantage of physical strength. Agriculture produces plants and animals increasingly dependent upon human intervention. Domesticated woolly sheep survive because humans select and protect animals whose excessive wool would otherwise be a hindrance. 

The same phenomenon appears outside biological evolution. Suppressed small forest fires and fuel accumulate, potentially contributing to larger fires later. Build levees to protect floodplains, and people build more extensively behind the levees; when the levees fail, the cost is greater.

Build a wall to protect the people inside, and the same wall that prevents others from entering prevents those inside from leaving.

Create weapons for defense, and you simultaneously create weapons for offense.

Create government to protect individual lives and freedoms, and you create an institution possessing the power and motivation to restrict those same freedoms.

These aren’t adequately described merely as unintended consequences. There is something more interesting about them.

Ironic Consequences.

The solution alters the system so that the conditions confronting the solution itself change. Sometimes the invention that frees us from a constraint ultimately becomes a new constraint. We become dependent upon the thing that liberated us. That leads to what always seems to be the hardest question—not “What?” or “How?” but “Why?”

Why are Ironic Consequences so Common?

It is tempting to imagine that nature somehow “resists” our interventions—that there is a natural order which, when disturbed, attempts to restore itself. But that introduces exactly the sort of anthropomorphism we must avoid. 

Nature doesn’t resist. Systems respond. Every system we encounter already embodies an enormous history. An organism, forest, river, society, government, climate, or ecosystem consists of innumerable interacting parts, each one having relationships that are consequences of everything that previously happened to them.

Then humans intervene. We tend to imagine the intervention simply as Single Problem —> Solution. But the universe doesn’t contain isolated problems.

When a solution enters an existing system, every component capable of interacting with the change responds according to its own structure, constraints, and history. Those responses alter other conditions and become new stimuli. Those produce still more responses.

Thus, every solution becomes a new stimulus to the system it changes. That may be the fundamental source of Ironic Consequences. We plan the one response we want and the universe supplies all the other responses.

The Boulder and the Twig

This doesn’t mean our interventions are fruitless. A boulder rolling downhill has an enormous, accumulated trajectory. Every bump and pebble affects that trajectory, however slightly, but most don’t reverse it. The boulder continues downhill.

A twig dropped into a raging river will go somewhere. We may predict the general result quite well while being utterly incapable of predicting precisely where the twig first will touch land.

Some future outcomes are heavily weighted. Living organisms eventually die. Rivers flow downhill. Unsupported objects fall. What remains difficult to predict are the supposedly “minor” details—when, where, how, and through precisely what sequence of events.

The important correction is that the boulder never returns to its original trajectory after encountering the pebble. The original trajectory no longer exists. Every second finds the boulder on a slightly different trajectory, incorporating the pebble into its history.

So, what appears to be nature’s inertia isn’t resistance to change. It is the weighting of accumulated history relative to the new stimulus.

That gives us a useful formulation: History does not determine the future. It weights it. Everything that interacts has consequences, and consequences themselves become initiators of further consequences. Every microscopic event becomes important by its weighting. Most tiny effects are overwhelmed by much larger interactions and disappear into the background.

Occasionally, however, a small event enters a system where feedback amplifies it enormously. Give the young Einstein a ball. Suppose he chases it into the street and is killed. An apparently trivial event—giving him a ball — could alter scientific history, military history, political history, and innumerable individual histories.

The ball doesn’t cause all those later events in any simple sense. It changes the history from which all subsequent possibilities emerge. Again, weighting determines whether a tiny perturbation disappears or becomes consequential.

History Cannot be Undone.

This also corrects the idea that nature eventually “snaps back.” Suppose humanity disappeared tomorrow. Forests might reclaim cities. Buildings eventually would collapse. Many obvious traces of civilization would disappear. But Earth never could become the Earth that would have existed had humans never evolved.

We have changed species, genomes, landscapes, atmospheric chemistry, migration patterns, ecosystems, and countless other things. Those changes already have caused still more changes.

Humanity eventually may disappear, but human history cannot unhappen. The future Earth would forever be an Earth whose history included humans. The footprints may disappear, but the effects of making them do not.

And Then There is AI

AI may become the largest example yet of an Ironic Consequence. We create AI because human minds cannot handle certain quantities and complexities of information. AI performs those tasks.

Humans consequently may cease developing or maintaining some of the abilities AI replaces. Then society reorganizes around AI’s availability. Eventually, we created AI because we couldn’t do certain things, and having AI caused us to stop doing many things. Therefore, we could no longer do those things without AI.

The tool created to overcome a limitation could make us increasingly dependent upon the tool. But that same problem provides perhaps the strongest argument for AI.

Human engineering generally solves problems discretely. We identify the individual result we want and try to produce it. But every solution becomes a stimulus to an immensely interconnected system.

A genuinely sophisticated AI might ask not merely, “Will this intervention solve the problem?” but rather. “How will the entire system respond to several interventions?”

“What constraints will disappear? What previously disadvantageous characteristics then may persist? What behaviors will change? What dependencies will develop? What second-order, third-order and n-order consequences will those changes produce?

And what new problems will exist precisely because the original problem successfully was solved?

Universal Engineering

This brings Ironic Consequences directly into the idea of Universal Engineering—identifying and weighting all the variables involved in any solution to a problem.

We never will predict everything. The interacting possibilities are too numerous, and every new response changes the conditions confronting the next event. We will use AI to analyze the consequences of our decisions further into the future.

AI wins at chess because it can “see” the results of more options, further ahead. A weak player thinks, “If I make this move, what happens?” A stronger player thinks, “If I make this move, what can my opponent do, what can I then do, what can my opponent then do…?”

The number of possible futures explodes with every additional move. Humans cope by intuition, experience, pattern recognition, and—very importantly—discarding possibilities. We simply cannot examine everything.

AI can examine vastly more possibilities and carry their consequences further forward before assigning weights to the alternatives. That’s Universal Engineering, writ small.

Suppose instead of a chess move we’re considering a new agricultural practice. The human specialist might ask what it does to crop yield. A better analysis adds cost, water consumption, fertilizer requirements, pests, soil, labor, nutrition, biodiversity, etc.

But every one of those consequences branches again. Change the insects—>changes the birds–>changes seed distribution—>changes vegetation—>changes erosion—>changes waterways…

Very quickly we have a “chessboard” whose number of possible positions is beyond human comprehension.

Now consider constructing the system that will allow permanent human residence on the Moon. How many interrelated variables are there? What are the possible problems, and how will the solutions to those problems affect all the other variables?

Have the “spaceship” people understood all the “living quarters” problems? Have the “living quarters” people understood all the “food” problems, the “psychological” problems, the “sun” problems, the “political” problems, and the countless others? And can every group factor the consequences of its individual solutions into every other group’s problems?

No human mind can manage that multidimensional chessboard. Perhaps no collection of human minds can. But an AI, given vastly more information than any individual human can comprehend, could examine far more of the interactions, follow their consequences further, and continually reweight the possibilities as one proposed solution changes all the others.

The question is not merely whether a solution works. The question is what happens to everything else if it is implemented. That is Universal Engineering. 

And here’s the especially important connection to Ironic Consequences: The chess program doesn’t evaluate a move solely by examining the immediate result. It evaluates the positions that the move makes possible. That’s precisely what we’re saying engineering needs to do.

We shouldn’t ask merely, “Does the levee prevent the flood?” We should ask, “What new world begins to exist if we build a levee or do something else?”

The funding has consequences. The construction project has consequences. The finished product has consequences. People build behind it. Property values change. Wetlands change. Sediment distribution changes. Political expectations change. Maintenance obligations arise. Dependence upon the levee develops. And each of those creates another branching tree.

AI wins at chess not because it knows one perfect move, but because it can examine and weight the consequences of more possible moves further into the future. Universal Engineering applies the same principle to the vastly more complicated chessboard of reality.

And to make things even more appropriate for AI, chess has fixed rules, known pieces, and a defined objective. Reality has none of those conveniences. We don’t even know all the pieces, much less all their interactions.

If the search space is already overwhelming on 64 squares, imagine it when everything potentially communicates with everything else, and each response changes the board. Chess resets the pieces after every game. History never resets the board.

But AI’s Universal Engineering is better at tracking and weighing the variables. AI wins at chess because it can see more options, further ahead. But perhaps AI can allow us to see considerably more of the river before we throw in the twig. 

The underlying principle is that nothing changes alone. Every change enters an already existing web of interactions. Every system responds to the changes. Those responses become new stimuli. Some consequences disappear. Others persist.

Persistence becomes History. History changes the weighting of subsequent possibilities. Then something else happens. History → Event → Response → New History.

And again. Our plans aren’t futile merely because their exact consequences are unpredictable. We constantly alter the weighting of future possibilities, often successfully.

But we should remember that we never change only the thing we intend to change. Perhaps that is the fundamental logic behind Ironic Consequences: Every solution changes the system that made the solution necessary.

And sometimes, by freeing ourselves from an old constraint, we create the conditions that make us dependent on our new freedom.

Oh, and remember that boulder rolling downhill, bouncing over the pebbles? What became of those pebbles? Did just one of them strike another, which struck another, which dislodged a rock, which eventually began a landslide? Or did nothing significant happen at all?

Those are precisely the kinds of consequences humans usually cannot follow. We see the boulder. We don’t see all the pebbles, much less what happens to each of them afterward.

AI’s Universal Engineering would be better able to see the pebbles, too.

Rodger Malcolm Mitchell

 

The more you consider all the things your brain and the rest of your body can accomplish, the more amazed you become that so little can do so much. Consider all the things a tiny mosquito can accomplish, from breeding to finding food to coping with disease, predators, and the changing weather and landscape — all this from a 2.5-milligram (0.000088184905 ounces) creature with a brain barely visible with the naked eye.

We have not found a way to create a machine as efficient as a mosquito — let alone a human being. Part of the reason is that nature is the ultimate multi-tasker.

This post began with recent articles about fascia and the interstitium. (Don’t worry. This won’t be technical.)

For centuries, fascia and the fluid-filled spaces associated with it were thought primarily to be mechanical. Fascia helped hold the body’s structures together. Interstitial fluid provided cushioning and allowed organs and tissues to move without damaging one another.

They were, in effect, scaffolding and lubricant.

But increasingly we are learning that this apparently humble system participates in communication throughout the body. Interstitial spaces and fluids provide pathways through which chemical substances, cells, mechanical forces, and perhaps other signals can travel.

Once again, something in a living body that seemed to be just a simple, single-purpose mechanism, turned out not to be a a complex, multi-purpose function.  Bones are not just supports. Blood is not just a carrier of oxygen. Skin is not just a covering. Muscle is not just a motor. Fat is not just stored energy. 

Structures, substances, movements, chemical reactions, temperature changes, pressure changes, and even what we call waste products frequently perform multiple functions. Life does an astonishing amount with comparatively little.

The Battery Problem

Human engineering traditionally approaches problems differently.

Suppose you wished to are design an electric bicycle. It needs stored electrical energy. So, you add a battery. The battery has a job: provide electricity.

Then you need someplace to put it, a structure to support it, wiring to connect it, a system to protect it, perhaps a cooling mechanism, sensors to monitor it, and electronics to control it.

If nature made your battery, you might expect to discover that the battery casing also strengthened the bicycle frame; its mass improved balance; its heat performed some useful function; its electrical system communicated information; its structure absorbed impacts; its sensors monitored other parts of the bicycle; and perhaps some substance produced by one of its operations became an input required somewhere else.

This suggests a different engineering question. Instead of asking, “How can we make a better battery?” you ask, “What else can the battery do?” Or an even harder question: “How else could an energy source be configured so that it performs additional useful functions?”

And then the almost impossibly difficult question: “How would every one of those configurations affect every other component and every other function of the bicycle?”

Thus, the problem explodes. An energy source can have countless shapes, sizes, locations, chemistries, voltages, structural properties and operating temperatures. Every variation can interact differently with the frame, wheels, rider, motor, brakes, electronics and environment.

Even a bicycle creates a combinatorial problem that rapidly outruns unaided human imagination. Now try designing a rocket, or an amoeba, and ask that same question, where the number of possible configurations approaches infinity.

The Engineering Opportunity for Artificial Intelligence

This may be one of the great opportunities for artificial intelligence. Human builders manage complexity largely by dividing it. One group designs the engine. Another designs the structure. Others design the electrical system, the communications, the cooling.

This is extraordinarily effective, but partly it reflects the limitations of the human brain. We cannot hold millions of interacting variables and possibilities in mind simultaneously. AI potentially can explore vastly larger spaces of configurations.

Instead of telling an AI, “Optimize this battery,” we might eventually tell it, “Consider the entire machine and everything it needs to accomplish. Find arrangements in which every component performs as many useful functions as possible. Then determine how every proposed change affects the performance of everything else.”

That is much closer to nature’s method. Not because nature thinks. It doesn’t. Nature has no intention. It just is.

Everything Communicates

The comparison becomes even more striking when we consider communication. Computers communicate primarily through deliberately constructed electrical and optical pathways. We build processors, memory, wires, buses, networks and communication protocols.

A living organism makes that look primitive. The human body communicates electrically, but electricity is only one channel. We internally communicate through blood, hormones, interstitial fluid, neurotransmitters, ions, protein shapes, molecular concentrations, and temperature.

Oh, and did I mention pressure., acidity, mechanical stretching, chemical gradients, timing, frequency, repetition, spatial location, light, sound, touch, odor, and taste?

The body doesn’t consist simply of components attached to a communications network. The interactions among the components are themselves communication networks. And communication need not be direct.

A cell in the toe need not send a private message to a particular cell in the liver. It can alter something that alters something else that changes a circulating substance that affects another system that eventually alters the conditions experienced by that liver cell. The effects propagate. In that sense, an organism is an almost incomprehensibly interconnected system.

Everything Has Its Own “Subroutines”

“Subroutine” is an analogy. An electron does not contain a little computer program telling it what to do. A more precise term might be “constraints” or “characteristics.”

Throw a ball against a wall and a great number of things happen. The ball deforms. The wall moves, however slightly. The air moves. Sound is generated. Heat is generated. Forces propagate through the materials. Atoms and molecules change positions. Electromagnetic interactions occur.

No central computer calculates all this and issues instructions:. Instead, every component simply responds according to what it is and the conditions affecting it. Fields, atoms, molecules, and each does their thing, and the macroscopic event we call “the ball bounced” emerges from all of them.

Nature doesn’t calculate the bounce. There is no calculation. The result is the accumulation of Stimuli —>Responses, trillions of times every second.

The Ten-Body Problem

We cannot write a general closed-form solution to even the three-body gravitational problem. Nature has no such difficulty. Put ten or a thousand gravitating bodies together and each will move in response to the gravitational situation created by all the others.

It would be misleading to say they consciously “coordinate.” There is no coordination center. Each simply does its own thing while simultaneously affecting the conditions under which the other nine do their things. Now replace ten bodies with trillions of interacting components.

Nature still does not require a central calculator. This may suggest an important principle for future engineering. Perhaps the ultimate solution to overwhelming complexity is not to construct a sufficiently enormous central computer capable of calculating everything.

Perhaps it is to design components with appropriate local rules and interactions so that useful global behavior emerges. Biology does this constantly. Cells self-organize. Immune systems respond without a central commander. Embryos develop structures without a central architect specifying the coordinates of every cell. Ant colonies exhibit complex collective behavior without a chief executive ant.

The whole emerges from interactions among parts.

But How Did Nature Find Any of This?

Here we encounter what may be the most interesting problem of all. The weighting problem. The number of theoretically imaginable arrangements of matter is staggering.

The familiar analogy of monkeys randomly typing Shakespeare illustrates the difficulty. If every possible sequence of characters were equally probable and every failed attempt had to be discarded completely before beginning again, producing Hamlet by chance would require an absurd amount of time.

Yet Hamlet exists. So does an amoeba. How?

The answer cannot simply be that nature tried every conceivable possibility independently. Even nature doesn’t employ infinite time.

The Dice Were Loaded From the Beginning

Nature begins with constraints. Atoms cannot do absolutely anything. Their structures permit some interactions and prohibit others. Charges attract and repel. Atoms form some bonds readily and others reluctantly or not at all. Molecular shapes permit certain interactions. Water favors some structures and destabilizes others. Temperature changes which reactions are likely. Pressure matters.

Catalysts make some pathways enormously more probable than others.

Long before Darwinian selection begins, physics and chemistry already have eliminated or enormously reduced huge regions of possibility. Permit and prohibit are the sieve through which possibility is strained.  Physics, chemistry, environment, weights, structure all weight the dice.

Then, once replication appears, Darwinian selection adds another extraordinarily powerful weighting mechanism: Persistence. A stable arrangement remains available for subsequent interactions. An unstable arrangement disappears. What persists becomes part of the conditions under which the next events occur.

Then Replication changes the game. A structure that makes copies of itself does more than persist physically. It multiplies its presence in the future possibility space. Variation creates alternatives. Selection changes their relative prevalence.

Successful structures become starting points for further experiments. Nature does not return to zero after every failure. It keeps what persists.

Inherited Solutions

This may be one of the principal ways nature escapes combinatorial impossibility. Each level inherits enormous amounts of already-solved machinery from below. Chemistry does not have to reinvent the proton. Molecules inherit the properties of atoms. Proteins inherit molecular chemistry. Cells inherit proteins and membranes. Multicellular organisms inherit cellular machinery. Nervous systems inherit cells. Brains inherit nervous systems. Language inherits brains.

Shakespeare inherited language.

Nature does not repeatedly search from fundamental particles. It builds upon what already persists. Solutions become components of subsequent possibilities. Yesterday’s result becomes part of today’s starting conditions. 

It’s as though every engineer knew of every success and failure in history, and so, began creating with many dead ends already avoided.

Shakespeare Wasn’t a Random Monkey

This also explains why Shakespeare could accomplish what random monkeys effectively could not. Shakespeare wasn’t selecting randomly from every possible sequence of characters. His possibility space already had been constrained and weighted by English grammar and vocabulary. He further was constrained by human psychology, Elizabethan culture, theater, his previous reading and writing, and his memories.

The sentence he had just written constrained what could plausibly come next.

After writing, “To be, or not to be…” not every possible subsequent sequence of letters remained equally probable. Hia history had weighted his next response. And that brings us to a much broader principle.

History

Everything begins with possibilities. Possibilities encounter limits, which create constraints, which weight what can happen next. Then something happens. Some consequences persist, which changes the conditions under which the next event occurs.

So we might describe the process:

Possibility—>Constraint —>Weighting—> Interaction/Stimulus—>Response—>Persistence—>History—>New Possibilities

Then repeat.

The River

History, in this sense, is not merely a record kept somewhere, but the physical result of what previously happened. The eroded riverbank is the river’s history. And that history affects where tomorrow’s water flows.

A river provides a useful analogy. The water constantly changes, yet we call it the same river, because the river has location persistence. It has banks, a channel, tributaries, gradients, sediment and an accumulated physical structure produced partly by its own previous flow. Today’s river is constrained by yesterday’s river.

But today’s flow also erodes one bank, deposits sediment elsewhere and changes the channel.

Therefore: Existing structure—>Incoming water—>Flow—>Changed structure—>Different future flow

The river’s history influences its response to stimuli, which changes its history. The water changes continuously while the pattern persists. This begins to sound remarkably like a “Self” or “Consciousness.”

From Nature to Consciousness

The same principles apply to a living organism. A stimulus arrives and organism responds according to its existing physical structure. That response changes the organism, however slightly. The changed organism now confronts the next stimulus differently. This can be expressed as: I Am = My History, and My History = My Physical Structure × (Stimulus —> Response) —>My New Physical Structure—> My New History

This is not intended as a mathematical equation. It is a conceptual description. My History is not merely a diary of what happened to me, but what has physically made me. Memories, habits, learning, and skills must be physically instantiated.

The tendency to respond in one way rather than another must exist somehow in the organism’s present physical state, or it could not affect the next response. Thus: History + Stimulus —>Responses —> New History may describe not merely learning, but something fundamental about that thing we call “Self.”

The Illusions of Perception

Consider vision. We casually say that we “see” an apple, but the brain never receives an apple. Electromagnetic radiation interacts with the eye. Neural activity results. The brain constructs the experience we call seeing an apple. We do not experience electromagnetic wavelengths. We experience red.

Likewise, we don’t experience oscillating air pressure. We experience sound. We don’t experience molecular binding to receptors. We experience taste and smell. Everything we sense is a translation or construction.

This becomes obvious when the external source disappears. Close your eyes and you still may see colors and shapes. Tinnitus produces an experience of sound without corresponding sound waves entering the ear from the environment. Dreams can generate entire experienced worlds.

During a dream, the dream may feel completely real. Only upon waking does the brain reclassify it: That was a dream. Occasionally even that classification fails. A remembered event may leave us genuinely uncertain: “Did that happen, or did I dream it?”

The experience was real. What may have been false was its apparent source. So perhaps the distinction is not real perception versus constructed perception. All perception is constructed. The difference is how strongly the construction is constrained by events outside the nervous system.

Reality As Weighting

What we experience as reality itself may involve weighting. Vision and touch say “X.” Hearing is consistent with “X,” and memory predicts “X.” Other sources confirm “X.” The mutually reinforcing signals give enormous weight to the belief that “X” is externally real.

With tinnitus, the auditory system says “Sound.” Other systems fail to corroborate. Your history says, “I have experienced tinnitus before.” Your conclusion becomes “The sound experience is real, but, an external sound probably isn’t.”

During a dream, internally generated systems may agree sufficiently with one another that the brain temporarily concludes that “This is real.” Then we wake and a flood of differently constrained information arrives and the weighting changes.

It was a dream.

And Then, Self

Among all the brain’s constructions, perhaps none is more powerful than: “I am.” I feel that I am the same person who woke yesterday. I feel related to the person I was ten years ago and I even feel partly like the person I was seventy years ago.”

Obviously, I have changed enormously, yet something persists. At my fiftieth high-school reunion, faces had changed enough to become difficult for me to recognize. Then an old classmate spoke, and suddenly recognition occured. The voice itself surely had changed too, but something about the manner, cadence, timing, pronunciation, emphasis or pattern persisted.

I recognized not identical matter but persistent patterns.

Is There a Fundamental Self?

It is tempting to imagine a fundamental Self surrounded by more changeable layers. Some characteristics seem extraordinarily persistent, while others change rapidly. I can be tired, angry, frightened, hungry or delighted without ceasing to be myself.

My beliefs can change. My knowledge, relationships, and my body changes, yet the feeling of continuity remains remarkably powerful.

A brain injury could present an important challenge. Damage to relatively small amounts of physical brain tissue can sometimes alter personality, memory, inhibition, preferences, emotional responses or the sense of identity far more profoundly than losing a much larger amount of tissue elsewhere in the body.

Physical magnitude and historical importance are not the same thing. A lost leg may represent kilograms of physical change. A microscopic neural alteration may profoundly change future responses.

Again, the factor is weighting.

Where Is the Self?

If Self is an illusion or a construction, the physical processes producing it nevertheless must occur somewhere., but “somewhere” need not mean one location. There may be no little chamber in the brain marked “SELF,” just as there is no cubic meter of water containing the essential Mississippi River.

The river has a distributed physical existence. So may Self.

Different physical systems may contribute differently. Brain systems involved in memory, body awareness, emotion, perception, prediction and social understanding all may participate. The endocrine system, the immune system and the gut all communicate chemically and neurally.

Sensory organs continually modify the state from which responses emerge.

“Self” therefore may have a physical location without having a point location. It may resemble a topographical map where some regions and systems carry enormous weight while others carry less. Some characteristics are extraordinarily resistant to change because they have been reinforced throughout a lifetime, while others can change in minutes.

I Am = My History

This leads to the proposition: I Am= My History, but “history” is not merely what happened. Rather, it is the present physical structure produced by what happened. At any instant: I Am = My Present Physical History.

Then: Present Physical Structure × Stimulus—>Response—>Changed Physical Structure—>New History

The next stimulus encounters that new history. Usually, the change produced by one ordinary event is tiny compared with the accumulated structure produced by decades of previous events. That may explain the extraordinary persistence of Self.

Yesterday changed me but yesterday was weighted against more than ninety years of accumulated history. A sufficiently powerful event, however, can alter the weighting dramatically. Trauma, learning, disease or brain injury can do it.

Every event modifies the river. While most barely move its banks, some change its course.

The Larger Principle: Universal Solutions.

We began this post with fascia and we ended with “Self.” But perhaps we have been discussing the same process all along. The universe does not begin each moment from scratch. It inherits, existing structure, which constrains possibilities which create weighting. Stimuli and interactions produce responses.

Some consequences persist, which becomes history, which becomes structure, and structure constrains and weights the next response. And complexity accumulates.

There is no central calculator. Everything has its own “subroutines”—its characteristic responses arising from what it is, where it is, what surrounds it and what has happened before. Higher levels inherit the machinery of lower levels.

Atoms inherit the results of physics. Molecules inherit atoms. Cells inherit molecules. Organisms inherit cells. Brains inherit organisms. Thought inherits brains. Culture inherits thought. Shakespeare inherited language.

And tomorrow inherits the results and rules of today. Perhaps that is how nature accomplishes the apparently impossible. It doesn’t examine every conceivable possibility. Every present state already contains the weighting created by the history that produced it. That weighting determines which possibilities are available, which are probable, which disappear immediately, and which persist long enough to become part of the next round.

The future of human invention may depend upon learning to imitate this—not merely by making better individual components, but by creating systems in which components perform multiple functions, communicate through multiple channels, respond locally, alter one another, and allow useful global behavior to emerge.

Artificial intelligence may finally give us the ability to explore some of that complexity beyond the limits of unaided human thought. We have designed rockets, but we are not yet able to design an amoeba from scratch. Perhaps the difference is not merely that the amoeba is complicated.

Perhaps it is that we still think like engineers designing separate parts, while nature operates through inherited constraints, weighting, interaction, persistence and history of the whole, with the whole including all surrounding stimuli.

Perhaps this is the largest lesson nature can teach human engineering. We tend to solve problems discretely. We identify a problem, isolate it, design a solution, and then attach that solution to the larger machine. Need electricity? Add a battery. Need cooling? Add a cooling system. Need communication? Add a communications system.

Nature could not work that way. The human toe was never designed separately from the foot, the foot separately from the leg, or the leg separately from the rest of the body. Every change occurred within an already existing network of interactions. A change in the toe altered forces transmitted through the foot, ankle, leg, pelvis and spine. Those changes affected muscles, circulation, nerves, balance, energy requirements and behavior. Their consequences could propagate throughout the organism—even, ultimately, to conditions affecting the hair on the head.

Nature did not calculate all those consequences. The interactions themselves performed the calculation.

Every variation encountered the whole system. Its consequences propagated wherever they could. Some consequences disappeared. Some persisted. What persisted altered the conditions confronting the next variation.

Thus, nature did not solve millions of independent problems. It produced universal solutions—solutions tested, not against one isolated requirement, but against the interacting consequences of the whole.

That may be precisely what human beings cannot do. Our intelligence succeeds partly by simplifying. We divide overwhelming problems into manageable pieces. We create specialties, departments, components and disciplines. That has made extraordinary technological achievement possible.

But the method has an unavoidable weakness: the world does not remain divided merely because our minds require us to divide it. Change the battery and you may change weight, balance, temperature, structural loads, aerodynamics, manufacturing cost, maintenance, safety, recycling, control systems and a thousand other things. Change any one of those and still more consequences propagate outward.

The number of possible configurations and interactions quickly exceeds anything a human mind—or even a collection of human minds—can simultaneously consider and weight. And that may be one of the fundamental reasons we need artificial intelligence.

Not merely because AI can calculate faster or remember more. The greater possibility is that sufficiently sophisticated AI may allow us to begin solving problems universally: to explore enormous numbers of interacting possibilities, follow their consequences throughout an entire system, weight those consequences against one another, and discover configurations in which one solution simultaneously becomes part of many other solutions.

In other words, AI may allow human engineering to move a little closer to the method that produced the living world. Perhaps the next great step in human invention will come when we stop asking, “How do we solve this problem?” and begin asking, “What solution does the whole system want?”

Rodger Malcolm Mitchell