Energy: The Ultimate Substitute for Land

Elon Musk’s recent talk focused on the Kardashev scale  – a framework for measuring civilization’s advancement by its energy consumption – and argued that moving up that scale requires expanding into space. Which invited comments like this: “I’m sick and tired of his shtick about Kardashev. I get it. The sun has lots of energy. But why would I care what % we’re using!?  Why would that be a goal?” The reason is simple: energy that is super-abundant (and that means cheap) makes a lot of other things possible. The entire progress of human civilization can be seen as harnessing more energy – from human muscle alone to animals, wind, sun, water, combustion, nuclear fission. If we find better sources of more abundant energy, we’ll make human life even better.

As an example, let’s look at farming. For some time in the 20th century, there were concerns about carrying capacity – the ability of the Earth to support a fast-growing population. Vertical farming was one possible solution. Vertical farming  is the practice of growing crops not in soil but in water or air in large buildings resembling hangars or warehouses, where crops are stacked many levels high. Recently, Matt Farrel made a video asking “Why This Vertical Farm is 500x More Efficient Than Farming”. The short answer is that while it is 500x more efficient by land use, this metric doesn’t tell the whole story. While vertical farming is immensely more land-efficient, it has struggled to achieve commercial success due to significantly higher costs compared to traditional farming:

• Energy: Traditional farming relies on a free, abundant energy source: the Sun. Vertical farming requires a massive energy input for lighting and climate control.
• Labor: It requires skilled tech workers who command higher wages than traditional farm hands.
• CapEx: There is a lot more machinery and infrastructure involved, and it is more expensive.

Vertical farming struggled to turn a profit. The anxiety about the Earth’s ability to support a growing population has largely faded now that it’s clear that there won’t be a runaway population growth. And yet, the mere fact that vertical farming can produce crops is great news for the long run, because it proves a vital point: when enabled by technology, energy can be a substitute for arable land. No land – no problem! This is great news:

1. For the Earth this means that carrying capacity in terms of agricultural land is not going to limit us. We can feed far more people; we just need energy.
2. Beyond Earth, this means that a human future on Mars and other space destinations is possible. Even without sufficient sunlight or native soil, we can still feed a planet.

As they say in the aerospace industry, “With enough thrust, even a brick can fly.” With enough energy, we can feed worlds. And we can do a lot of other things in other spheres of human activity. That’s why the Kardashev scale matters. A civilization that has a lot of energy is a more advanced and prosperous civilization.

Focus

As a Software Engineer in the tech industry, I understand the relentless drive to achieve. Many of us strive to eliminate disruptions, enter the mythical flow state, and simply create. Focus is crucial for getting things done and reaching important objectives. Many corporate calendar tools even include a “Focus Time” feature to support this.

This desire for total dedication is vividly captured in Robin Sloan’s Sourdough, where the protagonist, Lois Clary, a young woman fresh from college working for a robotics startup in San Francisco, has no life outside of work. This account of focused work in startup culture is set in a book that begins as an instant historical tale of the early 21st century.

This focus on work, particularly in startup environments, is not a thing of fiction. Indeed, in startups, working hard and long hours is the norm. Venture capitalist Chris Hladczuk, CEO of Hanover, recently asserted that building a billion-dollar startup is not achievable on a 40-hour work week. He caused quite an uproar but he certainly had a point.

But what if you could do even better? What if you could make all disruptions go away? Be completely dedicated to your work for all your waking hours, achieve true Focus with capital F? What’s the catch, you’re asking? Well, I said all disruptions and all the time when you’re not asleep. This idea of complete dedication is taken to an extreme in Vernor Vinge’s book A Deepness in the Sky, where a subgroup of humans known as Emergents push this concept to its terrifying limit.

Emergents’ Focus is a form of mind control, or mind enslavement. It works as a bioengineered virus combined with an active MRI hyper-activating parts of the brain on a millimeter scale. The result makes the target person a savant in their field of specialization but suppresses their individuality. They become obsessed with their work but they lose their social and emotional engagement. They don’t care about human relations and don’t respond to usual rewards and incentives any more. They even neglect their personal needs and hygiene. By the way, to me, Focus sounds somewhat like an artificially induced autism.

In the book, Focus offers a compelling promise. However, as practiced by the Emergents, Focus is undeniably evil. The main protagonist of the book, Pham Nuwen, harbored a lifelong ambition to establish an interstellar human community, aiming to prevent the devastating civilizational collapses that repeatedly plague human space. Upon learning of Focus, Pham initially believed it could be a means to achieve this grand objective. He was aware of the terrible cost, and for years, he wrestled with the ethical dilemma, weighing the price against the potential salvation of billions of lives. Ultimately, Pham concluded that any forced Focusing of people was unacceptable; any degree of slavery was too much.

Yet, Focus isn’t inherently evil. Some individuals could choose to voluntarily partake in it and harvest its benefits. Trixia Bonsol, a human linguist on an expedition to a world of intelligent alien spiders, is an example of that. After she is freed from her enslavement, she asks to be mildly Focused again. Her desire is to transcend being merely the best translator between the languages of humans and alien spiders. She aspires to translate gestures, context, and entire cultures at a level unattainable without complete dedication to the work. Therefore, Focus can be voluntary and can empower individuals to achieve their goals.

This leads to a troubling question: could a culture of job dedication and extremely long hours evolve into something more sinister? If we developed a technology akin to the Emergent’s Focus, would some companies favor hiring focused candidates? And, subsequently, might Focus become a condition of employment? If this were to happen, we would be well on our way to the Emergent’s system of slavery. A culture of forced Focus could even emerge voluntarily in highly competitive environments. Once enough people adopt Focus, those who wish to keep pace would feel compelled to do the same to avoid being left behind. A parallel to this phenomenon is rumored to exist today in some hyper-competitive universities where students use drugs to enhance memory and attention. Once a significant portion of the class does this, others feel pressured to follow suit to avoid losing out.

It’s interesting to consider how Focus works, especially when thinking about our current situation. The output generated by Focused workers requires verification by non-Focused individuals because it can be nonsensical. A system using Focus can’t function without constant, close supervision of non-Focused personnel. In the book, this was described by Trud Silipan, an Emergent manager of Focused personnel. This situation is very much like how AI works now, at least when it comes to writing code.

A further connection between the Focus described in the 1999 book and modern AIs is revealed through the character of Ann Reynolt. As Emergents’ Director of Human Resources, she oversees and manages Focus for the entire expeditionary group. She told Pham that obtaining useful results from Focused personnel requires asking the right questions, sometimes through trial and error. What is this, if not prompt engineering?

Bending over backward for productivity can actually backfire. We aim for efficiency, but sometimes we just create this productivity charade that’s really about control. When we’re chasing goals, we gotta be careful not to lose our humanity.

The Naked Sun or a Naked Threat?

Are the very rules designed to protect us from AI also the keys to its potential for harm? Read on and help me decide.

I’ve recently reread The Naked Sun, one of Isaac Asimov’s novels from the Robot series. one of Isaac Asimov’s novels from the Robot series. The entire series explores Asimov’s Three Laws of Robotics, and how these deceptively simple and seemingly foolproof rules don’t lead to the good outcomes one could imagine. In this book, the author asks this question: could a robot commit a murder? It should be impossible because the First Law of Robotics says “A robot may not harm a human being; nor, through inaction, allow a human being to come to harm”. Yet the answer in the book is positive. He describes a way for robots to commit murder. This constitutes a method by which robots can cause the death of a human. However, since the robots lack the necessary intent to kill (mens rea, or “the guilty mind”), their actions would not be properly called murder. This is critical to maintain compliance with the First Law – a robot can’t intentionally hurt a human, so they need to be manipulated.

Prompt Injection: Asimov’s Foresight

The genius of his concept is chillingly simple. One needs to break up a killing into a sequence of actions, each of them quite innocent if taken out of context. Then one assigns those steps to different robots without revealing to them the larger chain of events. One robot then makes a poison, another pours it into a glass, yet another gives the glass to the intended human target. Everything the robots do appears completely innocent to them yet the result is deadly.
Alarmingly, Asimov’s foresight echoes one of the most pressing challenges in AI security today: prompt injection. It’s especially tricky with something called “multi-turn prompting.” That’s when attackers create a bunch of seemingly harmless prompts. But when you string them together just right, they can trick the AI system into doing something it shouldn’t or spilling sensitive info. It works because the AI remembers previous conversations and uses that context, which makes it a real headache for AI developers and security folks. The great Isaac Asimov got many things wrong but here he was alarmingly spot on.

Autonomous Warfare: Beyond Sci-Fi

The book also describes another way that The First Law could be circumvented: uncrewed space battleships. It was Sci-Fi then but it’s a much more realistic prospect nowadays. We’ve just witnessed terrestrial warfare enter the new age of drone fighting. So far it’s mostly remote controlled devices but AI is reportedly being incorporated into all kinds of drones. So killer robots probably already exist. And modern spacecraft, even those that carry crews, can perform their functions autonomously. So technology already exists to create autonomous combat spacecraft. In The Naked Sun, the author develops an idea that a fully autonomous spacecraft can be fooled into using weapons on other spacecraft despite the First Law of Robotics because it can be convinced that other spacecraft are also autonomous, hence destroying them wouldn’t harm any human beings.

The Opaque AI: A Modern Dilemma

But wait – our situation might actually be worse than what Asimov described. Robots in The Naked Sun are well understood by their designers. Solarians delegated raising children to robots and considered it necessary for robots to be able to discipline children, which, at that time meant physically slapping. They considered it small immediate harm to prevent much worse issues in the future. Such were the times when the book was written (mid-1950s). But this task is problematic for a robot because according to the First Law, a robot can’t harm a human being. Solarian roboticist Jothan Leebig describes the alterations to the robots that he made to enable robots to discipline children as: “strengthening of the C-integral governing the Sikorovich tandem route response at the W-65 level”. Or, when asked to restate in simple terms, “a certain weakening of the First Law”. This behavior alteration, it seems to me, may have been done in software or in hardware – it’s unclear to me, but I don’t think it ultimately matters. What’s important is that the scientist knew exactly what he was doing. The change in the robots’ behavior was clear and fully transparent. That, I believe, is not the case with modern AI. Even before the advent of LLMs popularized by ChatGPT, the value of AI was in the ability to find patterns and solve problems that are too difficult to solve by giving specific step by step instructions. In simple words, AI can tell a dog from a cat without a programmer having to painstakingly code the cat vs dog differentiator. In fact, such a differentiator may be impossibly difficult. Modern AIs work by finding and exploiting patterns in the data. To a non-expert that I am, there seems to be a huge downside: we don’t know how exactly they do it. Worse, we don’t even know what an AI model can do. To a non-expert like myself, the core problem is that we don’t truly understand how they arrive at their conclusions, making them a veritable black box. We give AIs system instructions (example) and claim that they are the equivalent of the Laws of Robotics. But since the model is opaque, how do we know that it would actually obey those instructions? Have we got ourselves a robot without the Three Laws of Robotics?

How bad is it, really?

Finally, I’ll offer pure speculation from a non-specialist. I just use AI, I don’t design or implement them. I wonder if we may be even deeper in trouble. The AI models are trained by feeding them pretty much everything humanity has in a digital form. All the books, movies etc. A lot of books were written and movies produced about misbehaving AIs. But who would make a movie about robots / AIs doing exactly what humans want them to do? Such movies don’t exist for the same reason newspapers don’t report on traffic lights operating correctly. It is the normal and expected behavior and no one wants to read or watch about it. The result is that we feed our AIs a lot of material about how they misbehave and hope to make them behave by giving them instructions afterwards. I’m not sure that those instructions are enough to override all the bad examples they are fed during training. We may be creating a self-fulfilling prophecy of dystopia by training AIs on it. Certainly, AI developers are well aware of these risks and work to mitigate them. But in reality, like in Asimov’s function, this can be easier said than done. This leads me to an unsettling speculation as a non-specialist: by feeding our AIs a steady diet of fictional dystopias, are we inadvertently training them to fulfill these very prophecies?. It would be interesting to hear from someone knowledgeable in training the modern models.

The folly of Reversing a Linked List

Linked List is a popular data structure… on LeetCode. It is part of the standard library in C++ as a plain std::list which I find a bit pretentious given that it is not very useful in real life. The only thing it does well is inserting and deleting items in the middle of the list, if you have already iterated to that item. Under this condition, insertion and deletion are O(1) whereas they are O(n) for more typical lists structures like std::vector in C++ or ArrayList in Java. And that’s just the O-major time complexity analysis. However, this theoretical efficiency may not translate practically due to how linked lists store elements. Their non-contiguous memory allocation results in each element residing at a heap location determined by allocation time and algorithm. Consequently, consecutive elements are unlikely to be adjacent in memory, hindering performance. Traversal necessitates reading from disparate memory locations, increasing the likelihood of cache misses and slow RAM reads that stall the CPU pipeline. By comparison, std::vector is allocated contiguously, so reading it is likely to be all fast cache reads and traversal will be much faster in practice.

Linked lists are inherently unsafe for concurrent code due to their reliance on direct pointer access for efficient insertions and deletions. This direct access can lead to inconsistencies when the list is mutated, as there’s often no mechanism to prevent other threads from accessing the list simultaneously.

The danger lies in the saved pointers that enable efficient operations. While one thread is mutating the list, another thread might use a previously saved pointer to access an element, assuming it to be valid. Even using a mutex to protect the critical section wouldn’t solve this issue, as the saved pointer could still be used to access the list during the mutation.

Let’s explore a common interview question: reversing a linked list. This task is straightforward, achievable in just a few lines of code within a loop using standard programming techniques (temporarily store the next node, redirect the pointer to the previous node, and advance the current and previous pointers). Memory manipulation functions can further streamline the code. However, this operation is inherently risky. During the reversal process, the data structure is compromised. If another entity holds a pointer to the structure, they could encounter unexpected outcomes, at best, or crash the program. 

In C or C++, I don’t see an easy way to make linked lists safe. A different matter is Rust. One of the core concepts of Rust compiler is that if there is a mutable reference to something, there can’t be other references to it at the same time. Add to it the fact that the nodes of the linked list are owned by the list, and the result is that while the list is being mutated, no other references to it can be live, so no one can see a partially mutated list. This might make some code inconvenient to write, but it is much safer.

Miles Vorkosigan and Penric

After reading the Vorkosigan Saga Sci-Fi series by Lois McMaster Bujold, I decided to try her fantasy series, Penric and Desdemona (The World of Five Gods). I found the protagonist of the fantasy series, Penric kin Jurald, to be similar to the main character of the Vorkosigan Saga, Miles Vorkosigan, as both characters are empathetic and highly independent. 

For Miles, there are several great examples:

• Miles encounters his clone who was raised and trained by Miles’ enemies to impersonate him and assassinate his father. Despite this, Miles accepts him as his brother and insists on everyone calling him his brother, not a clone” Mark ultimately refuses to kill Miles. Later, Mark hijacks Miles’ fleet in a well-intentioned but hopeless hostage rescue attempt. Miles risks his life to save Mark during this attempt, and they eventually form a genuine brotherly bond.
• Miles was tasked with retrieving biosamples from the body of a bioengineered soldier. He found the soldier in a dungeon and, instead of just taking the samples, he rescued her. The soldier, now known as Sergeant Tora, joined Miles’ fleet and became one of his most loyal and driven troopers, and occasionally, his lover.

For Penric, one example is in “Penric and the Shaman” story where he is tasked to help capture and bring to justice an escaped magician (“shaman”). Penric, however,  develops a friendship with the shaman and gains an understanding of his perspective. As a result, Penric goes out of his way to assist the shaman in clearing the charges that have been brought against him.

Both Miles and Penric are pretty independent and prefer to work on their own. They do well when they’re given a mission and the freedom to do it their way, without someone looking over their shoulder all the time. This lets them get creative and change their goals as they go.

• The best example of Miles’ ability to adapt and overcome unexpected challenges is likely his mission on Dagula IV. Tasked with rescuing a key individual from a POW camp, Miles infiltrated the camp only to discover the person had died. Undeterred, he orchestrated a daring escape, liberating all the prisoners in the camp. 
• For Penric, it is his mission to Cedonia where his original task was to deliver a letter to a general of that country inviting him to the service of Penric’s employer. But when the general is arrested and blinded, Penric takes it upon himself to treat him, aiming to restore his sight and convince him to take the job. That’s quite a scope creep! 

When I read about magic, I always try to understand how it is implemented. For instance, in the book I’m reading, Penric has a “chaos demon” living in his mind which is a combination of experiences of around twelve humans that the demon previously inhabited. At first glance, it seems like an AI with a seamless brain interface, but it also gives Penric superhuman abilities like night vision and the ability to see diseases within the human body. These could be explained by sensors integrated with the AI, but he can also manipulate matter remotely, like picking a lock or instantly causing rust. This must be real magic because I don’t understand the technology behind it. However, it seems to follow the basic laws of thermodynamics since using magic causes a build-up of “chaos” within the sorcerer’s body, which must be released to prevent overheating. This implies there’s an internal power source, and the issue of heat rejection is a common challenge in many engineering projects.

But then the chaos demon has additional peculiarities. It can do an equivalent of a Jedi mind trick (“these aren’t the droids you’re looking for”). And doing that on an unwilling person causes the host human to bleed. I haven’t yet come up with a satisfactory explanation for that. If you have ideas, please share.

Link Visualizations in Vector Data Unit Tests

When writing unit tests for code that deals with vector geospatial data (points, lines, and polygons on the Earth’s surface), shape visualizers are a critical tool. These tools provide a visual representation of the data, which simplifies and speeds up the process of writing tests. Additionally, the tests themselves become easier to understand and less error-prone, leading to more reliable and maintainable code. It is crucial to remember to add a link to the visualization to your unit test once you have created test data!

Examples of public visualizers for GeoJSON data are geojson.io, geojson.tools or Keene State’s tool. For S2 cell visualization, one option is this.

Raster data (matrices) is often readable enough to understand the test data without visualization. For example, by simply looking at this matrix definition, we can easily make out the shape defined by it:

int labels_data[10][8] = {
 {4, 4, 4, 4, 4, 4, 4, 4},
 {4, 4, 4, 4, 4, 4, 4, 4},
 {4, 4, 4, 4, 4, 4, 4, 4},
 {4, 1, 1, 1, 1, 1, 1, 1},
 {1, 1, 1, 1, 1, 1, 1, 1},
 {1, 1, 1, 1, 1, 1, 1, 1},
 {1, 1, 1, 1, 1, 1, 1, 1},
 {1, 1, 1, 1, 1, 1, 1, 5},
 {5, 5, 5, 5, 5, 5, 5, 5},
 {5, 5, 5, 5, 5, 5, 5, 5},
};

It is more difficult to understand vector data (polygons, points) by simply looking at them lists of coordinates, so external visualization is more important for this type of data than it is for raster data (matrices). It may be possible to understand coordinates for very simple cases such as 2 clearly disjoint rectangles but anything less trivial is difficult to write and read. For example, what is defined in the example below? There are 2 shapes, but what is the relation between the 2 shapes? Are the shapes even valid? 

a1.set_polygon(s2textformat::MakePolygonOrDie("1:1, 1:1.001, 1.0001:1.0001, 1.0001:1"));
a2.set_polygon(s2textformat::MakePolygonOrDie("1:0.9997, 1:1.002, 1:1.0001, 1.0003:1"));

Once you visualize the shapes, you’d realize that you had a wrong vertex in the second shape. After removing it, you have this code and corresponding view.

a1.set_polygon(s2textformat::MakePolygonOrDie("1:1, 1:1.001, 1.0001:1.0001, 1.0001:1"));
a2.set_polygon(s2textformat::MakePolygonOrDie("1:0.9997, 1:1.002, 1.0003:1"));

Now you can clearly see that a square is inside a triangle, touching it on the left side.

Let’s be honest, it’s a pain to come up with good test data for location-based stuff. Just trying to think up coordinates in your head or write them down is a recipe for disaster. A way better approach is to use a visual tool where you can just draw the shapes and areas you need. This makes the whole process of getting test data way easier and less error-prone.
Visualization is a super helpful tool for simplifying tests and making them way easier to understand for code reviewers and anyone who looks at the code in the future. It’s literally a picture that shows how shapes are related to each other, which is so much clearer than trying to explain it with words or numbers.

The most crucial thing I learned about creating unit tests for vector data is to include a direct link to the visualization in the test code. This way others won’t have to either recreate the visualization themselves or simply accept your word for the accuracy of the test data.

If you’re concerned about exposing your company’s trade secrets by uploading test data to a public visualizer, there are 2 ways to address this. 

First, test data is usually too abstract to reveal anything about the code. So one test uses 2 disjoint rectangles and a partially overlapping triangle. What does this say about code under test? Nothing useful. Data can also be altered to avoid leaking sensitive information. 

Using a data visualizer doesn’t automatically make your data public. For example, geojson.io website’s Help section says: “if it’s secret and copyrighted, it will remain that way – it doesn’t have to be public or open source.”

Visualization tools are essential for effective unit testing of vector geospatial data, improving code quality and development efficiency. While data privacy concerns may arise, they can be addressed by recognizing the abstract nature of test data and using private visualization options.

Working with geospatial data: Raster and Vector

Geospatial data can be categorized into two distinct types: raster and vector data. Raster data is structured as a matrix, where each cell represents a rectangular area on the Earth’s surface. In contrast, vector data utilizes polygons defined by the coordinates of their vertices to represent geographic features. While lines and points are also vector data, they won’t be the focus of this discussion. In this post, I will share some insights gleaned from working with both raster and vector data.

Don’t mix raster and vector data

While it’s possible to write code that processes both raster and vector data, it can be challenging to understand and debug such code. It’s best to cleanly separate the parts of the program that work with raster and vector data. If you have both, convert to the one kind that is most suitable for your algorithm, do the work, and then convert back if necessary. For example, you might convert all vector data to raster and run an algorithm on resultant matrices, then convert back to vector.

Conversion requires care, especially from raster to vector

Converting polygons to rasters is typically straightforward, but there are some cases where a clever trick can improve performance. The reverse (raster to polygons) is trickier. Writing a conversion algorithm is easy but writing an efficient one might require a deep understanding of your vector data libraries (GeoJSON, S2 etc). Otherwise your code may be very slow. On one of my projects, a switch from a basic naive vectorizer to a well-optimized one improved performance by more than 1,000 times. 

Use S2 Cell Coverage as an Intermediate Step

For some use cases, using S2 cell coverage as an intermediate step (i.e. going matrix -> S2 cells -> polygon) can be very efficient.

Incremental Vectorization Can Be Slow

In my experience, incremental vectorization (i.e. updating your polygon / polygons after each matrix cell) is going to be slow. If your algorithm is based on unioning polygons or S2 cells, first collect all the parts and then union them all at once. This will be much faster than adding polygons or S2 cells one at a time.

Wing!

I’m excited to announce that I’ve joined Wing, the Alphabet’s drone delivery company. To work in aerospace industry developing a novel product – it feels like a dream!

Ownership of real property on Mars

In settling Mars, it would be important to understand who owns land and buildings on Mars. In economic systems that exist on Earth, protection of property rights has been paramount in enabling investment and development. As of now, the Outer Space Treaty prohibits claiming national sovereignty of any celestial bodies. Without sovereignty, land ownership is hard to establish because ownership and sovereignty are tied together by the key question: who issues property titles? Or, more importantly, whose authority to issue land titles will be recognized by other interested parties? Assuming that the Outer Space Treaty is with us to stay for the foreseeable future, what are other possibilities for establishing real property ownership rights on Mars?

In considering this question, I’ll first consult a fictional account of the settlement of Mars. Here’s how the story unfolds in Red Mars, the best fiction book that deals with the early stages of colonization of Mars. 

Red Mars makes an argument that other economic systems may exist. The book describes the growth of Martian colonies in a very humanly messy environment. Initially, all activities are supervised by UN’s “Office for Martian Affairs” (footnote: “UN Office for Outer Space Affairs” exists, but it is a small and inconsequential organization). No national sovereignty attaches, which makes sense because initial settlements are by design international with carefully managed balance of representatives from different countries. Next we see Mars a few short decades later, and commercial resource exploration is picking up. Big corporations from Earth are investing heavily in mining and terraforming. They certainly expect to own what they’re building, even though the legal status of their claims is unclear. At the same time corporations are lobbying the UN to change the law to their liking. It’s an interesting speculation and sometimes corporations can act on informal “understandings” with governments, especially when profit is huge. Miners acknowledge that shipping minerals, even most valuable, back to Earth is not going to be profitable until a Space Elevator is built on Mars, which would take decades. This uncertain investment is motivated by complete exhaustion of mineral resources on Earth, a condition that in the second book (“Green Mars”) of the Mars trilogy is called “full Earth”. Since there are no signs of full Earth in real life, it’s highly unlikely that corporations will heavily invest with no legal protection. 

The approach of mostly ignoring the governments is reminiscent of the world of Snowcrash, where the governments on Earth continue to exist, but their power has withered. This scenario is also unlikely to occur on Mars. Colonization of Mars requires huge investments, which means Earth needs to have a lot of free capital and an advanced economy capable of producing a surplus of high tech goods that are exported to Mars. If the Earthside economy is not doing well (for example, if we experience subsequent waves of deadly viruses), there will be little appetite for throwing trillions of dollars into space. Summer of 2020 has some matching indicators (success in space against a background of weak economy and civil unrest), but this combination can’t be sustainable. On the other hand, if things are going well, there would be no reason for the power of the US government to shrink.

Let’s now look at a couple of non-fictional precedents that relate to the issue of ownership of land on other planets. Even if a country that establishes a colony on Mars is tempted to do away with the restrictions of the Outer Space Treaty, they won’t be able to do so unilaterally. Their rivals who also have colonies or espouse ambitions to found their own colonies in the near future, will vigorously oppose any such claims. As of now, the two countries that are separating themselves from the rest of the field in the race to Mars are the US and China. Will we see a new Treaty of Tordesillas, dividing up the rights to the land of Mars, like Spain and Portugal divided up the New World in 1494? That seems highly unlikely as cozying up to China while giving a cold shoulder to their long-standing allies in Europe and Japan would be a stunning reversal of American foreign policy priorities.

Before humans go to Mars, they will go to the Moon. Preparing for missions to the Moon, NASA has recently proposed Artemis Accords – bilateral international agreements governing activities on the Moon. They seem to be a pretty straightforward elaboration of the principles of the Outer Space Treaty for the current realities. They are big on collaboration, interoperability and sharing of results, follow the general framework of OST and cite several of its articles. The advocacy article by Keith Cowing strongly emphasizes that Artemis Accords are continuing the lineage of Antarctic Treaty and Outer Space Treaty. His article mentions Antarctica 17 times!

One item of the accords that invited controversy is the proposal to establish Safety Zones around the areas where someone has landed on the Moon. However, it is quite common sense that once you land on the Moon and deploy mining equipment, you don’t want someone else to work in the same area without your permission. Industrial equipment is a hazard even on Earth! The idea of a safety zone is no different from fencing around mines and plants on Earth, which is not just uncontroversial, it is generally mandated by law. If you leave your industrial installation open to anyone, you’ll be liable when they’re injured on your site. 

So far no one has signed Artemis Accords. It’s interesting to see how things will develop. Will US allies join? Importantly, will China sign on? Will the US even want to sign an accord with China? If the US signs an Artemis Accord with China, it will make the Accord system a smashing success.

If Artemis Accords succeed, the Artemis model would become the presumptive blueprint to solving land ownership on Mars. The Artemis Accords themselves don’t address ownership in any way, but the value is in the approach of creating a web of bilateral agreements that could be extended as necessary. For example, Martian colonial powers could agree on a mutual recognition of homesteading claims. That is, someone who occupied and improved a plot of land on Mars would be able to claim an equivalent of ownership of it. Governments of key space faring nations would recognize this claim without asserting full national sovereignty over any part of Martian territory. One upside of eschewing national sovereignty claims is that it should make eventual transition to Martian independence easier.

Debates about land ownership on Mars are still a years to decades in the future, but Artemis Accords will happen in the next few months to years. Nominally about the Moon, these agreements should be of significant importance to the future exploration and colonization of Mars and beyond.

How Martian colony will retain its youth

A few decades from now, a nascent Martian colony will experience a serious challenge: retaining its young members. The situation will be in many details similar to the exposition of Ursula LeGuin’s The Dispossessed: Mars (in the book, Anarres) is a largely empty and harsh environment, sparsely populated, with austere life and very limited entertainment. One flight away lies Earth, with oceans, beaches, forests ranging from tropical to boreal, colorful butterflies, wild animals of all sizes, cities ancient and modern, concerts and sports contests on huge – and packed! – arenas, and of course, billions of human beings (oh, all the attractive potential partners). 

It will be very hard to explain to the naturally rebellious teens why they have to be tied to a tiny outpost on a barren planet. Even on Earth, keeping their youths is a losing battle for small and remote communities. As they grow up, new generations tend to move out of those communities into larger cities. In our case, parents will be rabid enthusiasts of Mars, but one can’t expect the children to blindly follow their ancestors. 

One possible way to deal with the problem is severe indoctrination and demonization of Earth. This is what Anarresti did in the Dispossessed. It worked for them because once the initial colony was set up, exchange of goods, information and people between the two worlds was limited to near zero by the mutual agreement of the two planets. This obviously can’t work on Mars. Anarres was settled by an ideologically committed group of people who rejected the culture of their planet of origin. Nothing like that exists on Earth. Beside cultural reasons, to cut itself off from Earth, Mars would need to become self-sustaining, which would take many decades if not centuries.

Another way is Areophany described in Kim Stanley Robinson’s Red Mars: an indigenous belief system kept in a tightly knit community. Problem is, Areophany only persisted because it was practiced by a reclusive subgroup of Martian colonists. We again see a motif of self-isolation, which can’t be healthy for the growing society. Even in the book, it was practiced by a tiny group and had little influence on the Martian culture at large.

The most likely solution lies in the opposite direction from Areophany or the Anarresti way: it is fast growth and ample immigration. Even the first Martian children might benefit from immigration: they can be guides to Mars and “big brothers/sisters” to the new immigrants, an exciting way to be important to the colony. Admittedly, this will be quite uneven because immigrants will come in waves once every 26-month synod. Later, as the colony grows to maybe 100,000 inhabitants, many of the “small town” problems will go away: dating pool will increase, labor specialization will create many job and business opportunities of every kind, local entertainment will be developed, different kinds of habitats will become available and so on. 

If the colony grows slowly, it will experience a long period of pain. Young people will bolt for Earth, making the growth even slower. But if the growth is fast, the critical period when it is too small for the youth will be shorter, so there won’t be enough time for a lot of people to leave and the growth will continue unimpeded. The faster the colony gets to the critical mass of people that feels like a city, the better, so the answer is to grow fast. Elon Musk’s stated goal of having a million person colony by 2050 is audacious to say the least, and, like much of Musk says, is more aspirational than real. Yet this aggressive approach is exactly what Martian colony will need to thrive.