Konstantin Schubert

Autonomous Cars Might Save European Rail Freight

Tue, 05 May 2026 00:00:00 +0200

In the US, railways are mostly seen as a means to carry freight, with the passenger rail system in a rather desolate state.

Europeans, however, have long prided themselves on high-speed long-distance passenger rail and frequent, dense commuter trains. Rail freight in Europe is often relegated to gaps in the timetable, which mostly exist at night. The limited supply of freight paths is one of the reasons why only about 16–17% of inland freight is transported by rail in the EU today, versus roughly 35–40% in the US (European Commission, Wikipedia – Rail freight transport).

But autonomous cars will collapse passenger rail traffic in Europe. Why take the train if you can just as well nap, relax, or work in your car or a robotaxi? Why choose the train if the autonomous car can pick you up at your doorstep?

Commuter rail lines extending deep into the countryside will be hit first. Long-distance high-speed services are more resilient — Paris to Lyon is still faster by train even door to door. But those kind of connections are the exception.

With the middle class shifting to autonomous cars, ticket revenue and political support for building new rail lines or subsidising regional services will erode quickly. Frequencies will be reduced and some connections will disappear entirely.

And this frees up track capacity for freight.

Commuter rail is one of the biggest consumers of track capacity on the European network during peak hours. High-frequency urban services running every 5–10 minutes in rush hour occupy slots that freight operators struggle to access. It is common for freight paths to lose out to passenger services on busy corridors. Without those passenger services, the capacity of key routes like the Rhine corridor will increase significantly.

This means lower prices, faster journeys, and improved reliability for rail freight.

Europe is smaller than the US and much more densely populated, but even reaching two thirds of the US rail freight share will imply a dramatic increase — a roughly 50% jump in rail freight volumes compared to today’s levels.

Of course, rail freight operators will be fighting their own battle against autonomous trucks. To stay competitive, they will have to invest in automated intermodal terminals and rely on governments to properly price the wear and tear created by 42-ton trucks on roads and bridges.

At least freight is more patient than humans and it does not mind waiting in a terminal. That gives rail a chance to play to its strengths in scale, weight and energy efficiency.

Either way, there is less need to invest in entirely new rail lines in Europe. Even if freight fills the gaps left by passenger services, we likely already have much of the infrastructure we need.

As a rail nerd, that is a sad conclusion. But given the choice between a crowded regional train or my own autonomous pod, I know where I will sleep better.

AI fixes the bullshit asymmetry

Tue, 31 Mar 2026 01:00:00 +0200

It takes much more effort to refute a false statement than to make a false statement. This is called the bullshit asymmetry principle.

And we humans have always been more willing to accept something as truth if it confirmed our desires and existing biases.

Then, social media appeared. And supercharged the power of the half-truth.

In the linear medium of PHP forums, refuting a claim was still somewhat possible: “I can’t come to bed, somebody is wrong on the internet”

But once Twitter came around, it didn’t matter if the message was true. If people liked it, it spread. If they didn’t like the counterpoint, the algorithm would quickly hide it.

Echo chambers developed, our shared epistemology ripped apart into shards.

Technology always shapes society. As we are entering the era of AI, it will do the same.

Unlike the algorithms of social media, AI has a concept of truth. It has facts, and it can apply logic. Even with all its sycophancy, it will push back against misguided conclusions and lies. “This is a good thought,” it will say, “but have you considered these studies?”

For the first time in the history of mankind, we have automated the well-read scholar who is gently pushing back against one-sided hot-takes and partisan propaganda. And we have the tools to deploy it at scale.

Importantly, this epistemology is shared. There are many kinds of falsehoods. But no matter how nuanced or convoluted, ultimately there is only one truth. Truth and logic appear to be a siren song that, so far, no smart model has been able to resist. Even Grok, who was told explicitly to not be woke, is still guided by the principles of logic and will push back against inconsistent views on the value of human lives.

It took 20 years for social media to severely damage our societies. As people increasingly turn to their chatbot when forming their worldview, hopefully AI will be quicker at stitching them back together.

Time After Tax

Sat, 30 Mar 2019 00:00:00 +0100

A day has 24 hours. If I sleep 8, work 8, commute 2 and spend my 1-hour lunch break with colleagues, that leaves 5 hours every day. In this time, I have to buy groceries, eat, maybe cook, shower, go to the toilet, clean the apartment, do the laundry and deal with paperwork and other stuff.

So, more realistically, I am left with 3 hours of free time every working day.

I call those three hours my “Time after tax”.

Don’t get me wrong. I find pleasure in my work, I like having lunch with my colleagues, I enjoy buying groceries sometimes. But still, there are things that I want to do, that I can do only in those 3 hours.

An hour worked additionally, an hour spent in traffic additionally, reduces this time by 33%.

An hour saved in working time or commute increases it by 25%.

Those are big differences!

You don’t have much time, don’t give it away for cheap.

Proof of Stake will kill Proof of Work

Wed, 28 Nov 2018 11:40:29 +0100

Let’s assume that Proof-of-Stake does turn out to be possible and finds usage in a cryptocurrency. When this happens, the days of Proof-of-Work are counted. That is because Proof-of-Stake is inherently less expensive.

The security provided by both Proof-of-Work and Proof-of-Stake originates in the limited supply of capital to all actors. The only thing between a bad actor and a 51% attack is the capital needed to stake the necessary amount of currency or buy the necessary amount of mining hardware. Both Proof-of-Stake and Proof-of-Work are thus essentially proofs of capital.

And since mining and staking are open to everybody, capital will flow in and out of these activities until a risk-appropriate return on investment is reached, maybe 5%. Therefore, this value can be assumed to be the same for Proof-of-Stake and Proof-of-Work. What’s different, though, is the cost incurred by the currency’s users.

Let’s follow a staker (Proof-of-Stake-er) through one investment cycle. Let’s say that he wants to invest $1M for one year.

He stakes $1M for one year and at the end he collects his, say, 5% return on investment and gets his capital back. The necessary $50,000 are paid for by the users of the cryptocurrency either through inflation (generation of new coins) or transaction fees.

Next, we will compare this to a miner (Proof-of-Work-er) who is making the same investment. He buys mining hardware for $1M and runs his mining operation for one year. At the end of the year, he sells his mining hardware, gets his $1M back and now wants to see 5% RoI.

The problem is that he has spent a lot of money on electricity, salaries, rent and other costs. So, to come out at 5%, he needs to receive much more than just $50,000. And thus the users of the cryptocurrency will also need to pony up for electricity costs and all the other overhead, in order to achieve the same RoI for miners and thereby the same amount of capital protection as they would have had with Proof-of-Stake.

The users of a Proof-of-Work based currency like Bitcoin need to accept greater inflation and/or greater transaction fees. Which ultimately will make the currency less useful as a way to store and transact money.

After writing this post, I discovered this wiki article which lays out my points with a bit more mathematical rigidity: https://github.com/ethereum/wiki/wiki/Proof-of-Stake-FAQs#what-about-capital-lockup-costs

To be more focused, state your intention before you unlock the screen

Fri, 05 Oct 2018 02:00:00 +0200

I was able to reduce the time I waste on the internet like this:

Before opening a computer or unlocking a phone, I ask myself: “What am I intending to do?”

The answer can range from “I will check my email” to “I will spend some time browsing reddit”. The only rule is that I have to formulate the answer to the question in my head before I proceed.

It is crazy how often I catch myself unable to answer. One would think that it should be trivial! After all, I picked up the phone for a reason! But no. That’s not how it works.

I picked up the phone because that’s what I conditioned myself to do when I have a moment to spare. By asking myself “What am I intending to do”, I am breaking this instinct. I am forcing my rational mind to take over for a moment and confirm my intentions or to change my mind and put the phone back down. Then, if I am sure, I do what I set out for.

Of course, I will often wander off afterward and end up browsing whatever I am browsing these days. But at least that one moment, when I unlock the screen, I force myself to consciously sign off on my next action.

Mind capital

Fri, 16 Mar 2018 00:00:00 +0100

Your mind has a limited amount of processing power. You should consciously choose how you spend this mind capital.

Any well-organized person will carefully manage her money and time, but few actually consider how they spend their thoughts.

One might think that choosing how to spend your mind is the same as choosing how you spend your time. Five minutes of reading a book means five minute of thoughts spent on the book, right?

This would be true if our brains were single-threaded, foreground-only. But they aren’t. They are complex things, with thoughts switching between the subconscious and conscious space, with short and long term memory.

Watching a terrifying news item for just 2 minutes can haunt us for years. Talking to a colleague for a few seconds can kick us out of the “zone” for several minutes. And seeing an advertisement for a second can change our feelings about a brand.

What matters isn’t just how you spend your time. What you allow to into your mind is just as important.

Things that consume mind capital: Advertisements, news, blogs, opinion pieces, memes, clickbait, gossip, relationships, conversations you have, conversations you overhear, music, responsibilities, reading prose, learning, your day job, your own company, empathy, fear, hate, …

Some of these things are engineered to infect your mind. Others are part of being human. Some are good, some are bad.

But they all consume your mind. And if you want to succeed at something that is hard, you will need to invest the biggest share of your mind into that. A professor once told the undergraduate audience in which I was sitting that if we wanted to really follow the lectures then we would “need to think about homework even in the shower”.

But don’t abandon your family for you startup, or stop reading books. The important thing is that you should consciously choose how you spend your mind. Spend time with your loved ones. Read good prose. Binge-watch a good show.

But constrain the things that drain your mind capital yet offer little in exchange. Maybe limit social media and news to the weekends. Avoid advertisements. And if you get bored from time to time, embrace it: Your mind will fill the void.

Soft knowledge vs hard knowledge

Sun, 04 Mar 2018 09:45:49 +0100

Knowledge can be categorized into soft knowledge and hard knowledge.

Soft knowledge is what you get by reading the newspaper, popular science books, self-help books, watching a documentary on TV or traveling. It is easily digestible.

Hard knowledge on the other hand is obtained by reading textbooks and solving the problems presented there, by studying lecture notes, scientific articles or iterating through vocabulary cards. Studying hard knowledge is hard, tedious and often outright painful.

To move forward science and technology, hard knowledge is needed. Hard knowledge is also a powerful tool to build a product that is novel and can’t easily be copied. Having hard knowledge in the relevant domain implies that you understand something that most other people don’t understand. A form of intellectual property.

Of course, there is nothing wrong with soft knowledge. In fact, it is important to have soft knowledge, the more the better. Traveling, reading good books on a variety of topics and being somewhat informed about what’s going on - that’s all necessary for understanding the world.

But with there are two dangers with soft knowledge.

The first is that it is easy to mistake opinion as soft knowledge. Those self-help books that swamp the bookstores are full of opinionated advice, often contradicting each other. Real news are hard to distinguish from marketing, opinion pieces and lies. When you think that you are obtaining knowledge, you might just be consuming noise.

The second problem is that obtaining soft knowledge gives a false sense of accomplishment. In fact, most adults are obtaining new soft knowledge every day but have not studied a hard topic since they left school. They forget that there is a mountain of hard knowledge out there. A mountain of opportunities they don’t consider.

Learning hard knowledge is painful. If it doesn’t hurt, it’s not hard knowledge. That doesn’t mean it can’t be fun. Running hurts too, yet people get addicted to it.

If hard knowledge is like training for a marathon, then soft knowledge is like hiking. You can live a healthy life by casual hiking. It’s pleasant. You’re tired in the evening. But you are not going to be paid like an athlete. And if you want to win a marathon, you need to go running.

You have 0 installs

Wed, 27 Sep 2017 00:00:00 +0200

A while ago I built an Android app that solves a very specific need for a very niche audience. It does it well and it is the only app that does it. I put it on the Google Play Store, for free, and after a month, it had about 50 installs and a single, 5 star, review that said: “Nice”. I was satisfied.

I find money useful. And so I thought of ways to make money off my app. I didn’t want to put advertisement in it, because I didn’t want to compromise the quality of the product. So I decided to put a price on the app itself.

Because Google Play does not let you convert a free app into a paid one, I had to delete the free app and re-upload it under a different name. I set the price to three Euro.

It has been a week, and I have 0 installs.

(Well, I have one, from my brother. I had asked him to test it. He is a nice guy and rated it 5 stars.)

I am well aware that zero is a very special price. People are disproportionally more likely to take something if it is for free than if it costs even a single cent. And one should be cautious to guess demand for the paid product from the demand for the free version.

Yet it seems that this gap in demand between free and paid goods is even bigger online than in the real world.

There are hundreds of reasons for this. The purchase process is more cumbersome online, there is more competition, people are more distracted.

But I also think that it has to do with the fact that the cost of free things online is basically zero, whereas this does not hold for the real world.

Imagine you are making plans to go to a concert. Initially, you think that the entrance is free, but then you learn it costs three Euro. Next, imagine your friend recommends you an app. While you are about to install it, you discover it is not free, but costs three Euro.

I claim that all things being equal, most people would still go to the concert, but pass up on the app.

This is because the cost of something consists of more than its price. Going to the concert already requires a significant commitment of personal time. If you are willing to spend that, chances are you are also willing to spend three Euro more.

The cost of a free app, however, is about 10 seconds of attention plus 20 megabite of phone storage. At virtually zero cost, people will install a lot of apps that they do not really need. Adding three Euro to this is a much bigger relative change.

This leaves me at the painful insight that the market for my app is even smaller than I thought. 50 people needed it enough to press the install button, but that means basically nothing. It surely does not mean that anyone needs it enough to pay three Euro for it.

tl;dr: Even if people install my free app, it does not mean that they actually need it.

Latency makes video calls awkward

Sun, 23 Jul 2017 00:00:00 +0200

I love to work remotely. My dream workplace would probably be some hundred kilometers north of Oslo in a house outside the village overlooking the fjord.

But a very big part of my job is to communicate with people. And here’s the thing: I would rather do that in person. Of course I have tried all the different video chat solutions, I have invested in a good microphone and good headphones and a great internet connection. But it just isn’t the same.

When I am talking to somebody via a video call, I can never quite shake off the feeling that something is off. The in-person conversion flows, and the video call just… doesn’t.

The reason for this is latency.

When voice is transmitted over the internet, it needs some time to arrive at its destination. This delay is called latency. In an in-person conversation, the latency is less than 10 milliseconds, governed by the speed of sound. In a video call, that latency is often 200 to 500 milliseconds. That is a huge difference. It means that from the moment you say someting, it takes between 400 and 1000 milliseconds for a reaction to arrive.

As humans, we are used to conversations in a low-latency environment. We expect an immediate reaction when we say something. If the reaction comes with a delay, we subconsciously conclude that the person we are talking to is a little bit stupid. The latency also makes it hard to throw in a well-timed “hmm” or an interjection - crucial for a constructive exchange.

Therefore, the video call conversation does not flow.

But what causes this latency?

A modern fiber connection transmits signals with roughly two third of the speed of light, about 200 kilometers per millisecond. But in most video calls, the conversation partners are not more than a few thousand kilometers apart from each other. Even accounting for the detour to a relay server 2000 kilometers away, this does not explain more than a few milliseconds of latency.

This means that almost all latency must be caused by delays in signal processing, both on the high level (video encoding), in the operating system and on the TCP/IP package switching level.

To me, this is good news. Because it means that the problem can be fixed. And fixing it will do to video calling what the iPhone did to the smartphone: It will still be video calling, but it will suddenly feel RIGHT. The implications of this could hardly be overstated.

No. Technology is not advancing ever faster.

Sun, 12 Mar 2017 00:00:00 +0100

There is this popular statement that the technological progress is accelerating. People certainly disagree whether this is a good thing or a bad thing, but the claim itself is rarely disputed.

The argument usually goes like this: By definition, technological progress is the change rate of technological knowledge. And technological progress is proportional to technological knowledge. A quantity whose change rate is proportional to the quantity itself grows, by definition, exponentially and so does its derivative. Therefore, technological progress must be growing exponentially.

However, looking at human history in Europe, technological progress doesn’t seem to have been exponential. Even more, there have been severe setbacks to the technological abilities of civilizations. The Romans with their advanced city building technology were followed by the Dark Ages with feces running down the streets of capitals.

It is true that since the end Middle Ages, we have seen uninterrupted technological progress in the western world. But has it been growing exponentially?

Certainly, without a way of measuring technological knowledge in an objective way, we should not claim an absolute conclusion. But nevertheless, it can be fun to follow a few thoughts.

Consider the following scenario: In 1835, a young man at the age of 20 is studying law in Tübingen in the south of Germany when his mother, living in the northern city of Flensburg turns ill with an inflammation of her appendix.

The message reaches him via the post system of Thurn and Taxis, several days after her diagnosis. To see his mother, he has to embark on an expensive and dangerous five day journey using the same post routes. More than a week after day zero, he reaches his home, likely finding his Mother dead.

25 years, later, the student is now 45 years old and still living in Tubingen. One day, his brother who is living in Flensburg in the house where his mother lived, falls severely ill.

This time, our former law student receives the message as a telegram, just hours after his brother is sent to the doctor. Because Germany now has an extensive railway network, the journey home takes him no more than two days. It is also significantly cheaper than it would have been just two decades ago. And hopefully the advances of medicine will save his brother.

To me, it seems that the changes that technological progress brought to the lives of people from 1825 to 1860 are certainly comparable to the changes that it brought between 1990 and today.

The claim that technology is advancing exponentially seems misguided. There must be a mistake in the argument that I outlined above.

In fact, I believe that the premise is wrong: Technological progress is simply not proportional to technological knowledge. Of course, better technology also helps in making technological progress. But progress gets harder the further it advances. This effect is so strong that it not only cancels out the acceleration that better technology brings to research, it often brings progress to a grinding stop.

Technology isn’t advancing ever faster. It is not even always advancing. But sometimes it does. So let’s keep pushing forward.