The Braintech Developer Stack, Time is Entropy and Machine Teachers
The "Future is not what it used to be" is not a newsletter. It's a movement. A way of life. This is why we're launching our clothing line. We're starting with clothes, but our vision goes further. Next we'll launch a drug subscription box (Nicolas will be the curator) so that microdosing and mushroom hallucination have no secret for you.
Ok, no, we were just kidding. We're not going to launch a clothing line nor becoming "Birchbox style" drug dealers. We're too lazy for that.
The Braintech Developer Stack
The past couple of years has seen the rise of user friendly and affordable hardware and software that enable developers to create their Braintech app. It’s only the beginning, but a real Braintech stack is emerging.
Let’s illustrate this trend by looking at two interesting products.
Hardware: Emotiv is an EEG headset that you can buy for 800$ and which can be used for professional applications. What’s interesting is that they also offer a SDK, as well as several APIs, that enable developers to interact with the brain in several ways: from accessing raw data coming from the headset to API “shortcuts” that abstract functions such as mental commands or emotion recognition (available “out of the box”). All of that through a subscription model (yes a good old SaaS model for Braintech).
Software: the Virtual Brain is an open source software developed by neuroscientists that simulates the human brain on your computer. This project is an attempt at emulating how the brain works to let anyone run their own simulation by playing with a variety of parameters and visualize the results.
Why does it matter ?
Similar to what is happening in the web developer tools space we expect more specialized products to emerge (e.g: an API focused on “mind control” only, specialised in gaming, in emotion monitoring etc.) and also more platforms.
We spoke about EEG in this newsletter and a previous one, but other technologies to interact with the brain (MRI, fMRI, brain electrodes) will also get more accessible in the years to come.
We're still at the beginning in terms of what's really possible and most of these tools have huge limitations (here's a great post on the current limitations of EEG headsets).
Our must reads this month:
[Data Point] Researchers have made a decisive step towards being able to simulate brain-scale networks on future supercomputers of the exascale class:
- [Article] "The image below show what a human saw and then three different ways an AI interpreted fMRI scans from a person viewing that image." Source.
Space/Time & Reality
Does time emerge from entropy?
What do most people think characterize “time”? First that time is universal, there’s only one time. We share the same “clock” across our whole universe. Second that time runs only one way: from past to present and future, we don’t go back in time. However in the past 200 years we’ve discovered that what I’ve just described is not true at a fundamental level.
First, the law of physics showed that time is a local variable (and is bound to space hence Space/Time). It doesn’t run evenly everywhere. It depends on where you are and at what speed you travel. We don’t see the difference at our human scale, but time is not running at the same pace for someone climbing the Everest and for someone enjoying a Sex on the Beach on the French Riviera. There is no universal clock shared across the universe, and the notion of “a single present” at that scale doesn’t even make sense. Second, in the fundamental equations of physics the parameter time (“t”), as I’ve described, is different. In these equations nothing forces the parameter “t” to go in only one direction (from past to future). It can go in any direction, it’s not the time as we know it.
At the quantum level the "t" parameter can go both way. But what we perceive at the macro level is a one way direction.
In reality the notion of time that we perceive could be an emerging phenomenon specific to us, created by the brain of living creatures, and not a fundamental phenomenon
But how and from where does time emerge in our brain? A possible explanation, speculative and not proven, is that it emerges from entropy. Entropy is a variable that quantifies complexity/disorder. According to the second law of thermodynamic entropy always increases in a closed system (such as our universe) until it reaches its maximum complexity.
This is my next tattoo. The second law of thermodynamics is one of the few equations in physics where the arrow of time actually matters.
And this process is irreversible. This is why, from our human perspective, the world is only getting more complex, more chaotic, it’s never going backward. The egg which you break to do an omelette will never go back in its eggshell. It’s a one direction process. And it’s where our notion of time could emerge and why we think it goes only one way: because from our human perspective entropy only increases and our brain interprets it as the arrow of time (a succession of ordered events).
Like a “chair” doesn’t exist at a fundamental level, but “emerges” from a particular configuration of atoms, what we think is time doesn’t exist at a fundamental level but could emerge from entropy.
Why is it important?
Because it could the next “Copernican Revolution”. If you go outside and look at the sky, the sun, stars and other planets seem to revolve around us. But it’s only an illusion due to our perspective. And it could be the same with time, it’s only an illusion due to our human perspective.
Our must reads this month:
- [Article] “A Debate Over the Physics of Time”: a great wrap up of the different theories trying to explain what time is.
- [Article] "Entropy explained, with sheep" a great interactive post to better understand entropy.
- [Book] "The Order of Time" the new must read book by Carlo Rovelli that inspired our mini deep dive.
Surveillance & Privacy
The Data as Labor vs Data as Capital or why we may soon receive a payslip with the machine teacher job title.
At the heart of the privacy/surveillance subject lies personal data. What is this data? Who owns this data? How much value does it have overall and marginally? An interesting debate is rising between two ways to look at it: is personal data a capital owned by the user who could then resell it, or is creating personal data a labor and users should be paid a salary for this job.
Personal data is not just the content you create directly. This is just the tip of the iceberg. Using any digital tool leaves a trail of data that we create actively (content, likes, share…) and passively (location, click/did not click, time spent, browser used...). Altogether this create our personal data stock, or trail would be a more accurate word. We all have one as soon as we use a digital tool whether we are logged in or not. Personal data is valuable individually to describe a user but also as an aggregate to detect patterns. A lot of online businesses are based on leveraging personal data to increase usage of a product and its monetisation. So should users be paid for the data they produce? If yes, how and how much?
There are 2 different way to look at this.
Data as Labor
We can consider that all users are producing data and they should be paid a wage for this task. The idea is that we are all employees of machine learning algorithm. The defenders of this idea consider it fairer because everybody gets paid not just the people proactively looking to get paid. Some people like Jaron Lanier in this paper link this idea to the Universal Basic Income.
Data as capital
Because we are producer of the personal data we then can be the owner of this data and thus should be allowed to sell it. Advocate of this idea are more the free market people. They consider everybody is responsible and can decide what to sell.
Microsoft research published a great paper on that topic in decembre 2017 and recently a “free market” think tank released a good report defending the data as property model. (funny enough this sparked a classical left vs right intellectual debate: marxist defending the labor angle and free marketer defending the capital :-) it seems that some controversies never get old.
How do you value and price personal data?
Whether you choose the data as labor or the data as capital argument you need to find a way to price personal data and this is getting tricky. Do you pay for the whole data set or just unique data point, for the dynamic evolution or for a large snapshot ... Researchers are finding way to point out the sub set of data that is meaningful to a learning algorithm ie reducing its error rate .That would allow a pure dynamic value based pricing but this may be difficult to apply. There are of course some blockchain / token based project around those ideas.
Why it’s important?
Personal data is already a business asset that can be traded and it will become more and more tradable. New business and organization will emerge from this trend. We could see union like organization where digital platform users will regroup to collectively bargain the value of their work. We may also see some tools acting as personal agents that will help users manage their data and sell them, brokers and marketplaces will emerge.
The taxation of digital economy is also a major topic that is still looking for better solutions. Considering data as labor or capital creates new ways for governments to collect taxes from digital platforms.
Our must reads this month:
[Data point] Brave recently hit the 2 million monthly average user mark, DuckDuckGo is seeing its highest traffic ever, with more than 24 million direct search queries per day (image below: average daily queries on DuckDuckGo). Source.
[Article] Good article in the Guardian on the personal data collected by telco operator. (The business model of Telco is changing to Surveillance Capitalism).
[Article] Data Poisoning attack on Deep Learning model. This may be the hacking of tomorrow. insert fake data in a dataset to fool the algorithm.