Pontifications: The Giving Tree

Walter shares his weekly thoughts, fears, and philosophies (or some bullshit like that).

I just got back from a long run with a good colleague of mine; we work at the same company and he's a German data scientist. Why his nationality matters, I'm not sure, but his general knowledge about the world is underpinned by his education of breaking down systems, measuring something about each component within the system, and then stringing those measurements together—using various algorithms—to provide some insight into the performance of that system. If he says this is what he doesn't do, then he's wrong or I'm just not doing justice in my description of his work. We share a lot of what we're thinking about, or working on. More so, he's able to focus more on specific working-ons; whereas, I'm constantly thinking—pontificating, right? I enjoy learning about what he's working on due to my appreciation of operations research and systems engineering; however, my education on the subject is undergraduate and not applied, and I feel I'm missing something fundamental to want to do this work myself...

Anyway, a couple of weeks ago, when I had just started listening to an aforementioned book The Coming Wave* [of AI, essentially], I recall discussing with this Data Scientist the idea of using plant or other regenerative tissues (proteins, I think) as media to replace the latest GPU/TPU chipset architectures that allow us to process large amounts of data to create the AI systems we have today. The current silicon chips are built using laser lithography now, and we're able to get down to some 25nm in size of the transistor in order to, generally, maintain a Moore's Law rate of "double the transistors in a circuit every two years."

Supposedly. According to a new book I'm listening to, Human Compatible by Stuart Russell, Moore's Law is expected to reach some limitation in 2025 with the current technology.  As transistors approach the size of just a few atoms, issues like quantum tunneling and heat dissipation become significant challenges. Once we reach atom-size transistors, we will need to unlock more capacity using 3D stacking of the current silicon-based architecture, and improving the heat dissipation issue using material science. This all leads the ultimate unlocker of of Moore's Law in the 21st(?) century—qubits.

The definition of a qubit sounds like some black fucking magic, where instead of having a bit state limitation of just two states (0 or 1), we will have access to double the amount of bits currently, exponentially increasing the density of compute/storage at the atomic level. The biggest limitations here appear to be concepts of interference, similar to (and including) what the human race has had to deal with regarding electro-magnetic interference (EMI). What are some of these interferences?

Thermal noise, electromagnetic fields, or even collisions with air molecules all lead to a problem of "decoherence," when the qubit essentially loses its quantum ability to provide those additional states. Other noise factors reduce the reliability of the computational gates to get into various states, which increases the potential for computational error. Just the act of trying to measure a qubit could, supposedly, change the state of the qubit. In a way, this doesn't seem so foreign to me. In my engineering capstone project, way back in the day, I used a cadmium cell to measure light interference to detect an object passing between the cell and ambient light. When I was being graded on this, the performance of the system degraded due to the ambient light! The sensor I used to measure the object needed to be re-calibrated if I moved the apparatus to a different room. I also assume, but never really cared to test it out, that a fully dark room would render the cadmium cell—by which I was measuring an electrical signal based on how much light was hitting it—completely useless in this application.

So, with all that said, I've wondered why we cannot leverage trees as a computing medium for some of the same, or similar, promises that quantum computer espousers say they'll get with reliable creation and measurement of qubits. Trees are extremely regenerative, so could they build the quantum computers for us, while drawing down carbon and effectively not require all the energy that is currently needed to keep data centers running? Though trees likely have some environmental interference which may negatively impact human's computational exploitation of them. Trees have molecular structures that seem to be stacked (I think?), allowing for more capacity for compute/storage. But are we able to measure the signals in the barks of trees? What manufacturing is needed to build these measurement tools? Does the make-up of the air cause defects in the bark's atomic layers? Would the electromagnetic interference of the earth and sun interfere with our ability to measure the states of the bits in the atomic levels of the tree bark? What if a fungus attacks the tree? Would we have enough redundancy to keep our modern compute/storage reliable for the public and corporate consumers to apply their capital to?

What I've described in the last paragraph is likely not at all not a practical realization of what is to come in regards to the future of technology. At least not until humanity reaches a precipice of exponential self-destruction. The thing is, what I describe is already happening, we just don't pay any attention to it in the grand scale of capitalism, and prefer to engineer our own simulation of what nature already does.

These thoughts continued to race through my brain, as Matthias and I rode the subway back from our run to our neighborhood, standing next to a man who is on his way to hussle-off his massive bags of collected NYC recycled cans.


*I realized I typo'd the title of the book last week by forgetting "Coming" in the title. Oops.