Typing at the Speed of Thought

I’ve recently come to the conclusion that moving individual fingers to press plastic squares is a fundamentally inefficient way to interface with a computer. It’s a tedious, mechanical process involving far too much physical exertion. Our bodies are already powered by electricity. Computers are too. It seems only logical to cut out the middleman.

So I'm going to try to build it.

The Grand, Deceptively Simple Plan

The idea is to strap some electrodes to my forearms and listen in on the faint electrical chatter my muscles produce when I think about typing. These signals, known as electromyography or EMG, are the literal precursor to movement. The plan is to capture this raw motor intent, feed it into a machine learning model, and have it gracefully translate my thoughts into perfectly formed sentences. Or, failing that, at least a reliable left-click.

I have never really worked with hardware before. It seems straightforward. You connect the pin that says "OUTPUT" to a pin that says "INPUT," and the magic happens. How hard can it be?

I am, however, confident this is going to be a fun weekend project.

A Brief, Humbling Google Search

Naturally, after coming up with this revolutionary concept, I did a quick search to see if anyone else had ever been so brilliant. It turns out, yes. A Canadian company called Thalmic Labs not only had the same idea back in 2012, they actually built a sleek, futuristic device called the Myo armband that did exactly this.

They successfully read forearm EMG signals to control computers, VR, and all sorts of other things. Eventually, they pivoted to smart glasses under the name North, and were then promptly acquired by Google in 2020.

So, on the one hand, this proves my idea isn't completely insane. On the other hand, a well-funded team of university graduates already built it, sold it, and moved on to bigger things that ended up being bought by Google. This is, I'm sure, a very encouraging sign for my $8.71 version held together with tape.

A Short List of Trivial Obstacles

I foresee a few minor hurdles:

Bad Signal: My body's electrical signals might be less of a clear broadcast and more of a staticky mess, drowned out by the 60Hz hum of every electronic device in a three-block radius.
Not Enough Training Data: The model will need to learn my unique muscle patterns. This likely involves me sitting very still and thinking about the letter "A" for an uncomfortably long time.
Low Accuracy: The difference between "thinking about typing 'hello'" and "thinking about an itchy elbow" might be tragically subtle, leading to less-than-ideal performance.
Errant Purchases: Accidentally ordering a pallet of dog food while swatting away a fly is a non-zero possibility.

The Tools of the Trade

To tackle this weekend project, I've procured a few key components. The star of the show is a tiny chip called the AD8232. It's an integrated signal conditioning module, which is a fancy way of saying it takes the faint, desperate whispers from my muscles, amplifies them, and filters out all the noise so they're clean and presentable.

This chip will be connected to an ESP32, a microcontroller I had lying around from a previous, equally well-thought-out project. The ESP32's job is to act as a middle manager: it will take the cleaned-up signal from the AD8232, convert it into a number, and dutifully forward it to my laptop for the real thinking to happen.

Here's the official shopping list to start on the POC:

1 x ESP32 Development Board (already on hand, cost: $0)
1 x AEAK AD8232 Physiological Measurement Module (cost: $4.04)
10 x TENS Unit Replacement Pads (cost: $0.99)
10 x Electrode Lead Wires Adapter Cables (cost: $3.68)

The components are on a slow boat from somewhere. In the next post, we’ll see if electricity plus optimism can actually be compiled.

What could possibly go wrong?