Functional Noise

What is functional noise? I did not attempt to explain the concept yet.

Let's start with the words. I know you love words.

Functional: "of or having a special activity, purpose, or task."

Noise: "random fluctuations that obscure or do not contain meaningful data or other information."

Hah, it seems we have a contradiction! What a silly blog name. How could we possibly give purpose to something meaningless?

Corrupted Neural Networks

Many years ago neural networks began to accurately predict written characters from images. Obviously people tried to make fun of the neural networks, by inventing corrupted images that humans could easily predict, but the neural networks couldn't. This proved that the neural networks were not so smart after all. See for example ways to corrupt the MNIST dataset.

Turns out this adversarial game can be used to improve the neural networks. Noise injection is often used in machine learning to avoid overfitting and generalize the predictions on unseen data.

One way is to add random variations to the input images. Intentionally distorting them. This leads to the creation of more training data and makes the neural network predictions more robust to those variations, and often also to other unseen variations. By augmenting input data with functional noise, you can improve the artificial intelligence.

But researchers looked for many other ways. One example was the dropout method, dropping out nodes in the network randomly during training. It's kind of distorting the data as it passes through the neural network. So functional noise can also be added inside the neural network.

Finally you can set up an entire Generative-Adversarial-Network (GAN), where one neural network tries to constantly fool the another neural network by generating less realistic data. But the other network will learn to see through the deceipt. Gamified functional noise to generalize the learning.

We're still playing these games with the AI up to this day.

Weird Dreams

Dreams seem like a mystery to neuroscientists. What is their purpose? They are clearly weird hallucinations, often not describing our waking reality accurately. Yet we spend a lot of time in them, up to many hours every day. Silly brains. Probably some evolutionary leftover, like the appendix.

However we can be creative in our scientific dream interpretations. I personally love this new Theory of the Overfitted Brain. It states that the weirdness of dreams is their intended purpose. Similar to the corrupted input data for neural networks, we hallucinate distorted versions of our reality, all to better generalize our predictions about the world and life.

Dreams are functional noise.

If you have boring dreams, you risk overfitting your current life experience. When life suddenly changes in an unexpected way, you feel unprepared and unpleasantly surprised, causing a great deal of suffering. You want to have exciting, weird, unexpected dreams. All to increase the flexibility of your mind.

How can you make your dreams weirder? I think you can start by slowly adding functional noise to your (inner) life. Read weirder books. Watch weirder movies. Think about weirder things. Talk to weird friends and colleagues. As your thoughts become just a little weirder, your dreams will amplify that functional noise, causing more interesting hallucinations. Just start small and go as far as you feel comfortable.

There's maybe a risk that you go too far? But fortunately you'll be testing yourself in the waking world every day. Does life still feel okay? As long as you can function well you are not at risk. And you can look for life improvements. Do small unexpected events feel less bad? The point is to reduce suffering and be able to deal with whatever life throws at you.

You don't have to start acting weird. People like you to have a predictable identity. You can keep that intact if you want. We're only preparing our minds for weirder eventualities.

Intentional obscurity

So functional noise is great for artificial and biological minds. Can we think of other purposes? I know at least one more. Remember: noise consists of random fluctuations that obscure information. You can use noise to hide signals, as a kind of cloaking device. Sometimes you want to intentionally make it harder to find something. I have already discussed this purpose in my beacon blog post.

Now you know what it is.

Functional Noise To Generalize Your Mind.