The holes in our knowledge are like bugs in a weather-sealed lens

A saw a story about bugs living inside a weather-sealed lens. That’s a place where there should be no bug of any kind, and yet they were there. What else do we miss? How often do we see profoundly rare outliers? More reading here, here, here, here, here, and here.

Visualizations created by life itself

Sometimes we visualize to remember, other times as a metaphor for some emotional process. The best visualizations are not created by us, but by life itself. We notice something extraordinary and unexpected. It catches our eye, but cannot find verbal expression. Suddenly one image is better than a thousand words.

At the same time, the limitations of our understanding become profoundly visible. We might fail to understand why a particular image captured our attention, or how particular situations could ever happen. Maybe we feel something profoundly non-verbal about life itself.

The magic of photography

I love photography. Good photographers tell stories with their images. These stories can be seen only in certain light from a very special location. It makes sense to pause for a while and think not about the location, the composition, or technical issues, but about the story itself. For example, what is the story of a bug in a water-sealed lens?

First, we do not see the bug on images by this camera except for some very specific settings. We might get some shadow. But when looking at the camera from outside the bug is very clear. It should not be there, by any reason of construction. Maybe there was some freaky crack or maybe the photographer created this story as a mystification. After all, photographers love to mistify combining multiple images from different times and camera settings,

Yet the biggest story for me: how blind we are to things that are not supposed to be there…

Hidden layers

Just like the lens is built from multiple optical elements, the neural networks in our brains are built from multiple layers.  We do not really know what is inside these layers, called “hidden layers” for a reason.  There might be huge bugs hiding within. This is an issue with different engineering solutions. In computer vision, generative networks can be used to fool detection networks, and together they learn to become better.

We need to experiment in multiple iterations until the neural networks in our brain converge to a meaningful solution. Sometimes it is enough to run mental experiments: at least for people as smart as Albert Einstein. Einstein’s IQ was above 180, my IQ is slightly above 140, so I often need computers to visualize my own mental processes. An average person with an IQ of 100 probably needs to actually perform the iterations physically in his garage.

Different people use very different tools to achieve basically the same thing: iterate. There is a science behind iterations.

Iterations

I will not go into the engineering of AI training. Instead, I will quote a somewhat different approach of the neuroscience:

• exploration, or “iterating around problem and solution while elaborating them concurrently”
• concretization, or “revisiting elements of the design while increasing their levels of definition, ensuring consistency”
• convergence, or “point-by-point improvement of parameters and details, at a fixed level of definition” 
• refinement, or “adjusting, improving, perfecting once primary objectives are met” 
• incremental completion, or “repeating a task on different information to incrementally arrive at a goal”

We iterate differently, based on the challenge we are facing and our level of progress. The bug will be visible only in some cases, in some iterations, and in some processes of experimentation. If the outliers are rare enough we will probably miss them. Sometimes we feel no visible progress and do not understand the results until much later.

The curse of resolution

My father is a sort of a hero, the quiet sort nobody knows about. His biggest achievement was during his internship when he saved a huge factory from an explosion. As an intern and a chemical engineer, he was asked to monitor some complex processes and measure the key indicators every 30 minutes.  My father has ADHD, so he measured every 30 sec instead. To his amazement, he found some peak response nobody knew about. This response would not be seen in 30 min measurement. His discovery saved a factory. For comparison, if he worked in Chernobyl there would be no catastrophe to discuss.

Why was he given the task to measure the result every 30 min? There were so many parameters that if all of them would be measured all the time, no processing power could handle this. As nobody knew about singularity points, the reaction was measured according to the sampling theory (Nyquist frequency). That was a reasonable response. If we have 5 dimensions, doubling the resolution in all the dimensions would result in x32 measurements. Taking x10 resolution would lead to x10000 measurements. For example, I work with image and video processing, there are often hundreds of dimensions. It is very easy to miss critical information.

The most critical decision my father made was not to hide and ignore his measurements but to confront his superiors and look for the root cause. The root cause in his case was some new quantum level that appeared due to a strong interaction between the ingredients.

Cannot be right

I remember a story I saw on TV, that claimed Einstein formulated correctly the general theory of relativity three years before he published it. He simply could not believe his result and used the next three years to eliminate all other possibilities. Possibly if Albert Einstein would still be working with his wife Mileva, we would see this and other results much earlier.

Power couples tend to work much better than talented individuals. In science, one member of the power couple usually comes up with preposterous suggestions, while the other systematically proves or disproves them. If we are forced to judge our own ideas, we might lose confidence and try to exhaust all other ideas instead. For most of us, this is an impossible task.

Our common sense does not handle the holes in knowledge. Otherwise, magicians would not trick us so easily. Instead, it signals that new results contradict everything it used so far. We feel unpleasant dissonance and usually try to avoid the subject. Or seek humor…

The punchline

Discovering holes in our knowledge may be uncomfortable. So our body uses special hormones to feel good. These hormones are produced when we get a punchline of a joke, and they are very similar to what we get when we experience awe.

When we experience awe, we feel something so much bigger than our everyday processing, that we need to cope with this revelation. No matter how much we read about it, our visual cortex might not be used to deal with this input. We are overwhelmed. It’s a sort of non-verbal punchline.

Contagious belief

When we truly believe something, our beliefs become contagious. People who love us, trust us, work with us will instinctively share our beliefs. These friends will pass the belief to their friends. Believing something without proof is profoundly dangerous. The more of us get invisible bugs in the way we view the reality, the harder it becomes to detect them,

The most rigorous science of them all, physics, is facing possibly the biggest existential crisis in its history. We know that most of the universe is made of dark matter and dark energy, yet we have no idea what they are. The nature of physics on the event horizon of the black holes is required to deal with singularities, yet even Hawking’s equations appear to be wrong. To explain the reality in all its complexity we need to introduce more dimensions, yet we cannot observe them.

Religion was a contagious belief that made people feel better, accept pain and death, and be better with each other. Science claims to check everything, yet the belief in that is contagious. We prefer not to check our limitations.

Nobody knows everything

Quite possibly Goethe was the last person to know “everything”. His IQ was definitely high enough. For the last 200 years, there was simply too much to know for a human. Possibly in 2060AD a computer will know everything for the first time, and this will be a singularity.

Our partial knowledge is the source of wonder, which is a sublime human feeling. I will address science fiction by Robert Sheckley : Computers are infallible. When they make a mistake, it is in fact correct, as the design included the probability of the error, and it performed as designed.

Probably it is best to meet our limitations with a sense of humor and awe… We perform as designed…