Thursday, February 6, 2020

What to remember

Prioritize Invariance


Only remember things that are true now, and that will always be true. An example of a statement that will not always be true:

Donald Trump is president.

An example of a statement that will always be true:

Donald Trump was elected president of the USA in 2016, and as of the 5th of February 2020, he held the position. 

So a statement that is not invariant can be made invariant by specifying a context to it. Why prioritize invariance? Simply put: so that you can trust your own memory. If you do not couple a conclusion with its assumptions in memory, you may miss when the assumptions change, and your knowledge is outdated. If you go to the movies and see The Wolf of Wall Street and come out unhappy, something like this can easily become a cached decision:

I do not like The Wolf of Wall Street. I will not agree to rewatch it.

Consider specifying this as an observation (about yourself)  instead:

When I came out of the cinema after seeing The Wolf of Wall Street, I remember feeling that the movie was too long. Also, the main character seemed unbelievable in the final 45 minutes.  

Suppose now that you are offered by a friend to rewatch The Wolf of Wall street some years later. The friend has a re-edit of the film with a shorter ending, that is meant to be more true to what really happened. With the first, emotion-based memory, you would not notice that this re-edit addresses exactly the issues you had with the movie. Furthermore, remembering things this way makes your opinions more interesting to others, since they are less subjective. Remembering reasons rather than personal general impressions makes it possible for people who hear your opinion to determine whether it applies to them as well. This is an example where the consequences are of little significance of course, but suppose the same thinking is applied to for example one's impression of a person, a new piece of technology, a travel destination, a political party, or a potential employer! One could easily miss things becoming desirable (or undesirable), despite having seen enough evidence to change one's opinion.

Years


When opening a book, the first thing one should do is to check the year of publication. Why? If you know the year of publication, you know the information that was available to the author when writing the book. How do you know what information was available in such and such a year? By knowing the year of publication of other books! So by remembering a linear amount of information (i.e. one integer per book), it becomes possible to consider a quadratic number of connections (i.e. between any pair of books).

The same argument can be applied to people (knowing their birth and death year) and movies (year of release). I have found it less useful to remember years for music. Perhaps it is because I do not know enough music, so that I have not been able to see the effects of the quadratic number of connections. Another reason may be that since I am not trained in music, I do not realize that there is an interesting connection between two songs. In that case, becoming more learned in music would open up a world of interesting observations and conversation topics for me. A third possibility is that the people who do claim that there are interesting connections between different pop music songs for example, are just kind of faking it. But even with this suspicion, it might be worth it to practice a better ear for music.

Another field where I do not find it particularly useful to remember years, is in technology. There are some key inventions such as the transistor in 1947, that one should know. A reason why it is not so useful to remember certain years is that the important inventions were often created over some time. So an approximate year or decade is fine. Big inventions also take time to reach their potential. An example is the transistor radio that was released in 1954, seven years after the transistor was invented. The first neural network was demonstrated by Frank Rosenblatt in 1958. But of course, the neural network was not very important until 2012. Another reason why years are not so important for inventions is that it is more specific to remember things about the invention itself, that explain why they were able to replace the previous way of doing things. The problem with the 1958 neural network was that it lacked multilayeredness and backpropagation. This was fixed around 1978. The next missing features to greatness for neural networks was perceptive fields and layerwise training, which was introduced in the early 1990s and 2006 respectively. The final touch was data augmentation, and increasing the total amount of data and training by a lot, which was done in 2012 after which the big break finally came.

Location


When someone tells me about a new development, one of the first things I ask is: who is doing this? Which company or university? This is often ignored in a news report of the new finding. The news segment may say "Italian researchers demonstrate such-and-such". Knowing that they are Italian is not enough: it is just as well to remember the university and/ or city. Why do I insist on this? 

One reason is it is more useful for networking. If you do read a good article from the University of Milan for example, and later meet someone in your field who studied in Milan, chances are quite good that they have met the authors of the article! This helps you ask more informed questions. It also shows that you have bothered to remember something related to them. In the best case they want to pay back in kind by asking about your research. 

However, I think the most important reason for remembering things by location, has to do with the concept of a mind palace. The idea of a mind palace is that you imagine this place with lots of different objects and textures and levels and whatnot, and you use the image of this place to connect different memories to each object. To me, using a mind palace seems so bogus. It seems totally arbitrary. Better then to connect things to an image that is not arbitrary at all: the map of the world! The world map is something that should be remembered anyway, so it costs very little extra. It also works well on different resolutions: one will know about more things in one's home country, but one also has a better idea of the geography of one's home country. 

Using the historical timeline as a third dimension to this world map-mind palace does also work very well. I don't see how it is possible to organize memories without having a good visualization of the historical timeline.  

No comments:

Post a Comment