How to search for the next Dalai Lama in 2030? using User embeddings?
TL;DR I am going to elaborate about a type of embedding of neural network(user embedding). And I am going to talk about how humanity could use this to our advantage. As well as, the human right of user embedding. All are just a speculation.
How a Dalai Lama is currently being selected?
Dalai Lama is about searching through a population for a unique character in a human. The traditional way is that a group of top monks go from home to home searching for a boy that would pick a recent Dalai Lama’s belonging among other artifacts. So, basically searching for a boy with the similar view, similar likings to previous Dalai Lama.
This is what I call inherit by characteristics or properties. There is the counterpart, inherit by blood, closeness. The selection of CEO, is also the same idea that main criteria are skills and competencies, like AMD and Google. While some companies(conglomerates), like Samsung, prefer to inherit by blood, closeness.
What is user embeddings?
User embedding is a high dimension vector/matrix contain numbers to use to represent/encapsulate who are you, what you like, what you value, based on your karma/behavior(what video/post you watch, how long,…) and your properties like age, location, income, education. Some info are directly ask you. Some info are indirectly infer from other sources (what app being installed in your phone[1], what devices are on your local network, what website you have been visit). e.g. what to recommend you in Netflix, YouTube, TikTok, Amazon, Facebook, Twitter or how search entry should rank in Google, what ads should you see. It is spooky true that sometime user embeddings are understanding you(and others) more than you understand yourself.
[1] used to happen in the past, using a special url, similar to this.
User embedding could be questioned using some math.
One day you might wonder if you are similar to dad or mom in terms of general tweeter habit. You could take user embedding and compare to each other by finding distance between dad’s embedding and your embedding against mom’s. Now. your know that you are closer to dad.
At this point, you might be able to relate to our topic which is actually searching for the next Dalai Lama using user embedding is within the realm of possibility! You could take the recent Dalai Lama YouTube embedding and see if any boy in Tibet have similar embedding as the recent Dalai Lama.
But, there is a catch. Comparing the whole embedding might not be a good idea of you focus on some attributes while exclude unrelated. How do we find user that more likely to helping others? Thankfully that many embedding also have a attribute of linear substructures.
This enable us to identify polarity and meaning of a subset of vector, see FastAI explanation for more details.
Also, if embedding from YouTube alone is not really who you are enough. You could combine embedding from Facebook, TikTok and as well as Twitter. It is also very possible to combine the NLP model(language embedding) to help in exploring a user embedding in a more descriptive manner. This is by no mean saying that it is easy or straightforward task.
Tweaking User embedding could make user closer to who we want to be.
We are likely to express our goal and direction by typing some query on Netflix, Google or YouTube. But, more of the time we just browse around suggestions create for us by our past. This might be good in some senses. but, not really good if we want to explore or dramatically change from who we are in the past. This is similar to what it is called echo chamber.
At some point, you might what to change the recommendation, There are several ways to do
- one way is to create new account and start from a clean slate embedding
- two is to keep typing new query and view those video, over a week or months
- three is to report YouTube what you don’t want to see more of this … because of …
- four is to ask your friend for a new playlist.
The existing solution is really stupid in some senses. but, I know that it could be better.
Imagine what it be like if we could use Elon Musk(or any of your model)’s user embedding to browse things. Imagine what it be like if we could explore our opinion e.g. republicans(or any of your view)’s user embedding to browse the other sides of ideas.
Imagine if we could just incrementally change only by a little day by day our existing embedding to get more and more closer to our desire characteristics(embedding).
That would be interesting things alternative to the existing ways, this is declarative express and intention of a human who might what to escape from a great gravity of embedding dark hole.
User embedding and self-actualization
On the individual level, knowing where you are going is really important. User embedding could help you to navigate or to keep track of where you are going. by analyzing the change of user embedding each week, you will able to see and evaluate where you are at and where you are going to toward.
The Right of User embedding
Although, User embeddings are the result of our whole community interaction with a platform. Currently, we are not able to get those embedding legally. It is not included in GDPR. This is up to a mercy of tech company/platform to give away.
Not to talk about right to directionally change our own embedding which is really futuristic and much much beyond the scoop of 10-year consideration.
This also could help startups and rivals to bootstrap user embedding, thus in order to keep competitive advantage instead of systematic win of community, reducing compute resources to compile/learn about personalization(causing global warming). I don’t see rationale for tech companies to share user embedding.
Dalai Lama? User embedding? and beyond
User embedding hold an immense power of personalization. On a surveillance government like China, it could easily use to understand and manipulate platforms on a greater scale. On a democracy government, it could be use to measure the result of election. Abuses are there. A way to protect is adopting Aeolia, extending its protection of personal data to user embedding as well.
Despite potential misuses, humanity will moving toward that our nature. At the bottom line, we, human, just want to have a peaceful time with his digital inventions. we want a good default so we could spend time with other important stuff.
Be transparent, Be expressive, And Be personalized.
