Topics
Creativity
Film & Literature
Music
Activities
Games
Food & Drink
Pets
Causes
Sports
16 Types
Enneagram
Astrology

Konstantin

1 month ago

INFJ
Leo

2w1

2

1

Conversations for Computer Scientists

Imagine a tree where every branch is a conversation. Every branch in this tree splits into more branches — more conversations. Some branches extend far, leading into more and more conversations — be it deeper and deeper into one. Some branches are connected to lots and lots of branches — many topics to talk about, but none that go deep. I guess am optimal conversation would be a tree that has many branches that extend very far, each branch having lots of branches and each one leading into more. We'll call this optimal tree a fruitful conversation. Fruitful conversations aren't good or bad, they imply that a conversation could be engaging, lots of deep things to talk about. That criteria can cover debates or a great conversation with a partner: lots of things to talk about, and with each one of them we can go on and on. For every person you talk to, you plant a tree. The tree represents your conversation, and it grows in the same manner you converse, navigate through a conversation. It starts out small, from a single branch with which one of you chooses to seed the rest of the tree. The "game" is to find the branches, conversations, that lead to more and more branches: either your conversation leads to more interesting topics to talk about, or your conversation is an interesting topic that can go on and on. Each conversation is interesting or not, and too much uninteresting conversation can starve the tree, just like how if a tree grows into the dark, it can end up dying, which is why it tends to grow towards the sun. Briefly, in computer science terms: a conversation-tree is a C = (V, E), where each vertex is a conversation, a 2-tuple containing a (your response, her response), and each edge is a relation that maps one conversation to many. Each edge has a corresponding 2-tupled weight with each element bring a real numbered value from [-1,1] that represents how the other received the response: (-0.8, 0.8) — they did not like my response in the same intensity that I liked there's. A path is a set of edges, and it's net sum of its weights represents how the conversation has progressed: after a thread of conversations, our net weight is (0.4, 0.2). Looks like we both enjoyed it overall despite some hit or misses on the way. Each agent in the conversation has a real numbered threshold [-1,1], that determines when a conversation will stop. If after a number of conversation your statement received a mediocre, overall reception that's lower than your threshold of what you can take, the next conversation will be a null conversation, a 2-tuple that is empty: ({},{}). If you understand anything up to this point: how would you traverse the tree, so that the net weight of each path is the maximum and uses the least amount of branches. If such an algorithm already exists, what is it's name?

favorite

2

11

Ty

1 month ago

ISFJ
Libra

Dating, it's called dating :P Jokes aside, I think my only question is why weight the edges from -1 to 1? This is my rookie computer science brain asking. But I would do a depth first search, choosing the highest minimum weighted branch (i.e. choose (0.8,0.7) over (0.9,0.5)).

favorite

1

9

Reply

Konstantin

1 month ago

INFJ
Leo

2w1

2

1

The calculations are much easier to do and can be scaled homogenously. The weights could be any arbitrary real number, but there would be no the sense of a maximum or minimum number, so the values of the net weights can easily get unwieldy after a number of iterations. It's much easier for a person to rate something from [-1,1] on how much they liked it or not then from (-inf, inf). They can also be readily put into terms of probabilities. So your strategy would be to engage in the deepest conversations first?

favorite

0

1

Reply

Ty

1 month ago

ISFJ
Libra

If there's a cap of 1 for each weighted edge, then the max value of any path to the terminal node is the number of edges traversed to get to that node. This is because I'm going under the assumption that we're aiming for the best first impression to get a second conversation tree.

favorite

1

1

Reply

Konstantin

1 month ago

INFJ
Leo

2w1

2

1

Yes! That's right, that would be one of the trivial cases. Cool of you to have spot that. Another reason is that one ought to have the same metric for each weight. [-1,1] should be a sufficient amount of numbers to represent trees that are of any arbitrary real number due to one of the property of the real numbers: (-1,1) and the set of reals forms a bijection. What do you think about an algorithm that emphasizes going down vertices with highest degree: choosing conversations with the most amount of topics to get deeper into?

favorite

0

2

Reply

Ty

1 month ago

ISFJ
Libra

Well, my thoughts is that most trees will have more nodes off deeper topics. Once you get you past the first few exchanges, you should have options if the topic was fruitful. Otherwise, if you didn't, then you should have a terminal node and have to go back up the tree. I think running both algorithms is promising.

favorite

0

1

Reply

AnhQuan

1 month ago

INTP
Sagittarius

You will likely find an algorithm that explains your phenomenon in some genetics algorithms that prioritize survival. Training Super Mario with machine learning comes to mind. Though dating is kind of a twofold problem. No one has the time to enumerate all the possibilities for conversation and you don't know what variables your answers may change because the system is sort of akin to a live GAN. Of course you could end up in a groundhog day scenario (controlling the seed), but thats unrealistic in reality, which is why we develop social heuristics. I've sort of tangented outside the original topic.

favorite

1

0

Reply

AnhQuan

1 month ago

INTP
Sagittarius

The main question is whether this is a thought experiment or we are trying articulate a practical application. We can theorize design and veer off into what constitutes heuristics to prune trees are discuss chatbot design. Though they aren't mutual exclusive.

favorite

1

2

Reply

Ty

1 month ago

ISFJ
Libra

I'm just trying to learn a few things by engaging in conversation :)

favorite

1

0

Reply

Konstantin

1 month ago

INFJ
Leo

2w1

2

1

I made the post with the intention to capture a few computer scientist or mathematicians who could make sense of it, and I'm pleasantly surprised to have found a few. It's really a thought expirement that is meant to explicitly define our social heuristics and their corresponding algorithms on a tree. I wanted to see if I could model people's mode of conversing, how they progress through a conversation, using just graph theory. Some of the corresponding algorithms, I thought, might have implications outside of the thought expirement. That is the beauty of thought expirements and their abstraction.

favorite

0

1

Reply

AnhQuan

1 month ago

INTP
Sagittarius

What you're essentially modelling is a Turing test I think. Let's change the reward system to say that the conversation terminates when the other party discovers that one side is a computer instead of just being disinterested in the topic. If a machine were piloting the model rather than a human, that would probably warrant a passing grade.

favorite

0

0

Reply

Midori

1 month ago

ENTP
Cancer

7w8

7

8

If this is some talk in dating then maybe i do need time off 😂

favorite

0

0

Reply

RELATED POSTS

Kevin

1 day ago

ESTP
Leo
True or false girls like big guys
favorite

0

4

Kevin

1 day ago

ESTP
Leo
Do y’all think I’m cute
favorite

0

2

Edward

1 day ago

ISTJ
Gemini
what is the trending tiktok that is just simple science??
favorite

0

0

Adam

1 day ago

ENFP
Aries

5w6

5

6

Sharks aren't your enemy
favorite

5