2d:null 2e:["$","title",null,{"children":"SM3809_A10 | John C | Observable"}] 33:Tbdd,html`

SM3809_A10

Documenting Works

Read

Point of Collection by Mimi Onuoha: In my high school statics classes, one of the first few things we learned is that how to collect data matters a lot. If we conduct a survey about how much do Hong Kong people spend on food every week by counting food checkouts in Taste supermarkets, we will have a number that is expected to be bigger than the fact. That's called sample representativeness. In this case, people with lower income are not represented because they are unreachable in the data collecting process. This kind of bias sometimes is made by mistakes or limits in the collecting process, but it can also be made intentionally to mislead others. Thus, when we are going to believe or use a set of data, it is important to figure out how it was made (or at least who made it, and if the maker is trustworthy or not).
Aside from the bias, one interesting thing this article mentions is that when we made a set of data, we tend to make it fits the algorithm(s). I think this idea is thought-provoking: we use data of something mean we quantify that thing, the first question is, of cause, is it 'quantifiable', but also, how do we decide that it is 'correctly quantified'. I think an example can be how do we evaluate students' performance. Our usual 'data' approach is to quantify them into marks and grades (so we can apply a very simple sort algorithm to them), but as we all know now, this approach is very doubtful as it is not inclusive and fair enough. I think it is important to have similar concerns in mind when solving other problems with data.
Excavating AI by Crawford & Paglen:
- A big question: to train an AI we need an incredible amount of data, thus we can not look at them one by one to ensure they are all OK, then how can we completely prevent (which means make the possibility drop to 0) problematic data learned by the machine? (also, how do we can possibly label the huge dataset?)
- I found it somehow related to Foucault's idea of episteme: we are creating an episteme for the machine because the labeled data set not only tell that this is an image of an apple and this is an image of an orange, but also implies a boundary of knowledge that this AI model can know, and what 'knowledge' means to the AI.

Review

My stupid neural network

Collect

Leung Man Tao: "Naai Gong" or "Gang Du"
In Liz's

Just another old work of mine that is related to the topic: Another Day at Work

`34:[] 31:{"id":"37254456ef32a010","slug":"sm3809-a10","title":"SM3809_A10","thumbnail":"6f3520bf0334f7252a7a653eeeec82f7a24511b89da263dbe741c0e180b23d72","access_level":"public","create_time":"$D2021-03-28T16:09:48.898Z","update_time":"$D2021-03-31T19:05:38.393Z","view_time":null,"first_public_time":"$D2021-03-29T13:35:14.039Z","has_unpublished_edits":null,"fork":false,"stars":0,"starred":false,"owner":{"type":"team","id":"b2e9840be7b1b607","login":"real-john-cheung","name":"John C","avatar_url":"https://avatars-next.observableusercontent.com/1d9ef6121a0bbd1a75fab479f9891fe1d78cc65f76c394585ae1a7dad0d3abf6","tier":"starter_2024"},"version":53,"publish_version":53,"role":null,"authored":false,"authors":[{"type":"individual","id":"e8e99acdbe3aebfb","login":"real-john-cheung","name":"John C","avatar_url":"https://avatars-next.observableusercontent.com/1d9ef6121a0bbd1a75fab479f9891fe1d78cc65f76c394585ae1a7dad0d3abf6","tier":null}],"members":[{"type":"individual","id":"e8e99acdbe3aebfb","login":"real-john-cheung","name":"John C","avatar_url":"https://avatars-next.observableusercontent.com/1d9ef6121a0bbd1a75fab479f9891fe1d78cc65f76c394585ae1a7dad0d3abf6","tier":null,"role":"editor"}],"comments":[],"custom_thumbnail":false,"trashed":false,"lapsed":null,"subscribed":null,"forks":0,"collection":null,"type":"notebook","body":{"title":"SM3809_A10","cells":[{"id":0,"value":"$33","pinned":false,"hidden":false,"mode":"ojs","author":"$undefined","summary":"$undefined","output":"$undefined","database":"$undefined"},{"id":4,"value":"html`\n