"We recommend mitigating these risks by budgeting for curation and documentation at the start of a project and only creating datasets as large as can be sufficiently documented.
""When we rely on ever larger datasets we risk incurring documentation debt, i.e. putting ourselves in a situation where the datasets are both undocumented and too large to document post hoc."" p.1
""Without documentation, one cannot try to understand training data characteristics in order to mitigate some of these attested issues or even unknown ones."" p.1
""LMs are not performing natural language understanding (NLU), and only have success in tasks that can be approached by manipulating linguistic form."" p.1
n-grams
In the fields of computational linguistics and probability, an n-gram is a contiguous sequence of n items from a given sample of text or speech.",0.9999964237213135
"Perhaps a person wearing Google glasses and annotating the experience of wlaking would be more interesting.
DS: Sounds like an over-estimation of reality.",0.999995231628418
"There isn't a lot of reasoning behind these piectures/ They seem quite random.
R: It is scraping any informatino from the url> ??",0.9999942779541016
The n-grams typically are collected from a text or speech corpus.,0.9999821186065674
"Other methods of biodiversity conservation that women contributed to were through tribal wild food festivals where women followed traditional recipes.
[16:18] Michelle: https://canal3lavictoria.cl/
[16:20] Sieta: haha sooo nice to hear the capitalist mindset of ever more and 'better' is completely dismissed in this regards Danae ;)
[16:21] danae: YES <3
[16:21] Linda: ugh! wish I could stay.",0.9999747276306152
"But the matrix and methods have a lot of problems.
Agathe Balayn
3:56 PM
http://aif360.mybluemix.net/data#
A: I still cannot share my screen.
R: The language of 350 degrees.",0.9999736547470093
Generally they don't save any context information.,0.9999595880508423
R: I dont' want to hold engineers to a higher moral standard than myself.",0.9999527931213379
"While ""bag of words"" might well serve as a cautionary reminder to programmers of the essential violence perpetrated to a text and a call to critically question the efficacy of methods based on subsequent transformations, the expression's use seems in practice more like a badge of pride or a schoolyard taunt that would go: Hey language: you're nothin' but a big BAG-OF-WORDS.""
(different to the bags of words of ursula le guin)
carrier bag of fiction
""While the average human is responsible for an estimated 5t 퐶푂2푒 per year,2 the authors trained a Transformer (big) model [136] with neural architecture search and estimated that the training procedure emitted 284t of 퐶푂2.",0.9999395608901978
"Anotherinteresting observation is only10%of the images in MSCOCO have only one category per image, in comparison,over60%of images contain a single object category inImageNet and PASCAL VOC (Source?)
[scanning through several graphs]
[Fig 6: Samples of annotated images in COCO dataset]
[Person Bike Horse]
S: So COCO is better.",0.9999200105667114
"Conundrum.
Cristina: reading from a text by Mimi Onuoha:
https://github.com/MimiOnuoha/missing-datasets
""Missing data sets"" are my term for the blank spots that exist in spaces that are otherwise data-saturated.",0.9998918771743774
Is so alien of the situaiton of walking around on the street.,0.9998674392700195
""In text indexing and other machine reading applications the term ""bag of words"" is frequently used to underscore how processing algorithms often represent text using a data structure (word histograms or weighted vectors) where the original order of the words in sentence form is stripped away.",0.9998548030853271
"The nod, the shaking of the head, a facial tick, too subtle in a grid.",0.9998061060905457
"Mostly these conversations (discussion the socio-technical limitations) do not take place.
R: Arts have limits as well.
DS: This reminded of my classes on distance reading.",0.9997729659080505
Extrasensory signals are not legible.,0.9997721314430237
"To avoid images of single objects in isolation, the authors used combinations of object categories as search terms.",0.9997689127922058
"When the grain is at 45 degrees to its warp and weft threads it is referred to as ""true bias."" Every piece of woven fabric has two biases, perpendicular to each other.",0.999764621257782
The car was detecting shapes and energy but there was nothing there.,0.9997240900993347
Bias is a systematic error.,0.9996728897094727
"And when there’s the delay of bits transferred per second, so what you’re seeing has happened sometime in the past, how do you react to each other in a timely manner?
"""""" Generated text: (from https://deepai.org/machine-learning-model/text-generator)
Reading from both halves: What causes this delay?",0.9996379613876343
That which should be somewhere is not in its expected place; an established system is disrupted by distinct absence.,0.9995997548103333
"Or looking at it differently.
[END
|||STOP||||When you try to arrive at a definition of intelligence, much falls through the cracks.|||STOP||||
|||STOP||||Where does the organism begin or end?",0.999566376209259
Eg: what is a cancerous cell?,0.9994860887527466
"No capital M meaning.
(Michelle) if Turing was to rewrite this text now, how would he re-write it.
(Renée) I think he would use different pedagogy.
(Danae) Yes, but also, he's talking about how this question is not even really relevant.
(Michelle) Yes, this goes to the question of what is it to be human?",0.9994780421257019
"Seen through the optics of the future subsistence of the human species, as well as economic, social, and environmental injustices, this cannibalistic logic ought to be thoroughly criticised.""
28/04/2021 - session 13
[15:08] Linda: p. 13
[15:09] Agathe Balayn: ahah, the train just arrived in Delft finally
[15:10] Cristina: Hello sorry I am late! A meeting before went on 10min longer
[15:19] Renée: does the line break occassionally?
[15:19] Renée: or is it my headphones
[15:19] Agathe Balayn: for me, yes
[15:25] danae: (i love linda's voice <3)
[15:25] Linda: <3
[15:26] Cristina: 40% of Romanian households have no access to the internet
[15:27] Sieta: https://cat.org.uk/
[15:27] Renée: thanks for that statistic Cristina and thanks for the link Sieta
[15:34] Cristina: in the ether :P
[15:34] Michelle: https://guifi.net/en
[15:36] Cristina: (we have a community networks working group if anyone wants to join :P https://pad.vvvvvvaria.org/wg.communitynetworks )
[15:36] Sieta: cool!!
[15:36] Renée: awesome!
[15:36] Noemi: the mentioned visualzier app: http://www.architectureofradio.com/
[15:36] Renée: thanks
[15:37] Cristina: https://www.wired.co.uk/article/google-project-loon-balloon-facebook-aquila-internet-africa
[15:37] Renée: it's like the star
[15:41] Linda: Lesbians are gonna save the world
[15:41] Renée: Information Activism\
[15:41] Cristina: by Cait McKinney
[15:42] Michelle: https://anarchy.translocal.jp/radio/micro/howtotx.html
[15:43] Michelle: http://www.tacticalmediafiles.net/videos/4556/Telestreet_-The-Italian-Media-Jacking-Movement;jsessionid=059CF4D59F1C46374B8215623418BD50
[15:43] Cristina: amazing ref, thanks for sharing
[15:43] Renée: OMG, there are so many good references
[15:43] Linda: coooool
[15:43] Michelle: https://criticalengineering.org/projects/deep-sweep/
[15:46] Michelle: https://mazizone.eu/
[15:47] danae: i think lesbianism becomes super central in mega patriarchal societies such as those in latin america, in the past they became nuns to avoid marrying men, now they're driving cars across the desert installing antennas <3
[15:57] Cristina: Wireless Leiden (NL) : https://wirelessleiden.nl/en
[15:57] Michelle: https://berlin.freifunk.net/
[15:57] danae: https://calyxinstitute.org/
[15:58] Cristina: https://buildyourowninter.net/index.html
[16:00] Renée: These approaches are somewhat designed for working on the fly
[16:02] Renée: rapid deployment of networks
[16:02] Michelle: Some of the critical concerns in the region are loss of traditional knowledge on agro-biodiversity and indigenous crop cultivation, and the impact of climatic change and weather patterns on crop yields and biodiversity.",0.9994762539863586
"Unsurprisingly, this lack of data typically correlates with issues affecting those who are most vulnerable in that context.
The word ""missing"" is inherently normative.",0.9994671940803528
There were pictures of people annotated with character traits. ... associate and full professor.,0.9994106292724609
"Lynn Margulis', evolutionary biologist, research has shown that the evolutionary tree is not an accurate model.",0.9993501305580139
"It explains why the objects in COCO frequently seem neither common, nor in context.""(Philipp Schmitt, https://humans-of.ai/editorial )
Cristina
3:37 PM
""What I am suggesting is that, as a result, COCO represents more than images of objects.",0.9993472695350647
A bad connection?,0.9992490410804749
"There is no rational, logical, reasonable truth.",0.9992367029190063
"A doctor would have the expertise to say which it is, but someone else's expertise might have another opinion.
Anna: machines are unethical.",0.9990365505218506
It is only about the object.,0.9990031123161316
"There has to be a contradition.
[Army photos, people on phones, playing baseball, chopping onions, holding a cell phone, computer sccreen]
R: It raises the question who is being trained actually?
A: The workers train themselves to have the same strategy.
[Who and where are the workers?]
A: A soup is generally in a bowl.",0.9988694787025452
To force somebody into speech is yet another act of violence.,0.9985126852989197
"No support is offered for these statements, acc.",0.998478353023529
"Would it be possible to have the screen share on so I can look at the code while I am writing?
[brings up Microsoft COCO: Common Objects in Context]
Agathe: Let's look through this link, then the pdf.
[opens dataset tab]
S: do you see it?
A: here is a summary of the data set s... persons, people, objects.",0.99845290184021
"The desire to use more neutral language.
R: Interesting in that text, demonstration and not explanation.",0.9982534050941467
It is a bit of a design problem because you have to assign the attributes.,0.998033344745636
"Maybe bringing it out of the rationality. ['Algorithms as Cartomancy', Flavia Dzodan]
Renée
4:29 PM
algorithms as divination... taking them out of the ""rational"" language
Sieta
4:30 PM
my feelings exactly haha
Agathe Balayn
4:30 PM
an escape game to make people aware of AI bias https://www.tomokihara.com/en/escape-the-smart-city.html
L: We need to take it out into another type of field.",0.9979799389839172
"There are not many people doing interdisciplinary work.
(Agathe): it's very frustrating, when I work on more critical stuff, I get told that I shouldn't because I should be finding solutions to things and not coming up with more problems.",0.9978814721107483
"Contrary Views on the Main Question
We are listening to the sound of the above server.
And then we will read the 9 Contrary Views.
Hello :)
Reading: Computing Machinery and Intelligence by Alan Turing.",0.9978200197219849
But there's very few people who are actually doing that.,0.9975645542144775
What about doing?,0.9974269270896912
"Practitioners need to make a choice between the threshold.
R: So this is almost a Mechanical Turk moment.",0.9973836541175842
"There are questions which of course require more, and machines may get stuck.",0.9973746538162231
Then another richer slower interaction interface.,0.9968786239624023
"Bias can come in at any of those junctures.
Renée
4:10 PM
So in terms of the arc... the process is:
Gathering
Categorizing
Searching/ Training
Pattern Recognition
Conclusions
A: There are historical biases in the data as well.",0.9968379735946655
"If he would be embracing today's decentering, anti-racist practices, etc.",0.9967330694198608
"Knows(p, k)∀𝑘𝑘.Commonsense(k) ⟺Knows(ANY-FOOL, k)
Agathe Balayn
5:24 PM
people really believe they can collect ""common sense knowledge"" objectively: http://conceptnet5.media.mit.edu/ (they made a database of common sense knowledge)
possible branches to be unravelled further:
what is intelligence, the artifice?
notions of refusal
curriculum of ai
13/04/21
13/04/2021
We are reading: https://faculty.washington.edu/ebender/papers/Stochastic_Parrots.pdf
measured by leaderboards?
what are language models
Winograd schemas (""The man could not lift his son because he was too fat"" --> who was too fat?! Not too easy to say).
benchmarks are now criticised for their ambiguity and claim to objectivity
baking ethics or a certain perspective into the luangage model
NLP has sought larger datasets with the larger LMs.
""While investigating properties of LMs and how they change with size holds scientific interest, and large LMs have shown improvements on various tasks (§2), we ask whether enough thought has been put into the potential risks associated with developing them and strategies to mitigate these risks.""
(https://searchengineland.com/welcome-bert-google-artificial-intelligence-for-understanding-search-queries-323976 - BERT used for Google search)
There are competing scales to be taken into consideration.
""documentation debt"" = when datasets are so large it is difficult/expensive to document them?
In collecting ever larger datasets we risk incurring documentation debt.",0.9964132905006409
"Crawley
Text: I cannot decide the stability of my plan, depends on dynamics, dispositions, contextual constraints.",0.9959685802459717
"But then, with the case of machine learning, the computer seems to say: “Here you go, that’s your answer.” With Ifá divination there is no certain truth.",0.9959421753883362
It becomes more constrictive due to what the system allows for queerness.,0.9958974123001099
"Deterministic result of perception.
Renee: difference between wink and blink?
Sonia: its confusing.",0.9957922697067261
"Viruses are living forms too|||STOP||||
|||STOP||||Are dualities gradients?|||STOP||||
|||STOP||||Thinking through parallels to the Genesis story, how it also resulted in differenciations|||STOP||||
|||STOP||||Descartes' bad rep.",0.9957851767539978
"But now that attributes are not universal...
R: And they are also trying to design themselves out of a problem.",0.9954558610916138
"We already know the answers, that machines are going to be racist and sexist, which are valid points.",0.9951802492141724
"Humans can also give wrong answers, depending on the type of question and how it is posed.
(Danae) But this is the problem of discreteness, indeed, how do you get these static answers.",0.9948050379753113
"The items can be phonemes, syllables, letters, words or base pairs according to the application.",0.9943752288818359
Which one is top down & which one is bottop up?,0.9935810565948486
In comparisonImageNet and PASCAL VOC both have less than 2 cat-egories and 3 instances per image on average.,0.9935252666473389
Uncertainty?,0.9935110211372375
Do you include nonhumans?,0.9933763146400452
Categories: the amount of sports are quite limited.,0.9931046366691589
"Chekov writing is not too different from the captions of COCO. We also always try to be purely descriptive, which is hard when today everybody is trying to giving opinions, and advancing their objectives.",0.9930288195610046
"What appears to be out sight.
mi
Michelle
6:05 PM
I am here, but I don't want to be included.
mi
Michelle
6:07 PM
Freedom in the silence
mi
Michelle
6:07 PM
The assumptions that you make when you think you are being inclusive.
mi
Michelle
6:09 PM
The freedom of not being nameable.
mi
Michelle
6:09 PM
And how does this work in an AI system?
cr
Cristina
6:10 PM
https://points.datasociety.net/the-point-of-collection-8ee44ad7c2fa#.y0xtfxi2p
cr
Cristina
6:11 PM
ah sorry, it's here
https://github.com/MimiOnuoha/missing-datasets
re
Renée
6:14 PM
https://www.youtube.com/watch?v=Dwr5ax3zek8
Notes: 16 March 2021
[A PROPOSAL FOR A PLAY]
Live Transcription - Michelle
Renee: We are reading through the data sets.
Reading through data sets quality assurance (equality and inequality) ISO quality assurance and how it is also used in education.
ISO verification language.
Thanks for show me that.
Sonia: ISO is the standard verification language.",0.9910876154899597
"As the Zapatistas say, “The world we want is one where many worlds fit.”
[17:37] Renée: who are the curators and what is their agenda - whether formulated forthrightly or through neglect?
[17:39] danae: caracoles :) https://freight.cargo.site/w/971/i/e8998287750263b2bf6b46b316c1dddb61ad72e0d74b4bc807e5c8e94acbb41d/caracolezln.png
[17:42] Sonia: Simon DeDeo
[17:43] Renée: It is who shows up and as a result who is showing up.
[17:48] Renée: using harm rather than bias
[17:48] Renée: some use ""fairness""
[17:49] danae: now the infosec community uses ""harm reduction"" a lot
[17:49] Renée: systematic bias
[17:49] Michelle: how about a care brief?
[17:51] Renée: the word ""harm"" eliminates the mathematical response
[17:52] Michelle: Work on synthetic human behavior is a bright line in ethical
AI development, where downstream effects need to be understood
and modeled in order to block foreseeable harm to society and
different social groups.
[17:53] danae: i guess the text fails at computer science AND at politics :P
[17:53] Sonia: hahaa
[17:53] danae: tbh i was expecting more science
[18:03] Cristina: I’d like to suggest that machine learning is not new.",0.9905669093132019
"What appears to be inaudible, what appears to be out sight still influences the system.
https://www.youtube.com/watch?v=Dwr5ax3zek8
Linda: I also have this problem of being involved.",0.9901575446128845
"What does it mean to recall and remember?
""Tool not visible unless it's broken"" dictum (Heidegger, but also McLuhan in a different way, with the fish quote) -- reminds me of Susan Leigh Star's work on infrastructure --> also, yeah!
What are we forgetting today?
Is it possible to speak of ""reconnection"" at all?",0.9898669719696045
Example: Tesla car near a cemetary.,0.9892633557319641
The delay of an extra stop because you gotta swing by the unmute button on the way to action.,0.988723635673523
Turing calls bullshit.,0.9882431626319885
The delay between thought and action.,0.9881630539894104
Hesitation?,0.9873998165130615
"Without incorporating our experiences, any response to this disaster will fail to change the complex ways in which social, economic and political systems shape our lives – offering some an easy pass in life and making others pay the cost.",0.986914336681366
"Use a headset to avoid causing background noise for others.To join this meeting by phone, dial: +31 15 201 0064Then enter 57941 as the conference PIN number.
[16:21] To invite someone to the meeting, send them this link: https://bbb.tbm.tudelft.nl/b/aga-9ab-8g7-tu1
[16:24] Renée: Agathe - do you feel like joining from the perspective of computer science
[16:25] Renée: So since I'm not doing well on the mic situation - shall I do the glossary
[16:25] Agathe Balayn: https://faculty.washington.edu/ebender/papers/Stochastic_Parrots.pdf
[16:27] Cristina: https://commoncrawl.org/
[16:27] Sonia: commoncrawl.org/
[16:27] Michelle: Common Crawl
[16:28] Cristina: The Internet is a large and diverse virtual space, and accordingly, it is easy to imagine that very large datasets, such as Common Crawl (“petabytes of data collected over 8 years of web crawling”,11 a
filtered version of which is included in the GPT-3 training data) must therefore be broadly representative of the ways in which different people view the world.
[16:29] Cristina: (under Size Doesn’t Guarantee Diversity)
[16:30] Cristina: https://twitter.com/Abebab/status/1309137018404958215
[16:41] Michelle: The Colossal Clean Crawled Corpus
[16:44] Michelle: We instead propose practices that actively seek to include communities underrepresented on the Internet.",0.9859368205070496
Breathing is automatic and out of our control.,0.9828603863716125
I refuse to make my language palatable to your ears.,0.9824673533439636
It’s simply digital divination.,0.9819814562797546
"These are the possibilities of the near future, rather than Utopian Dreams
6) Lady Lovelace's objection: a machine can do whatever we know how to order it to perform! Refutation to the conjecture: yes, but this doesn't meant that we know exactly what the consequences of our order are.",0.98089200258255
"Not sure how to go about it.
[This is the starting point]
A: Now they have their data set and they are going to predict their model.
R: It is great to see 'privileged groups', and 'unprivileged groups' to get an idea of semantics.
A: Mostly they are picking and defining values based on race or gender or a combination of the two.
[scanning through code written in Python]
Protected attribute names
('RACE;)
Cannot do more than two groups at once (0 and 1) binary
Privileged and unprivileged protected attribute values
3.2.2.",0.9795517921447754
"For example, what would he think of drones?
Mr. Turing
Hormones.",0.9759359955787659
"In order to understand you need to take into account the meanings and context where the communication lies.
The basics. depending on the information you are trying to get you will either undesrtand 'wink' or 'blink'
Renee: adjacent reading. question to Danae
Sonia: randomness in terms of quantum floctuation is what gives life to consiousness and free will. and if we are determinsit beings it doesn't really matter based on ethic.
there's a conversation, you become a person out of 2 people's cells, but then you become something that has an influence on your cells. (food, habits...) It used to be very DNA centric, but epigenetics came by to kind of dismantle that idea.
in the case of singular progress, as we become more autonomous (zigot) it becaomes more self sufficient, but as organisms we are inclosed in an environment. we adapt. so the more autonomous an organism is the more it needs to externalise its needs. (discovering fire to be able to make meat more digestive)/ that puts the stomach out of the body. (car are the legs) and this is relateed to observing related to environment.
example about the super inteligenc book.",0.9757429957389832
Critics of the methods of distance reading.,0.9748100638389587
"Rationalism. |||STOP||||
|||STOP||||Karen Barad's Meeting the Universe Half Way is not putting away rationality.",0.9747478365898132
To predict whether there will be expenses.,0.972797155380249
"It all depends on the context. information theory ex. you try to define signal and noise, when communication is made.",0.9705168604850769
"Something which remains unresolved, neither proved nor disproved, when we need ""yes"" or ""no"" answers.",0.9671815037727356
"Consolation would be more appropriate than refutation in the case of this argument...
The anticipatory potential of doom-scnearios, it's not always technophobic (Danae)
3) The mathematical objection: in logic we can see that there are limitations to discrete state machines, e.g.",0.9660007953643799
"Websites, tweets, reddit etc are more problematic because they are even more contextual.",0.965435266494751
I think about delays.,0.9618654251098633
"Sometimes a system of morality is also inhibiting action
Linda: so the freedom we are talking about then, is it the freedom to be a finger while knowing that you are attached to a body?
Sieta: people who stick to one ideology often think that there's only the finger, it seems to me.",0.961746871471405
"Whether that url is adding, framing the dataset. if you have the url there will be textual framing.
[guy with home cooking blog]
R: The circle is a cake.
A: Most datasets scrape images from the web.",0.9603193998336792
Glance: we only had a few seconds to look at the images.,0.9596940279006958
"I am here but I do not want to be included, to be counted.",0.9584084749221802
"Sometimes the case is ethics-washing.
Agathe: how to relate machine learning to this complex system theory?",0.9583925604820251
"Give ppl safe spaces where they can say: I have a voice, but I refuse to make it heard to you.",0.9557873010635376
"When reading the group through the screen, we search for disturbances, anomalies in the field.
Reading between the lines, reading between the actions/unmuted words: What occupies that “between”?",0.9538184404373169
Novacene: The coming age of hyperintelligence.,0.952522337436676
"He wouldn't have many incentives to do that.
Sonia wonders about reading Kittler (thanks) and suggests reading Alicia Juarrero.
Agathe - mentions the trenches of disciplinarity
For next week we are going to read: Alicia Juarrero's Dynamics in Action (Chapter 15):
https://aliciajuarrerodotcom1.files.wordpress.com/2012/02/dynamics-in-action-pdf1.pdf
Michelle suggests and adjacent reading of Juarrero-and...?",0.9432322978973389
"One has not yet observed a machine that can do X, so a machine ""can't do X"".
our notes drop out as Sonia reads half of Arguements from Various Disabilities and Renée picks up.",0.9422363638877869
ML has its own constraints.,0.9412145018577576
We have to take dymanics and context-dependence seriously.,0.9405133724212646
Choosing between the threshold between 0 and 1 to determine whether a person should go to prison for example.,0.9399721026420593
"But the context is makes the object. \
[Context matters]
S: Cars in the background.",0.9396147727966309
"The dialogical in lieu of the dialectical. guerra di movimento vs Guerra di posizione (Gramsci).
---break----
Text: one advantage of dynamical systems theory is that it can handle self-cause.
Interjection: everyone looking into Ruth Benedict guilt-based versus shame-based.
Anna: apparently Benedict talks about Americans as guilt-based, Japanese as shame-based.
Text: the implications of dynamical systems for human development are vast. ""We still cannot identify those thresholds of psychological or
emotional disequilibrium when intervention would be most effective
. Making matters worse is the fact that since in practical terms, each
run of a complex adaptive system is unique, educational and child -rearing
techniques that might work for a Hannah may not work for a Gabriel .
And techniques that might have worked for Hannah yesterday may not
today ."" (Juarrero P?)
Renée: interesting how knowledge acquisition is different at every stage, every day you are a new learner.
Linda: I'm trying to wrap my head around attractors: at the conscious and unconscious level, there are things are we decide upon and there are unconscious things that are happening in the background.",0.9368321895599365
"If we release the framework of failure, we can really that we are in iterative cycles, and we can keep asking ourselves—how do I learn from this?
Emotional growth is nonlinear.",0.9365696310997009
"If we release the framework of failure, we can really that we are in iterative cycles, and we can keep asking ourselves—how do I learn from this?
Emotional growth is nonlinear.",0.9365696310997009
The system should be buffered from perturbations of all sorts.,0.9361927509307861
"White long dress.
[Normative, Westernized images]
R: Mechanical Turks and where they might reside.",0.9307928681373596
One person gets three images.,0.9263993501663208
"Maybe something to chat about later)
What is the best way of framing a conversation around positive versus negative (or other forms) of exposure?
> predatory inclusion
> evitable technology
> ""they tried to bury us, they didn't know we were seeds""
> subaltern pattern making
What kind of infrastructure is needed for subaltern pattern making?
Chat:
[16:08] Michelle : Undrowned
[16:08] Michelle : Black Feminist Lessons from Marine Mammals
[16:08] Michelle : Alexis Pauline Gumbs
[16:10] Cristina : hello sorry i am late, i was caught up in another meeting!
[16:11] Renée : I guess so
[16:11] Renée : So then we hit mute?
[16:11] Renée : or is it playing the same way for everyone
[16:12] Cristina : do we hit play?
[16:12] Renée : meaning is the timing similar
[16:13] Sieta : questions here
[16:14] Renée : We will do the questions here:
[16:14] Sieta : did you press play?
[16:14] Sieta : mine did not move
[16:14] Sonia : I pressed play to start it
[16:14] Agathe Balayn : mine either
[16:14] danae : same here
[16:15] Renée : press play yourself
[16:15] Cristina : mine either, i pressed play myself
[16:15] Sieta : haha oke
[16:15] Sieta : yes it starts!
[16:15] Cristina : it started again yes
[16:15] Renée : now we are synched
[16:19] Renée : What brings you joy in this time? (Remember to share that on the pad)
[16:21] Renée : What kind of technologies do we want in the first place?
This is not about access as that works with something predefined – to be given entry.
[16:21] Renée : How would we embed our highest values?
[16:21] Renée : What is depth in the context of deep learning?
[16:23] Cristina : (side-question: did anyone read the Indigenous AI position paper? https://www.indigenous-ai.net/position-paper/ )
[16:23] Renée : What is the role of social and cultural memory within the context of technology?",0.9248291254043579
"MAybe we do not desire the system to be autonomous, because it has a political effect.
Sonia: she's not necessarily connecting freedom as 'good'. but she's more quantizising the levels of freedom.
Danae: its easy to connect 'free' with 'more objective'
Sonia: in anarchist systems i was working with systems.
Cristina: i was curious about containerisation. and how would that be worrying?
Danae: containeristation requires an amount of new skills, so we are in the midst of this problem, sustaining a more conservative system or embrace a new one?",0.9210001826286316
"There is a temporal acceleration.
https://commoncrawl.org/
Renee: datasets as Rapuccini's daughter (?) https://en.wikipedia.org/wiki/Rappaccini%27s_Daughter
visible vs invisible biases
Sonia: bias is unavoidable, better embrace it, acknowledge it
(nicolas maleve's presentation on current trends in the bias discussion:
https://en.wikipedia.org/wiki/Grain_(textile)
The bias grain of a piece of woven fabric, usually referred to simply as ""the bias"", is any grain that falls between the straight and cross grains.",0.913240373134613
"Image recognition with people recognized as animals, and...
A: Pictures ... recently of a hand holding a thermomator ..",0.9087361693382263
An action will already exclude other possible actions.,0.9058907628059387
"In order to have anything, there first must be differentiation or difference.|||STOP||||
|||STOP||||How is this system relational?|||STOP||||
|||STOP||||Embodied shift.|||STOP||||
|||STOP||||Programatic language present in the text |||STOP||||
|||STOP||||Historical bias manifested in dualist thinking, relationship to ""unities"".",0.9034404158592224
"Training a single BERT base model (without hyperparameter tuning) on GPUs was estimated to require as much energy as a trans-American flight.""
Can you imagine if a similar amount of energy was expended on educating a child?
Can you put that comment in the chat Agathe....
I'm not sure, but I believe that there is a tax to pay depending on the carbon emissions you create, and academic researchers probably don't have the financial resources to pay this tax directly, and so to do this research. https://arxiv.org/pdf/1906.02243.pdf
Look up: SustainNLP workshop
https://sites.google.com/view/sustainlp2020/home
This! ""Among their recommendations are to run experiments in carbon friendly regions, consistently report energy and carbon metrics, and consider energyperformance trade-offs before deploying energy hungry models.",0.9002392888069153
"That the silence can have some influence or power is interesting.
Sonia: When you think you are being inclusive, you may also exclude the people you are not aware of.",0.8999819159507751
"In addition to these calls for documentation and technical fixes, Bietti and Vatanparast underscore the need for social and political engagement in shaping a future where data driven systems have minimal negative impact on the environment [16]."" (p.3)
“Feeding AI systems on the world’s beauty, ugliness, and cruelty, but expecting it to reflect only the beauty is a fantasy.” (Ruha Benjamin)
""When we perform risk/benefit analyses of language technology, we must keep in mind how the risks and benefits are distributed, because they do not accrue to the same people.""(p.3)
""Is it fair or just to ask, for example, that the residents of the Maldives (likely to be underwater by 2100) or the 800,000 people in Sudan affected by drastic floods pay the environmental price of training and deploying ever larger English LMs, when similar large-scale models aren’t being produced for Dhivehi or Sudanese Arabic?""
https://www.redpepper.org.uk/an-open-letter-to-extinction-rebellion/
""An open letter to Extinction Rebellion""The fight for climate justice is the fight of our lives, and we need to do it right."" By grassroots collective Wretched of The Earth.""
Our communities have been on fire for a long time and these flames are fanned by our exclusion and silencing.",0.8991383910179138
"We start with the 9 objections to the view.
Introduction to Turing and his life.
Can machines think?
Opinions will differ on this problem, we need to pay attention to all views.
Turing's own belief: the question in 50 years will be different because we will have faster computers, which can play the imitation game, they will fool us.
The question is meaningless, at the end of the century (20th) we will be speaking of thinking machines.
Conjectures are of great importance, so let's consider the opinions which are opposed to my own.
1) The theological objection: God-given characteristic of intelligence.",0.8945223689079285
The way to very forefully introduce this information into theis machines for events in the future is completely at odds of the experience of walking throught the city.,0.8940938115119934
"How does that work as well?
Renee: HOw does the camera frame the recognition?
Sonia: Let's look through several introductions.
Michelle: I am doing live transcription and intend to write a play from this meeting.",0.880581796169281
"We emphasize that
crowd labeling is only necessary for images containing
more than ten object instances of a given category. (Source?)
DS: A group of human judges evaluating whether the caption is right or wrong.",0.8738699555397034
The scandal wit Google.,0.8727152347564697
"Is this where we are at?
A: Making a system to predict whether somebody will have to go to the hospital or not.",0.869774580001831
"Paradigms of categorisation.|||STOP||||
|||STOP||||Sympoesis - nothing is autopoiesis, things are never alone.",0.8691532015800476
"One way to understand this is through the
lens of “appropriate technology”, defined as being
small-scale, affordable by locals, decentralised,
labour-intensive, energy-efficient, environmentally
sound, and locally autonomous.6
In this definition
we find similar dynamics in land stewardship and
small-scale agriculture insofar as the appropriate
technology movement grew out of the energy crisis
of the 1970s, similar to land-based approaches that
promote environmental conservation by seeking to
“close the cycle”, such as permaculture.
[16:44] Sieta: a really nice book about permaculture and society is Cultural Emergence by Looby Macnamara
[16:45] Cristina: the phrase ""appropriate data"" from Timnit Gebru still haunts me
[16:46] Welcome to Slow reading!For help on using BigBlueButton see these (short) tutorial videos.To join the audio bridge click the phone button.",0.8682141900062561
"The cold eye, what would it do... 15 years ago I had a Polish student in PZI and was interested in political media images.",0.8665051460266113
"In literature we are not interested in patterns, but exceptions.",0.8633744120597839
Induction is the key issue w/ this conjecture.,0.8451166749000549
"Validating LR model on original data
unprivileged groups=unprivileged groups,
privileged groups=privileged groups
print(explainer_orig_panel19_train.disparate_image))
from collections import defaultdict.
C: Do you think they developed it themselves?
def text(dataset, model, thresh_arr)
A: This validation model.",0.8448487520217896
"How are the icons developed?
A: Interesting to see what they decide to annotate ... home images... subjective.",0.841025173664093
Or let the workers annotate whatever is important in the image.,0.8153179883956909
"A person dressed in colourful clothes. \
R: Almost like Haiku
S: The situation of being in front of the computer.",0.8095386624336243
That is where the interesting story exists.,0.8065817356109619
"Speech vs silence: if you have speech you are an author, if you are silent, it is a pathology which needs to be cured.",0.8041458129882812
Then a group of three coming up with a common description.,0.8032424449920654
"Allen Lane.]
2) The ""heads in the sand"" objection: scared of machines.",0.8026077151298523
"Why do people become who they are?
Sonia: time, trauma, experience and the influence of media: https://asc-cybernetics.org/2008/HM-08WienerComments.pdf
Michelle: ""
Transformation doesn’t happen in a linear way, at least not one we can always track.",0.8021067976951599
"And she connects it to her work on oracle bots.
Rather than a problem of the mathematical capabilities of the machine, it can instead (missed this part) (be a problem of the receiver?)
machines produce mathematical fictions
If Turing was to re-write this text in the present, how would he re-write it?
What is it to be human?
Would he reframe his work under more post colonial framework.
What would be the questions you would ask him if he were here?
We can't speak for the dead.",0.791527509689331
"There was a dataset on people who had been to the hospital for long periods of time to develop a dataset based on fairness.
A: This whole section.",0.7866371273994446
"Just because some type of data doesn't exist doesn't mean it's missing, and the idea of missing data sets is inextricably tied to a more expansive climate of inevitable and routine data collection.
Renée: next session: what about allocating the first 20 mins to wrapping this text up.",0.7623746395111084
"But it is not enough of this approach that finds something in this that can advance the arts, fine arts and scholarship of what these things can provide.
Renée
3:39 PM
So they are purely descriptive as opposed to interpretative.
Cristina
3:36 PM
""For COCO, the authors wanted, say, pictures of cats on couches, but not of cats posing in front of a white background for a feline studio shoot.",0.7592524290084839
"In the context of fairness, we are concerned with unwanted bias that places privileged groups at a systematic advantage and unprivileged groups at a systematic disadvantage.",0.7579811811447144
"Protected attributes are not universal, but are application specific.",0.7502331733703613
"I wonder how this was in his time.
(Danae): nowadays things have become very segregated, and in the humanities we demand interdisciplinarity.",0.7469510436058044
"Many don't question subjectivity.
-(Sonia) Leopard and tree example.",0.7398055791854858
"It's a political question in the end, people have agendas.
(Sonia): yes, the main line of attack from the humanities is indeed this idea that we have to counter this scientism-solutionism... etc etc.
4) The argument from consciousness: ""Not until a machine can write a sonnet...."" The refutation for this conjecture is solipsism: we also don't know what and if people feel when they feel.
5) Argument from various disabilities: (Sonia:) the modern version of this is the ""AI is whatever we still can't do with machines"".",0.7380810976028442
"But if he were in the room, what would be the questions we would ask him?",0.7313891053199768
What does silence or inaction mean in this context?,0.7138232588768005
"Therefore, you will battle between the level of freedom and systems.",0.7087022662162781
"But looking at randomness in pi, we cannot actually distinguish between the behavior of a human or a machine.
8) The argument from informality of behavior: the endless possibilities of behavior.",0.7078030109405518
"Name(a) = ANY-FOOL ∀𝑘𝑘.Knows(ANY-FOOL, k) ⟺∀𝑝𝑝∈Persons.",0.7043132781982422
No connotations.,0.698525607585907
But it doesn't discuss that different people have different biases.,0.6911488771438599
Some people in this network are trying to produce some type of computation under the logic of containerisation.,0.6757782101631165
"Pictures with computers is important to annotate.
[cats and computers, sinks, floors, cats in bottles, cats on a chair, cat on a sofa]
S: Our future is imagined through these flat surfaces instead of 3D spaces (with sounds)
R: COCO comes from the Gorilla, Like COCO wanta a banaan?
S: Perhaps happy coincidence.
S: Should I click on something else?
[brings up horses, in landscapes, hills, towns, etc]
A: These have been critized because of these connotations.
S: Of course it is going to be pretty safe.",0.6713740229606628
"In order to envision a future in which we will all be liberated from the root causes of the climate crisis – capitalism, extractivism, racism, sexism, classism, ableism and other systems of oppression – the climate movement must reflect the complex realities of everyone’s lives in their narrative.
parallels between deciding what to do with AI models and what to do with institutions; reforming institutions
ref to How Not To Teach (??) from a compedium of anarchist pedagogy texts
linda: how can you talk about refusal when you don't even have a base understanding? which feminism can actualise in data science? is it only liberal feminism? does the radical have space?
renee: what are the leaderboards/benchmarks for this text?
the meta elements (structure, refs) of the text are what are interesting maybe more so than the context
agathe: how do these topics become accepted through the peer review processes?
danae last week: the decline of american imperialism
the very last sentence of the article: ""Thus what is also needed is scholarship on the benefits, harms, and risks of mimicking humans and thoughtful design of target tasks grounded in use cases sufficiently concrete to allow collaborative design with affected communities.""
https://www.metamute.org/editorial/articles/californian-ideology
for next time:
https://www.apc.org/sites/default/files/sneak_peek2020_final.pdf (maybe rhizomatica? page 13)
cristina: a quote by Mariame Kaba that seems to fit our next reading:
""When something can't be fixed, then the question is what can we build instead?""
The Chat - April 20th
[16:14] Michelle: somebody is sending scratchy noises :)
[16:15] Renée: I'm seeing if it is me
[16:21] Welcome to Slow reading!For help on using BigBlueButton see these (short) tutorial videos.To join the audio bridge click the phone button.",0.6701288223266602
One of the challenges of a slow reading group during a pandemic: how to read each other if we are to read together.,0.659231960773468
A fairness metric is a quantification of unwanted bias in training data or models.,0.6459302306175232
"Also AI is a form of complex system, it learns from what it knows previously.",0.627545177936554
Hormones are literally what wire and rewire us.,0.627469539642334
In CS nowadays you need to build new tools and not reflect on existing ones.,0.6086239814758301
White and black drawing.,0.5871201157569885
"Aren't we disconnected, per definition?
(it's so cute that someone is agreeing in the background all the time)
(haha yes i thought it was one of us in the beginning)
Can technology help us to remember what we have forgotten?
What is consent in the context of data extraction?
How do you deal with the desire for social domination?!?!! (Age-old question, still not solved by things like democracy..)
(We were just talking about NFTs with Anna, something that has the same potential dangers as what she's talking about right now.",0.5834068059921265
Both the Ifá system and computers are trying to make a decision amidst great uncertainty.,0.5675846934318542
So Twitter started to say that it was not a problem of fairness.,0.5657301545143127
"The imitation game is also a refutation to this conjecture.
(Sonia) explains the Gödel problem, barbers.
(Agathe) nowadays we deal with different problems in AI, we're talking about patterns in data and less about logic problems.",0.5281524062156677
"To abolish Big Data would mean to put data in the hands of people who need it the most.
[16:42] Cristina: transgressive potatoes
[16:43] Cristina: While this digital divide must be addressed,
information and communications technologies
(ICTs) can and must be employed and deployed
differently.",0.5220670104026794
"What happens in the web context where images can land in so many different contexts.
S: Segmentation. 2,500,000. 22 worker hours per 1,000 segmentations.",0.5214776992797852
"Sonia and Danae lean towards Maturana.
Sonia suggests we read the last chapter (15) of DiA.",0.513395369052887
Violent images but absolutely neutral.,0.5035126805305481
"When we talk about fairness in ML the decision is usually to ""de-bias"" the system but without reflecting that you will not be able to take into account all possible contexts.",0.5034852027893066
"To organise systems today, which can be related to the idea in the text, the logic that if the goal is met the system will be freer.",0.5014795660972595