DS (Anti MRA)Nov 13, 2019
Anupam Guha
Wrote something after ages for a newspaper, on why we should resist reifying AI governance and pushing blame on algorithms w r t human decisions, in the light of the birdsite going after @sanjayuvacha and @ladkidyutiful.
I get nervous with non academic writing, as I don't want to dilute the argument but still want to make it as accessible and as non jargony as possible. Do have a read.
https://www.thenewsminute.com/article/artificial-scapegoat-why-algorithms-cant-be-blamed-political-or-corporate-decisions-112248
Artificial scapegoat: Why algorithms can't be blamed for political or corporate decisions

Recently, the social media site Twitter saw a spate of suspensions which have seemingly contradicted its own Terms of Service (TOS). A lot of Kashmiri, Dalit, journalist, and progressive accounts were slapped by tweet deletions and account suspensions, some permanent without offering much in the way of explanations.

Nov 13, 2019 at 1:06pmWeb
Show threadAnupam GuhaNov 13, 2019
Am interested to know if people find these kind of explainers useful and will read them. Thinking of doing a series on AI, history policy and future stuff
Show threadashwinNov 13, 2019
@Anupam_Guha yes. I am a technologist working on FAT AI systems at my firm. This is extremely useful discourse and should be publicly available. The academic papers on these matters are too dense for a layman to follow. :)
Show threadashwinNov 13, 2019

@Anupam_Guha while I agree with your article in many ways. We can also use AI based systems to detect inherent bias and bring it to the attention of human actors. While mitigating bias maybe the ultimate goal, I often tend to see AI systems as human-enabled and not independent decision makers.

An AI system that gives an output along with interpretable/explainable models are useful compared to a black box.

Show threadPrav🦸Nov 13, 2019
@Anupam_Guha would love to read and also support it any way possible
Show threadMusafirNov 13, 2019
@Anupam_Guha positive. I definitely would love to know more on AI and governance. Keep me in the loop please.
Show threadNNov 13, 2019
@Anupam_Guha Yes. Interested.
Show threadskybrianNov 14, 2019

@Anupam_Guha it seems very free of jargon but also rather introductory? But since I've been reading about this topic for years, perhaps I'm not the intended audience.

A few points:
- It seems like saying "the machine did it" should be treated not as blaming a scapegoat, but as admitting an error, though a different kind. An organization must take responsibility for its machines.

But admitting this is just the first step. How are the errors remedied and new errors prevented? Is there a robust process to investigate and fix mistakes? Admitting a serious error and doing nothing about it is either arrogance or helplessness.

- An army of workers would also make mistakes when they don't have the background knowledge to understand the meaning of a photo or poem. I don't see how it can be avoided unless the moderator is sufficiently part of the community they are moderating to know the people and understand what they are saying. This seems incompatible with Twitter's flat, global structure and fits better with Mastodon, Reddit, or other places that have small subdivisions with their own moderators.

- Machine filtering seems to work well when combined with local overrides. Consider Gmail's spam filter, which seems to work for many people, because it allows the user to train the algorithm and override it. This early success perhaps made people too optimistic that it could be used globally for social media?

Show threadAnupam GuhaNov 15, 2019
@skybrian extremely introductory
Show threadModerngypsyNov 16, 2019
@Anupam_Guha three different topics.... but yes interesting. Would be interested to hear about possible AI applications to Management Control Systems...
Show threadModerngypsyNov 16, 2019
@Anupam_Guha history policy? History is what it is .... History policy is what interests the RW 😁
Show threadAnupam GuhaNov 16, 2019

@moderngypsy meant AI-history, AI-policy, AI-future trends

should have used a colon instead of a comma

Show threadModerngypsyNov 16, 2019
@Anupam_Guha ok makes sense for lesser mortals like me 😁
Show threadMusafirNov 13, 2019
@Anupam_Guha insightful indeed. @sanjayuvacha @ladkidyutiful
Show threadPrav🦸Nov 13, 2019
@Anupam_Guha @sanjayuvacha @ladkidyutiful Well written article on a point that most struggle to understand including the developer and scholar community. But there is human intervention on top of the AI decisions including the Twitter or in any other platform. So AI being biased should never be a reason for these issues and the companies who implemented should own up their mistakes even in the current scenario. Or else is transparent that something happening behind the screens.
Show threadThe Last of her NameNov 13, 2019
@Anupam_Guha @sanjayuvacha @ladkidyutiful
That was a very enlightening article, Sir. I've been interested in the sociological analysis of AI, ever since I read Virginia Eubank's Automating Inequality. I'm just wondering if you can suggest more readings on this topic.
Show threadAnupam GuhaNov 13, 2019

@Lawandemotions @sanjayuvacha @ladkidyutiful

Some readings to start with

Begin with Kate Crawford's https://anatomyof.ai

Then Evgeny Morozov's critique of Zuboff's book on "surveillance capitalism", read https://thebaffler.com/latest/capitalisms-new-clothes-morozov (not exactly on AI but the arguments hold)

Also Cathy O'Neill's "Weapon's of Math Destruction"

Anatomy of an AI System

Anatomy of an AI System - The Amazon Echo as an anatomical map of human labor, data and planetary resources. By Kate Crawford and Vladan Joler (2018)

Anatomy of an AI System
Show threadashwinNov 13, 2019
@Anupam_Guha @Lawandemotions @sanjayuvacha @ladkidyutiful also let me plug https://ainowinstitute.org their report on AI ethics, labour and policy is a great read.
Home

We challenge & reimagine the current trajectory for AI.

AI Now Institute
Show thread@[email protected]Nov 13, 2019
@Anupam_Guha @sanjayuvacha @ladkidyutiful That is why linguists say AI, Deep learning are a product of the biases of the humans behind it. A look at the data on which these tools are being trained is very revealing and disturbing.