We used fictitious resumes to find employer bias. We used statistics to find bias in a healthcare algorithm. Side by side these studies show why discrimination perpetrated by algorithms is so very different from discrimination perpetrated by people
People are biased. Data is biased, in part because people are biased. Algorithms trained on biased data are biased. But learning algorithms themselves are not biased. Bias in data can be fixed. Bias in people is harder to fix.
There's plenty that I agree with in this article but the thing that I cannot abide is the concept that finding discrimination in privately held, IP-protected algorithms is easy is just plain wrong.
โ€œBiased algorithms are easier to fix than biased peopleโ€. But, in some cases, I don't see these as entirely different problems. Sometimes fixing a biased algorithm first requires fixing biased people.
. in makes a key point about algorithmic bias: "Changing algorithms is easier than changing people: software on computers can be updated; the โ€œwetwareโ€ in our brains has so far proven much less pliable" #AI #2MA
We keep blaming machines for bias. But the real problem is us humans. provides a nice summary w/ some of his seminal work w/ #AI algorithms
Essential reading from : Yes, algorithms can be just a biased as human beings. But... "Changing algorithms is easier than changing people: software on computers can be updated; the โ€œwetwareโ€ in our brains has so far proven much less pliable.
"It is much easier to fix a camera that does not register dark skin than to fix a photographer who fails to see dark skinned people." Awesome, personal essay by on their recent paper on racism in health algorithms
Biased Algorithms Are Easier to Fix Than Biased People
A must-read for all decision makers: since biased algorithms are far easier to fix than biased people, shying away from imperfect algorithms is a huge mistake. From the always-insightful writing for
Biased Algorithms Are Easier to Fix Than Biased People - insightful piece by โฆโฉ in โฆโฉ ๐Ÿ‘๐Ÿพ โฆโฉ โฆโฉ โฆโฉ โฆโฉ โฆโฉ โฆโฉ
this is just superb โ€” on a lesson in contrasts: algorithms and humans. when it comes to bias/discrimination, it may be easier to โ€œfixโ€ the camera than the photographer. But weโ€™re going to need both to improve the human condition.
.: "It is much easier to fix a camera that does not register dark skin than to fix a photographer who fails to see dark skinned people."
"Pinpointing discriminatory behavior by a particular group of people โ€” in this case, hiring managers โ€” is often very hard. By contrast, uncovering algorithmic discrimination was far more straightforward" via
"Biased Algorithms Are Easier to Fix Than Biased People" in the every time I say it, people look at me, offended. Glad someone with much much more credentials is explaining it with the caveats it needs. (It still needs more caveats).
Recommended read by : If we want present & future #AI to take just decisions, we need to tackle human & algorithmic #bias side-by-side. Technical solutions won't suffice: Deeply-entrenched human biases will be hard(er) to detect & overcome.
"Compared with the intransigence of human bias...discrimination by algorithm can be more readily discovered and more easily fixed" on how algorithmic bias is a relatively tractable challenge.
Biased Algorithms Are Easier to Fix Than Biased People
Finally, a nuanced take! Biased Algorithms Are Easier to Fix Than Biased People
Nice column by on bias in algorithms. Read to the end for a nice family touch.
None of this is meant to diminish the pitfalls and care needed in fixing algorithmic bias. But compared with the intransigence of human bias, it is a great deal simpler. Discrimination by algorithm can be more readily discovered and more easily fixed.
Biased machines can be *detected* faster and easier than biased people. Biased machines can be also be *fixed* faster and easier than biased people. This is important, often overlooked, and beautifully articulated here by the Man Himself,
Biased algorithms are easier to fix than biased people. With proper regulation, algorithms can help to reduce discrimination. But the key phrase here is โ€œproper regulation,โ€ which we do not currently have
Biased Algorithms Are Easier to Fix Than Biased People via โฆโฉ
โ€˜We must ensure all the necessary inputs to the algorithm, including the data used to test and create, are carefully stored.โ€™ โ€˜We will need a well-funded regulatory agency with highly trained auditors to process this data.โ€™
Biased Algorithms Are Easier to Fix Than Biased People
This is a very interesting discussion by , but how can we get the unbiased people to correct the biased algorithms? And how do we know who are the unbiased people ex-ante?
Humans and algorithms are complements not substitutes
"[B]etter algorithms can help to ensure equitable treatment in our society, though they wonโ€™t resolve the deep, structural bias that continues to plague the United States. Fixing the biases of society is no easier than fixing the biases of people."
This is a smart and thoughtful reflection on countering human vs. algorithmic bias, but what about the relative impact of human vs. algorithmic bias? That is, the rate and scale at which each is likely to harm folks?
Discrimination perpetrated by algorithms is very different from discrimination perpetrated by people
Biased Algorithms Are Easier to Fix Than Biased People: via
Biased Algorithms Are Easier to Fix Than Biased People, By Sendhil Mullainathan /
โ€œIt is much easier to fix a camera that does not register dark skin than to fix a photographer who fails to see dark skinned people.โ€ via โฆโฉ
Good points here. Much work to do on bias. . Biased Algorithms Are Easier to Fix Than Biased People