The Truth About Algorithms | Cathy O'Neil
ฝัง
- เผยแพร่เมื่อ 6 ก.พ. 2025
- We live in the age of the algorithm - mathematical models are sorting our job applications, curating our online worlds, influencing our elections, and even deciding whether or not we should go to prison. But how much do we really know about them? Former Wall St quant, Cathy O'Neil, exposes the reality behind the AI, and explains how algorithms are just as prone to bias and discrimination as the humans who program them.
Speaker: Cathy O'Neil (weaponsofmathd...)
Animator: Niceshit Studio (niceshit.tv/)
Intro and outro animation: Cabeza Patata (www.cabezapata...)
Producer: Abi Stephenson
Extracted from a free talk given at the RSA in London, 2017.
Subscribe to our channel for more!
Follow the RSA on Twitter: / rsaevents
Like RSA Events on Facebook: / rsaeventsoff. .
Listen to RSA podcasts: / the_rsa
See RSA Events behind the scenes: / rsa_events
Algorithms are systematic ways to solve problems. What they bring is consistency, which can mean being consistently wrong.
Exactly! Without making sure that we account for bias, algorithms will end up compounding that bias!
This should be required viewing for anyone who uses algorithms. Thank you.
youtube thinks they can calculate what content I want, but instead I end up bored and feeling empty inside. I found this video by remembering the channel from a previous experience years ago and searching it up manually to feed my curiosity. my curiosity is not nourished and openness is being hampered with. the sad part is that they dont even realize that it is ultimately bad for business aswell, in the long run.
"All models are wrong, but some are useful." George Box
As an artist I use algorithms to draw trees, mountains, clouds, ect. A series of sensations that are constantly updating and referencing what is being put down on paper. Make a scribble then if I want it to be a cloud the next set of scribbles will be what i programmed my kinesthetic sense to be what i define as "cloud-like" (which are circular puffy shapes) then i move these groups of scribbles around randomly varying in sizes. You can do this with anything that has a fractal like structure
That's nice
Very good video! I love the content and the animations! Made me want to subscribe and watch all the videos on this channel.
Loved ur book Cathy. 5/5 on goodreads. Salutory incisive yet accessible work.
There is a very important line at the end that is clearly skipped in the audio but that shows in the subtitles: (0:22) "THEY ARE INHERENTLY PICKING UP WHATEVER BIAS WE'VE GIVEN THEM". 'Bias' is right. Why we can't hear that line but only read it? Just curious.
Great video - thank you very much!
the animation is phenomenal!
I'd love to see honesty in the sources to cite for further research, but in reality, the vetting & certification seems to be lost in who I can trust.
I thiught so!! Thank U
Well said 🙌🏾
Guys, I have two questions:
1.) - can an algorithm be built on the perfect past data? Perhaps that can't be possibly done because everyone of us has different of opinions and different definition of what's perfect.
2.) - can an algorithm update its database by itself and get better and better with time?
I've searched all the music stores and can't seem to find any of these Al Gore Rhythms
I prefer the traditional definition of an algorithm as a definite transformation that succeeds when it reaches its purpose in a finite number of steps. Your family judges the meal not by the algorithm that produces it, but by the result. Algorithms only work on computable problems, a class of problems that excludes many of the market decisions. The deception is not in the algorithms but in the pretence that they can do things beyond their limited capability.
Good point. For pseudo-algorithms that work to get "some answer" for things that really are outside the realm of properly defined computable problems, do you have a preferred term? Back in the day we would say "heuristic" or "heuristic algorithm" or "fuzzy logic" these weren't exactly the same, but I'm not sure what term to use. Many people are calling whatever they use to come up with some kind of answer for things really outside the realm of properly defined computable problems are just calling them "algorithms" in practice, no?
she is a genius
The truth about algorithms is that they may illuminate commonalities the ignorance of those looking at them do not allow for individuality and anomalies.
They can b useful although not 100% accurate.
This is quite insightful. Should you crave more information, a suitable book is recommended. "Game Theory and the Pursuit of Algorithmic Fairness" by Jack Frostwell
Thanks
👍 thanks
Look at the directing studio's name. Hah nice.
Very good, thanks! =8)-DX
Did your parents know you would grow up so smart. There must be corporations that see you as the enemy. I see stuff on Quora, LinkedIn that is so biased and ignorant that it really winds me up. A classic was a company that claimed it had cracked sarcasm detection in semantics. It even put the code on github. There was a multitude of comments that said how wonderful it was and nothing when I replied that solving such a culturally complicated question was a joke using a simple algorithm. Give me strength.
"There was a multitude of comments that said how wonderful it was" - did their algorithm detect the sarcasm? :-)
No, it was on vacation that week.
This information was absolutely beautiful! I recognize how propagandized data can be misleading.
How honest! I used to think I got wrong impression that some algorithm were made on purpose for their rationalization. I didn’t mean this is bad, but I just want to say we have to keep the ability to recognize what kind of algorithm we need. And of course, the animation is wonderful as well.
I do not think the issue is an algorithm bur rather its use and interpretation. The big issue nowadays is people drawing the wrong conclusions from particular algorithms, either because they are not relevant for the task or because their results are wrongly understood. THIS is the big problem, there is nothing wrong with the algorithms themselves!
the algorithm hides the labor and opinions behind it so in a sense, yes, it is the mystification included with the construct of the algorithm that is the issue
@@blindlight what alghoritms? All alghorithms or just some of them? Maybe you start use more precise alghorithm to express your thoughts...
@@nonseans its not the algorithms its the mystification/anthropomorphisation of them that's the issue here. "Clever AI outwits creator" vs "creator leaves imprecise instructions" the problem is always a human one.
@@blindlight I think she definitely misuses term to make bold statements. I suppose we all know that data is interpreted by humans and there are a lot of traps hidden that sometimes we are not even aware of. But stating that alghoritms emeed human intepretations is just BS...
@@nonseans i think shes saying they embody human judgements not that they *need* them
Who else was sent here for ALR?
hmmm
had to make it 50 comments sorry
I couldn't leave it at 49
Maneiro😂😂😂😂😂😂😂😂😂😂
P.S. How long have you been an ilk? Do you meet at special ilk conferences and talk about ilky things 😂
I AM THEREFORE I Think 💔 I THINK Therefore Am I 💭OOOoowwwwwww Mmmmmmm🙏❤❤❤ 💭⚔️⚔️⚔️💭💔👀
As an Independent I can easily see that the filtering algorithms in social media are Very biased against conservative voices.
Can I ask the basis for suggestion that algorithms are fair on it's own? I never heard that. I feel like someone just invented a wheel... To say something is fair or not first you have to define fairness... but that opens the whole different cosmos I don't want to step into so I stop here.
Nonsense.
An algorithm, in everyday language, is a precise recipe for processing data. It is certainly possible to create an algorithm that produces flawed, or even totally incorrect results, but that is the fault of the designer. To oversimplify and claim that all algorithm are “opinions” is just as silly as reading a dessert cookbook and then claiming that all recipes are fattening.
False. The opinion is the definition of success. In the video, she gave her example - in her opinion, her kids eating healthy meals is "success". Her son, on the other hand, would define a successful meal as one entirely consisting of dessert. These definitions of success - these opinions - are baked into the models they would independently produce. Blindly accepting either model without knowing the definition of success of the person who created it can lead to wildly terrible results.
And even the algorithm can contain a bunch of opinions (magic numbers, statistical skews, probably others and more recently, training set choices).
You are wrong on this. The filtering algorithms in social media show a definite bias. They are Not impartial.
in your case “fattening” is something your algorithm is built to avoid. that’s an intention - and an opinion
Joel Neely, she appears to be a data opinionist, not a data scientist.
flawed belief that veg are good #meatheals
How honest! I used to think I got wrong impression that some algorithm were made on purpose for their rationalization. I didn’t mean this is bad, but I just want to say we have to keep the ability to recognize what kind of algorithm we need. And of course, the animation is wonderful as well.