The journalists created a fake job posting on two AI interview platforms. They specified the traits of the ideal candidate and provided the questions that need to be answered during the interview.
Then they started experimenting... π
AI Job Interviews - another good example of bias in ML π€¦ββοΈ
Two journalists tested some AI tools for assessing job candidates. Even when the candidate read a Wiki article in German instead of answering questions in English, the AI systems gave them good scores π€·ββοΈ
Let's unpack π
The journalists created a fake job posting on two AI interview platforms. They specified the traits of the ideal candidate and provided the questions that need to be answered during the interview.
Then they started experimenting... π
One of them did a fake interview giving all the right answers and predictably got very high scores - 8.5 out of 9 π
Then she tried something different... π
In a second interview, instead of answering the questions in English, she just read the article on psychometrics from the German Wikipedia π
One system gave her a score of 6 out of 9, while the other determined she is a 73% match for the job.
Oops... π
Interestingly, one of the systems generated a transcript which was obviously meaningless.
This means that the machine learning model behind the tool likely captured nuances of the intonation of the speaker instead of the meaning of the actual words.
π
These tools are biased - they incorrectly put more weight on the intonation instead of the actual meaning of what is being said.
This is shown when one of the components is artificially removed - in this case, the meaning (German instead of English).
More examples π
Another AI interview tool was reported to give candidates different ratings when they used different backgrounds or accessories during the interviews.
Why does this happen? π
One likely reason is the dataset used to train the machine learning models. It may not contain enough data for the model to correctly learn to discriminate important features. It overfits on some non-essential features, like the background.
π
Another common problem is the use of black-box models where it is difficult to interpret why the algorithm makes a certain decision (for example in the case of neural networks).
More work is needed to inspect what the model is paying attention to.
π
Furthermore, it seems that the systems weren't tested in scenarios like this one and weren't prepared to meaningfully handle these cases. The question is then, what other cases will not be handled correctly?
π
Interestingly, human interviews are also subject to many biases and this is a known problem. The hope is that AI systems will help reduce those biases, but one needs to be careful to not just replace them with different ones.
For more details, you can read the full article by @HilkeSchellmann and @sheridanlwall on MIT Technology Review here:
https://t.co/uHAlrA2vKw
Follow me @haltakov for more!
https://t.co/Vy73ydcp61
The legend @Kasparov63 once said that the best outcome occurs when technology and humans work together and not as a replacement
— Guzman Ojero (@GuzaUy) July 20, 2021
https://t.co/7hBveNTpHb https://t.co/gZkJ6zol9W
More from Vladimir Haltakov
Let's talk about a common problem in ML - imbalanced data βοΈ
Imagine we want to detect all pixels belonging to a traffic light from a self-driving car's camera. We train a model with 99.88% performance. Pretty cool, right?
Actually, this model is useless β
Let me explain π
The problem is the data is severely imbalanced - the ratio between traffic light pixels and background pixels is 800:1.
If we don't take any measures, our model will learn to classify each pixel as background giving us 99.88% accuracy. But it's useless!
What can we do? π
Let me tell you about 3 ways of dealing with imbalanced data:
βͺοΈ Choose the right evaluation metric
βͺοΈ Undersampling your dataset
βͺοΈ Oversampling your dataset
βͺοΈ Adapting the loss
Let's dive in π
1οΈβ£ Evaluation metrics
Looking at the overall accuracy is a very bad idea when dealing with imbalanced data. There are other measures that are much better suited:
βͺοΈ Precision
βͺοΈ Recall
βͺοΈ F1 score
I wrote a whole thread on
2οΈβ£ Undersampling
The idea is to throw away samples of the overrepresented classes.
One way to do this is to randomly throw away samples. However, ideally, we want to make sure we are only throwing away samples that look similar.
Here is a strategy to achieve that π
Imagine we want to detect all pixels belonging to a traffic light from a self-driving car's camera. We train a model with 99.88% performance. Pretty cool, right?
Actually, this model is useless β
Let me explain π
The problem is the data is severely imbalanced - the ratio between traffic light pixels and background pixels is 800:1.
If we don't take any measures, our model will learn to classify each pixel as background giving us 99.88% accuracy. But it's useless!
What can we do? π
Let me tell you about 3 ways of dealing with imbalanced data:
βͺοΈ Choose the right evaluation metric
βͺοΈ Undersampling your dataset
βͺοΈ Oversampling your dataset
βͺοΈ Adapting the loss
Let's dive in π
1οΈβ£ Evaluation metrics
Looking at the overall accuracy is a very bad idea when dealing with imbalanced data. There are other measures that are much better suited:
βͺοΈ Precision
βͺοΈ Recall
βͺοΈ F1 score
I wrote a whole thread on
How to evaluate your ML model? \U0001f4cf
— Vladimir Haltakov (@haltakov) August 31, 2021
Your accuracy is 97%, so this is pretty good, right? Right? No! \u274c
Just looking at the model accuracy is not enough. Let me tell you about some other metrics:
\u25aa\ufe0f Recall
\u25aa\ufe0f Precision
\u25aa\ufe0f F1 score
\u25aa\ufe0f Confusion matrix
Let's start \U0001f447
2οΈβ£ Undersampling
The idea is to throw away samples of the overrepresented classes.
One way to do this is to randomly throw away samples. However, ideally, we want to make sure we are only throwing away samples that look similar.
Here is a strategy to achieve that π
Machine Learning Paper Reviews ππ
Check out this thread for short reviews of some interesting Machine Learning and Computer Vision papers. I explain the basic ideas and main takeaways of each paper in a Twitter thread.
π I'm adding new reviews all the time! π
AlexNet - the paper that started the deep learning revolution in Computer Vision!
DenseNet - reducing the size and complexity of CNNs by adding dense connections between layers.
Playing for data - generating synthetic GT from a video game (GTA V) and using it to improving semantic segmentation models.
Transformers for image recognition - a new paper with the potential to replace convolutions with a transformer.
Check out this thread for short reviews of some interesting Machine Learning and Computer Vision papers. I explain the basic ideas and main takeaways of each paper in a Twitter thread.
π I'm adding new reviews all the time! π
AlexNet - the paper that started the deep learning revolution in Computer Vision!
It's finally time for some paper review! \U0001f4dc\U0001f50d\U0001f9d0
— Vladimir Haltakov (@haltakov) September 28, 2020
I promised the other day to start posting threads with summaries of papers that had a big impact on the field of ML and CV.
Here is the first one - the AlexNet paper! pic.twitter.com/QNLPIMZSIa
DenseNet - reducing the size and complexity of CNNs by adding dense connections between layers.
ML paper review time - DenseNet! \U0001f578\ufe0f
— Vladimir Haltakov (@haltakov) October 15, 2020
This paper won the Best Paper Award at the 2017 Conference on Computer Vision and Pattern Recognition (CVPR) - the best conference for computer vision problems.
It introduces a new CNN architecture where the layers are densely connected. pic.twitter.com/DuHytaoXia
Playing for data - generating synthetic GT from a video game (GTA V) and using it to improving semantic segmentation models.
Time for another ML paper review - generating synthetic ground truth data from video games! \U0001f3ae
— Vladimir Haltakov (@haltakov) October 5, 2020
I love this paper, because it pushes the boundaries of creating realistic synthetic ground truth data and shows that you can use it for training and improve your model.
Details \U0001f447 pic.twitter.com/fBgORYG8Lz
Transformers for image recognition - a new paper with the potential to replace convolutions with a transformer.
Another paper review, but a little different this time... \U0001f937\u200d\u2642\ufe0f
— Vladimir Haltakov (@haltakov) October 5, 2020
The paper is not published yet, but is submitted for review at ICLR 2021. It is getting a lot of attention from the CV/ML community, though, and many speculate that it is the end of CNNs... \U0001f447https://t.co/bh6wUxYfxu pic.twitter.com/dZGBYB8A5U
More from All
You May Also Like
First thread of the year because I have time during MCO. As requested, a thread on the gods and spirits of Malay folk religion. Some are indigenous, some are of Indian origin, some have Islamic
Before I begin, it might be worth explaining the Malay conception of the spirit world. At its deepest level, Malay religious belief is animist. All living beings and even certain objects are said to have a soul. Natural phenomena are either controlled by or personified as spirits
Although these beings had to be respected, not all of them were powerful enough to be considered gods. Offerings would be made to the spirits that had greater influence on human life. Spells and incantations would invoke their
Two known examples of such elemental spirits that had god-like status are Raja Angin (king of the wind) and Mambang Tali Arus (spirit of river currents). There were undoubtedly many more which have been lost to time
Contact with ancient India brought the influence of Hinduism and Buddhism to SEA. What we now call Hinduism similarly developed in India out of native animism and the more formal Vedic tradition. This can be seen in the multitude of sacred animals and location-specific Hindu gods
i wonder if you can make a thread bout witchcraft in malaysia.. or list of our own local gods/deites..
— r a y a \U0001f319 (@lcvelylilith) February 20, 2020
Before I begin, it might be worth explaining the Malay conception of the spirit world. At its deepest level, Malay religious belief is animist. All living beings and even certain objects are said to have a soul. Natural phenomena are either controlled by or personified as spirits
Although these beings had to be respected, not all of them were powerful enough to be considered gods. Offerings would be made to the spirits that had greater influence on human life. Spells and incantations would invoke their
Animist ceremonies of a religious or magical nature were normally held for the purpose of divination or making a request. This would either be done at a keramat or at a shrine similar to the Thai spirit houses or Chinese roadside shrines pic.twitter.com/I1hliyi0x3
— \u2745\u1710\u170b\u1713\u170e (@uglyluhan) June 16, 2019
Two known examples of such elemental spirits that had god-like status are Raja Angin (king of the wind) and Mambang Tali Arus (spirit of river currents). There were undoubtedly many more which have been lost to time
Contact with ancient India brought the influence of Hinduism and Buddhism to SEA. What we now call Hinduism similarly developed in India out of native animism and the more formal Vedic tradition. This can be seen in the multitude of sacred animals and location-specific Hindu gods
