Celebrating AI with PRIDE🌈 + Free AI Events😍
Trending news on Racial bias in AI + How Instagram/Facebook/others use AI + AI 101 Learning Resources (and more!)
Celebrate AI with Pride🌈
Pride, for us, has always meant “not changing to be accepted”. This June, we’ve been supporting our favorite groups by participating in their virtual events and joining the discussions around inclusion and diversity. One of the recurring themes that came up was around bias, such as racism, still prevalent in key AI technologies.
And it’s not that the technology is biased, but we can add in our own unconscious biases in the technology, making it biased.
Trending in AI: Example of Racial Bias in AI
This week saw an example of societal/racial bias in AI come to light when Duke University researchers launched PULSE, a computer vision model that claimed it can generate realistic, high-resolution images of people from a pixelated photo.
For training, it used the Flickr Face HQ data set compiled last year by a team of Nvidia researchers. The same data set was used to create the reconstructed output. It seemed to work fine on white people, but when someone input a dipixelated image of former US President Barack Obama, PULSE upsampled his photo into that of a white man.
Once raised, this started trending on Twitter where many people got the code working locally and tested it out with other ethnicities. You can see below an example of an Asian woman being transformed into a white woman. This pattern was observed across many other minority groups.
And *Surprise, surprise*, Marilyn Monroe (or other white individuals) were still white.
How do AI/ML systems and algorithms get biased?
Bias in Data Collection
This clearly shows an example of racial bias in ML systems caused by unfair data sets that were not based on a representative sample. This is one of the most common ways biases can creep into ML systems.
However, there are also other common ways where our societal biases (or other unconscious biases) can be introduced into a ML system:
Bias in Data Preparation and Preprocessing (e.g. Labelling)
People (Typically engineers) who add features/labels to the data set and design them can be biasedBias in Modelling (Model training, Evaluation and Testing)
The architecture of the model where one makes decisions such as what algorithm(s) should be used, what evaluation metrics (objective function) should be selected, and how the algorithms learns can be biasedModel Deployment and Testing
The way users interacting with a system, or are impacted by it can further reinforce the bias
As the future leaders and consumers of AI/ML, It’s important to remember that change has to start somewhere.
We must seek to understand if our technology could be used to marginalize groups of people and take action to prevent it.
So stand up for what you believe in and say it with pride!Free AI 101 Resources:
🧠A introductory course on Artificial Intelligence and Machine Learning This online AI/ML course covers the necessary concepts from scratch
📕Learn Deep Learning from scratch Deep Learning sounds overwhelming to a lot of folks. The only prerequisite for this course is high school level math
📚Springer has released 65 Machine Learning and Data books for free
Hundreds of popular books are now free to download💻 70+ Data sets for Data Science and Machine Learning projects:
A lot of you have asked for project ideas. Use these to build your next ML project💡Foundations of Machine Learning course offered by Bloomberg
AI in the World Around You
🤦Instagram DOES prioritise ‘sexy selfies’ in your news feed with a “soft porn algorithm”, study reveals
📽AI in film and TV An Australian screen agency is developing a movie and TV show using artificial intelligence
🙏Google’s Guide on using AI/ML for social good aimed at helping nonprofits and social enterprises apply AI to social, humanitarian and environmental challenges.
👤How Does Facebook Use Machine Learning to Deliver Ads? Facebook provides an official answer to the question you’ve been asking yourself for a long time
🤔Our weird behavior during the pandemic is messing with AI models — forcing humans to step in to set the models straight.
Free AI Events and Career Resources
🎉Featured Event: Registration is now open for HobbyHacks, on July 25-26th
TechTogether (Boston’s largest female and non-binary hackathon) is now virtual, and open to people around the world! Collaborate with a team to design, build, and pitch a project in just 48 hours. With over 20+ workshops and 24/7 access to career & technical mentorship, this is an event you won’t want to miss.
Visit hobbyhacks.techtogether.io to learn more and register.
🤗Call for Code is not your average hackathon! Take on impactful problems such as battling COVID-19 and addressing climate change (among many more). They’ve got Starter Kits for beginners and big $$$, opportunities for winners.
📈Back to Business with AI (7/14) is a free, live, virtual event for business professionals covering key AI areas such as autonomous vehicles, computer vision, smart manufacturing, NLP, etc.
🎨Art-A-Hack 2020 has a challenge around the COVID-19 “DANCEDEMIC” based on reimagining interactive live performances in new ways. Think wearable sensors + biometric data + entertainment = Fun learning experience
🤷♀️What’s the difference between the data scientist and data analyst roles? A great visual breakdown between the two roles (this is a FAQ we receive!)
🔎What does a typical AI/ML project look like? a great article covering a typical AI/ML project structure, with stages, roles, and tools
As always, thank you for reading! Through this newsletter we bring you a curated list of the latest and greatest (and most fun) updates in the world of AI and ML. We also share resources and events for you to dive deeper.
If you have a link you’d like us to include in the next newsletter, let us know.
Have a question or feedback for us? Want to join AI for Her’s mission? Want to share your journey in AI/ML with others to inspire them to get started? Find out ways you can get involved.
Stay safe, and stay strong!
Yours,
The AI for Her team