Welcome to the 42nd edition of Hold the Code! As the quarter slows to an end, we will share with you information on AI decision making and the new TensorFlow Lite.Our weekly feature discusses one researcher's mission to bring diversity into the field of AI research.
Happy reading!
Bringing Deep Learning to Android
TensorFlow Lite, a newly developing variation on the current open source machine learning library TensorFlow (both created by Google’s Google Brain Team), is being designed for a new capability: to work specifically and efficiently on android systems. This new framework would not only advance AI capabilities for phones in the short run, but also set up the entire field for a long term shift towards smarter artificial thinking.
What is TensorFlow?
While TensorFlow libraries work for a variety of different types of machine learning, they are especially powerful at what is called deep learning and deep neural networks, a computationally expensive method of machine learning where a computer is given astronomical amounts of data and is told to develop its own pattern matching for it. This type of machine learning is often applied to traditionally very difficult problems like object detection and gesture recognition, both of which are offered as pre-trained models by TensorFlow.
What does this mean for the future of smartphones?
Efficiently implementing these algorithms on smart phones would open a world of opportunities, and TensorFlow Lite goes a step further in their design by maximizing processing efficiency specifically on the types of digital processing chips often seen in phones. This specialization then works both ways, where as AI tools become increasingly important for smartphones, processing chips can then be further designed to optimize their capabilities with these types of computations. Over time, this could hopefully lead to a new generation of AI optimized mobile hardware, further increasing Android phone capabilities in the areas of visual and speech processing, as well as in augmented reality.
To help developers get started, the TensorFlow website contains content such as guided curriculums with recommended courses, books, and videos to help get people started. They also offer a variety of pre-trained models for common problems, eliminating the user’s need for astronomical data sets and computing power.
AI Making Life or Death Decisions
Future Developments
Last year in August, a suicide bomber killed 183 people in Kabul International Airport. When a tragic event like this happens, limited hospital space has to be rationed for hundreds of individuals that need medical assistance, creating difficult decisions to determine who gets care. The current triage process that is used to make these decisions relies on clinicians to go to each soldier to assess the urgency of their needs. This process, according to the head of the medical branch for NATO’s Supreme Allied Command Transformation, is outdated and could use some innovation.
To address this issue, the Defense Advanced Research Projects Agency (DARPA) is looking to innovate the triage system with AI. DARPA has been influential in developing the internet, GPS, weather satellites, and Moderna’s COVID-19 vaccine. A spokesperson from the agency said:
“DARPA envisions a future in which machines are more than just tools. The machines DARPA envisions will function more as colleagues than as tools.”
In 2018, DARPA committed $2 billion to a program dedicated to incorporating AI into over 60 defense projects. Today, through a new program called In the Moment, it wants to develop technology that would make quick decisions in stressful situations using algorithms and data, arguing that removing human biases may save lives.
To accomplish this, algorithms that assist military personnel with small unit injuries and mass casualties will be created and evaluated. Further down the line, algorithms for aid disaster relief may also be developed.
Matt Turek, the leader of the In the Moment program, said that the algorithms will be able to access more information than the most “highly trusted humans”, such as identifying all available resources at a nearby hospital:
“That wouldn’t fit within the brain of a single human decision-maker. Computer algorithms may find solutions that humans can’t.”
Risks of Relying on AI
Despite the ability for AI to access and store more information than a human, some are worried about the ethical implications that it could have when making important decisions. Sally A. Applin, a research fellow and consultant who studies the intersection between people, algorithms and ethics, said in reference to the DARPA program:
“I think it could set a [bad] precedent by which the decision for someone’s life is put in the hands of a machine.”
Additionally, there is a concern that algorithms will be biased in favor of certain groups, like previous algorithms that have seemingly favored white patients over black patients. That being said, it is entirely possible that the human making a decision about soldiers’ care has underlying biases, including racial biases, so to what extent do algorithms discriminate more than people?
Weekly Feature: Building a Diverse AI Pipeline
A Tipped Balance
Dr. Tanya Misha has been an AI researcher for over a decade, yet she still doesn’t see herself represented in this field that she “thinks of as her own”.
The statistics echo what Dr. Misha feels:
Less than 20% of AI professors are women
Only 10% of AI researchers at Google, a leading AI giant, are women
Only 2.5% of Google’s entire workforce is Black
“We want to see more women, we want to see more people of color, we want to see people with different lived experiences, and the problem is that the funnel is not wide enough.”
So, Dr. Misha did something about it – she decided to build a sustainable pipeline bringing underrepresented students into AI. Founding SureStart, a program that engages students from various backgrounds to learn, get mentorship, and build hands-on AI projects, Dr. Misha has since been helping numerous students from underrepresented backgrounds see themselves thrive in the field.
Go, Myth-busters!
A large part of Dr. Misha’s work is centered around myth-busting: No, you don’t need a PhD to work in the field. No, you don’t need to come from a computer science background to break into AI. Yes, you are good enough.
AI is fascinating. Many working in AI find it exciting, not only because of its power but also because they love the challenge. However, we ought to consider how the “challenge”, or sometimes the perceived challenge, can pose barriers like the imposter syndrome for those from underrepresented backgrounds. Diversity in the AI workforce is crucial to companies and societies, and let’s start prioritizing that.
Love HTC? ❤️
Follow RAISO (our parent org) on social media for more updates, discussions, and events!
Instagram: @Raisogram
Twitter: @raisotweets
RAISO Website: https://www.raiso.org
Written by Arielle Michelman, Jake Connell, and Larina Chen.
Edited by Dwayne Morgan