Original article was published on Artificial Intelligence on Medium
Janelle Shane is an optics and artificial intelligence research scientist, as well as the author of the AI Weirdness blog, where she writes about the sometimes hilarious, weird ways that machine learning algorithms get things wrong. She received her PhD in electrical engineering — photonics from UCSD. In this episode, she shares about her current research, her educational background, and her writing endeavors and perspectives on AI.
As only a middle schooler, Janelle became very interested in electrical engineering and optics thanks to her aunt, then an optics professor at The Ohio State University who ran a laser lab. The fun of the lab helped to inspire Janelle’s further academic pursuits in electrical engineering and optics.
Though she always had many interests, Janelle primarily stuck with the field of electrical engineering throughout her education, proceeding to obtain an MPhil in Photonics from the University of St. Andrews as well as a PhD in electrical engineering — photonics from UCSD. During her experience at UCSD, she designed microscopic lasers with the purpose of sending information at faster speeds on computer chips. Her PhD thesis at UCSD involved conducting simulations with “coke can lasers,” where a laser material is encased in a metal shell while being amplified.
Janelle now works at Boulder Nonlinear Systems, an optics company that specializes in non-mechanical beam steering, move or shape light without physically moving parts like a mirror. Janelle explains that the field of optics “covers a really broad bunch of areas” such as physics, chemistry, and electrical engineering. It essentially can be summed up as the science of light. At her company, their current goal is to get quick updating speeds to keep up with processes such as brain activity. Janelle and her collaborators “are using computer generated holograms…to zap individual brain cells in the brain of a mouse” in order to figure out how different brain cells interact with each other. Optics is used to read the signals that come off the brain cells because a lot of them are engineered to fluoresce when activated.
As for how she integrates AI in this research, Janelle describes that “AI is a useful approach if you don’t know much about the problem you’re trying to solve.” For instance, she found AI to be useful when, among very many possible shapes, she needed to identify what shapes might be useful when breaking apart molecules in a particular way. In this instance, AI helped her to recognize the pattern of simply adjusting the power of the laser.
When attempting to apply AI to her projects, however, she often discovered that AI really wasn’t necessary to reach the desired result efficiently. “AI really is, in some cases… an approach of last resort.” One major concern for the problems with AI is that we can’t always tell how the AI got to the answer it did. She goes further to state that “the danger of AI is not that it’s too smart, but that it’s not smart enough.” We shouldn’t rely on AI nor assume it’s perfectly accurate. Especially since AI can and often picks up on human bias and takes advantage of programming loopholes which provide inaccurate or unintended results. Janelle explains that these potential errors or blind spots of AI make it more essential for people to incorporate human judgment and use discretion when designing and using the results of AI models.
On the topic of writing, she originally started her AI weirdness blog to document her electrical engineering material and projects — even when tests failed, the results could still be interesting and cool to document. She then branched out to write about and share funny or weird outcomes of different AI experiments she conducted.
One of her favorite, unexpected examples of this experimentation with AI weirdness is human collaboration with AI to create something silly. For example, she might use AI to come up with strange combinations of words and then human artists will draw those words to result in some humorous drawings.
Her general advice for students interested in AI is to utilize existing resources or programs such as runway ML and get started working with AI in our own time.
Co-written by Emily Zhao