SciArt Spotlight: Janelle Shane
Bouncing butterstrange bars! We have AI humorist Janelle Shane on the blog today, and things are about to get wonderfully weird! Janelle runs the extremely popular, interesting, and often hilarious blog AI Weirdness where she posts the results of her experiments training neural networks.
So how does training neural networks work? Simply put, Janelle "feeds" a neural network a dataset (a bunch of real-world examples of a certain type of thing), the neural network learns from the dataset, and then it produces what it "thinks" are appropriate additions to the dataset. The outcomes are often priceless, and I personally don't think I'll ever come up with a better name for a guinea pig than "Fuzzable." Simply outstanding. At Style Engineers Worldwide we love it when a person's work in #SciArt really shows us the amazing and unexpected magic that occurs at the intersection of technology and creativity, and Janelle's results on AI Weirdness do precisely that!
In case working with artificial intelligence wasn't cool enough by itself, Janelle is also a research scientist working in optics — heck yeah, lasers! (She runs AI Weirdness as perhaps the world's most amazing hobby.) Janelle has a PhD in Electric Engineering – Photonics, has been featured by outlets including Gizmodo, The Atlantic, New York Magazine, and TED Talks chose her as a speaker for their TED2019 Bigger Than Us series.
We love that Janelle is making artificial intelligence accessible through her particular brand of #SciArt, and we were thrilled to ask her all about AI, her upcoming book which we can't wait to read, what advice she has for people who want to play with neural networks, too, and more! Meet Janelle Shane.
When did you know you were a scientist?
There were early signs. I have a bar graph that I made as a very small child, quantifying and comparing the attributes of several My Little Ponies.
Can you describe your work as a research scientist working in optics?
I get to design equipment that scientists use to study the brain, or that engineers use to help self-driving cars see.
What’s the weirdest/most fun experiment you’ve ever done?
I got to do a project once where I designed a virtual reality arena for mantis shrimp. These shrimp have amazing eyesight, and we were looking at ways to build a theater for them where we could test what their eyes could actually detect.
How did you end up embracing artificial intelligence as a hobbyist?
I saw an experiment where someone trained a neural net to generate new cookbook recipes, and I thought they were hilarious. Shredded bourbon, peeled rosemary — I wanted to learn how to generate them too!
What delights you most about the nature of AI -- in that it can be a tool for automation and efficiency, and simultaneously a hilarious mess where computers get things wrong to a ludicrous extent?
I like the sheer unexpected weirdness that AI can come up with. It has so much in common with the wild things that you see in nature.
What do you enjoy most about sharing your super amusing neural network results with the world on AI Weirdness?
I love it when people embrace the wacky stuff the neural net comes up with. People have invented recipes for the neural net's cookies, knitted its knitting patterns, and even turned its paint colors into real nail polishes.
Should we be worried about AI eventually taking over humanity and enslaving us?
We may never have to worry about this - it's really hard to build AI that can do simple tasks, and even then it reveals just how little it understands - like a cat-generating AI that can do realistic eyes and fur but doesn't reliably generate the correct number of legs. I like AI researcher Andrew Ng's take on it: worrying about an AI takeover is like worrying about overcrowding on Mars.
Is there anything we can learn from machine learning that could make us better humans?
Since so many AIs learn by copying us humans, in a way it's holding up a mirror. Some of the stuff that it's learning from us is stuff we don't want it to learn — like bias. We can learn a lot from what AI learns from us.
Can you tell us a little bit about your forthcoming book: You Look Like a Thing and I Love You: How AI Works and Why It's Making the World a Weirder Place?
It will have cartoons and jokes! It'll be a really fun look at what AI actually is today, and all the strange stuff it does.
Can you give us a bit of a preview of your upcoming TED Talk?
I'll be talking about the disconnect between what we want AI to do, versus what we actually ask AI to do.
When not working with lasers, what are some of your non-neural-network-related hobbies you enjoy in your free time?
I love to hike in Colorado! I also play the Irish flute.
Which scientists (modern day and/or historical) inspire you?
I'm really inspired by Rachel Thomas, who co-founded fast.ai and is bringing free online machine learning courses to anyone who wants to take them. It's a tremendous resource that is doing so much for AI, and so much to make it accessible.
You have some great info and resources on your FAQ page for people who want to get started playing with neural networks. Do you have additional advice for young people interested in pursuing machine intelligence research as hobby or maybe even career?
Pick a project that inspires you, and start working on it! There are so many great free resources (like fast.ai) and even free computing resources. There's nothing like the experience and recognition you get from doing a hands-on project.
If people want to send you a dataset for future neural network training exercises, what should they consider before getting started?
Try out some of the resources on my FAQ page — you can train your very own neural network on the data of your choice! The more data you have, and the narrower the kind of data you're training on, the better your results will be.
Do you have any favorite fictional scientists and/or AI characters in books/movies/other art forms?
There have been some really great AI characters in the last couple of years. Two of my favorites are Murderbot (from Martha Wells's Murderbot Diaries books) and Breq (from Ann Leckie's Ancillary Justice series).
Piggybacking off of that: What do you think the media/Hollywood gets most wrong about AI? Most right?
Hollywood AI is much different from real-world AI. In science fiction, AI is almost always general AI, which is at least as intelligent as a human and can do lots of different things. In real life, what we have is narrow AI, which can really only do one thing well at a time. As a result, people tend to be more worried that AI is too smart, when instead we should be worried that it's not nearly as smart and context-aware as people tend to assume.
If you were a superhero, what would your go-to wearable tech device be?
I would like a bionic eye that can auto-label stuff for me. I would primarily use it for identifying birds.