Daniel Angelov is a Ph.D. student at the University of Edinburgh, Scotland, in the Robust Autonomy and Decisions (RAD) Lab, focusing on the intersection between Robotics and Machine Learning. He received his undergraduate and master’s degrees from the University of Reading. Daniel has participated in several Google Summer of Code programs and is an avid mountaineer with a passion for the skies.
Why did you decide to do your internship at PARC?
In my work on robotics and machine learning, I’ve realized that robotics is less about the robots and more about understanding how humans work, emulating their lives and plainly trying to figure them out. I thought I’d have a good opportunity to explore that further as a PARC Intern in the Machine Learning group.
Can you tell us more about the project you’ve been working on?
I’ve been involved mainly in the COGLE (COmmon Ground Learning and Explanation) project, which is a project under PARC’s explainable AI (XAI) DARPA program. As more and more machine learning systems come into existence, people naturally start to question how these “black boxes” decide the correct answer. My work involves creating agents that not only learn how to do a specific task, but are capable of giving insight about the particular reason behind their actions.
How do you hope your project could impact the world?
Long term, I think explainability is going to help not only verify and confirm the correctness of artificial intelligence systems (we all want our cars to drive safely) and extend the interaction between AI and humans, but also contribute back knowledge and transparency to the human experts. We are currently bound by the length of our existence, spending a large part of it on obtaining knowledge, but machine learning systems can absorb all that people have contributed so far and, maybe by explaining their reasoning, we will gain helpful insight for the next generation of human experts.
How are your studies related to this project?
Well, I’m doing my Ph.D. in Robotics and Autonomous Systems at the University of Edinburgh and I’m trying to teach robots basic concepts about the physical world, how to understand it and change it in a way to fulfill its goals. By giving the robot the ability to imagine or dream of possible sequences of things to do, or by observing how humans perform specific tasks, it can learn some causal structure about the world that we as humans use so intuitively in our lives.
What have you enjoyed most about interning at PARC?
I think people are the most important resource a company can have, and at PARC I have definitely met some great researchers and mentors and fellow interns and friends. As a researcher and entrepreneur at heart, I have always felt that the right place would allow you to both work on state-of-the-art innovations and apply it to an actual customer-facing product. And seeing that happening in front of my eyes, while also having fun, picnics and a good time, is incredible!
What are you reading at the moment?
After my daily brief of ArXiv papers with a morning cup of tea (it seems all Ph.D. lives come down to this), I usually spend a few months concentrating on a specific topic. I recently decided to interrupt my current batch of logic, maths and reasoning and spend some time reading stories from professionals in other areas. Right now it’s “Alone on the Wall” by Alex Honnold, and hopefully I’ll gain some insight into his process of thinking.
If you could invent anything, what would it be?
A machine that can ask the right questions at the right time. If we assume knowledge is a non-stationary process, and there is a positive trend (as we are observing), humanity will eventually invent anything we can think of, and having it by default is kind of boring. So, we are limited by our known knowns and known unknowns. But if we get an oracle, just like Clippy the old Office assistant, it would not only speed up the process, but also allow us to explore those areas of knowledge that would otherwise remain “dark” due to a lack of evidence or enquiry. If we ignore the possibility of annoyance, people will get these magnificent moments of revelation, not only because they got an answer, but because they were nudged to explore in the right direction.
Time travel or teleportation?
If I could bring Clippy, definitely teleportation!