Below is the fifth installment in our series, AIML: Big Changes, Big Conversations. Taking AIML (Artificial Intelligence and Machine Learning) as our example, we explore methods for accelerating and amplifying conversations around major changes in ways that enable us, as individuals and as a collective, to understand and direct their impact.
Katherine Gorman is a Podcast Producer at Collective Next and one of the founders of the popular show Talking Machines, which she co-hosts with former Harvard Professor Ryan Adams. While not a scientist herself, Katherine has long been fascinated by the topic of Artificial Intelligence and Machine Learning (AIML) and wanted to create a forum for conversing with experts in the field. One of her goals has been to broaden the conversation about AIML and the show is notable for having “human conversation” about this highly technical material. We sat down with Katherine to discuss the ways in which Talking Machines catalyzes conversation among AIML experts and how non-experts can meaningfully engage this significant topic.
Marsha Dunn: You are in your second season of Talking Machines. How did the show begin and what was your goal in developing the podcast?
Katherine Gorman: A great deal of fear and excitement surrounds AIML, but it doesn’t necessarily reflect the research really going on. One of our goals has been to inject scientifically based-information into the public understanding. We want to help pivot the public discourse away from hyperbole towards fact-based conversations. In addition, we felt that illuminating the lives and passions of AIML researchers would bring the conversation to life.
Marsha: Your tagline is “Human Conversation About Machine Learning.” Can you elaborate?
Katherine: Ultimately, AIML is about developing tools intended for human use. So we need to have human conversations about this technology; conversations in which everyone can engage. We all need a working knowledge of what AIML is and what it means for our future. This technology is changing our lives—shaping how information is presented to us and how we make decisions. However, despite the magnitude of its impact, we need to avoid discussing AIML in terms that imply it is magical or larger than life. We need to keep our conversations grounded in human experience.
Marsha: One of the central fears people have about AI is that machines will one day overtake humans. How do we value what is unique about being human while embracing the benefits of AIML?
Katherine: We need to recognize that machines are good at the things humans aren’t good at and humans are good at things machines cannot do. Machines are good at small repetitive tasks and consuming vast quantities of information. Humans excel at creative thought and extrapolation. It would take a lot of forward movement in the field for machines to begin to touch these human capacities.
In fact, the field is currently focused on getting machines to do things that 2-year olds do with ease. Andrew Ng, Chief Scientist at Baidu, has a fitting response to concerns of a machine takeover: “worrying about the rise of evil killer robots is like worrying about overpopulation and pollution on Mars before we’ve even set foot on it.” In other words, it is all hypothetical and distracts us from solving the problems right in front of us. Right now we need to address questions like: What does data privacy mean? What does it mean to give someone your data? What does it mean for someone else to make decisions for you based on this data?
Marsha: As co-host of Talking Machines you interview many of the biggest names in AIML. How does the podcast help to catalyze conversation among professionals in the field?
Katherine: One way is by promoting faster communication among researchers—a critical need in the field. Developments in AIML have outpaced the inherited models of communicating research in academia (e.g. publication in scientific journals).
Cornell’s website arXiv has responded to the need to accelerate the publishing process in the sciences and technology by allowing researchers to post findings immediately. However, while this modality promotes speed it also creates a firehouse of un-intermediated information. With the Talking Machines podcast, we can respond rapidly to new developments while also curating and contextualizing the content.
Marsha: You have also mentioned that by discussing the passions and motivations animating the work of senior researchers, you provide a sort of mentorship function.
Katherine: We discovered that illuminating the life’s course of those at the top of the field provides inspiration and direction for those still early on in their careers.
Marsha: You speak to experts in AIML with the goal of identifying some of the big questions in the field. What are the most significant questions non-experts should be asking about AIML?
Katherine: How do these machines and tools already function in our daily lives? How do they help us make decisions from Facebook friend suggestions to Amazon book recommendations? Non-experts should look out on the world with an eye towards understanding how AIML is already changing our thinking or understanding.
Marsha: Discussions about AIML can get very technical very fast, what are some of the ways experts and non-experts can engage in meaningful, interdisciplinary conversation on this topic?
Katherine: The conversation seems highly technical, but researchers approach their work from a very human place. For example, a researcher solving a problem in machine learning begins by asking, “How do children learn about the world?” This is work being done by humans and thus is explorable by everyone at a certain level.
People are often intimidated by science and technology because they employ language most of us are not fluent in. However, if you sit down with someone who is fluent in this language they can distill the discussion down to simple core principles. What is more, the AIML community is excited to make their work accessible to the larger population. I encourage people to ask questions, to be open to trying to understand this evolving field. On Talking Machines, we take questions from listeners, many who are looking to understand fundamental terms and concepts.
Marsha: Can you give an example of a term in AIML that might throw us non-experts for a loop, but which actually represents an accessible concept?
Katherine: The term “topic modeling” sounds intimidating, but simply refers to a sorting process that employs key words. For example, if you have a huge corpus of data, let’s say all of the articles published by the NY Times news from 1965 to the present, topic modeling allows you to use key words to categorize articles according to similarities and differences.
Marsha: Before you leave us, what are the most exciting things you have learned about the future of AIML through hosting the show?
Katherine: First of all, I find science reporting, particularly in the field of AIML, to be akin to watching the Olympics. You get to witness people engaged in activities that you have the ability to perform on some level, but which their skill and training allows them to elevate to an art form, and that is riveting.
One of the most profound things I have learned in covering AIML is that it offers us the opportunity to build the world we want as opposed to the world we currently have. Let me explain. In order to train machines you need training data. However, our real-world data is full of implicit bias stemming from humanity’s prejudices and limitations. As we help machines learn we can compensate for these biases; we can provide machines with data and models that reflect our better aspirations.
[Photo: Katherine Gorman and Ryan Adams interview Andrew Ng via live video at TEDxBoston 2016 - Flickr, TEDxBoston]