Online learning platforms have grown tremendously in recent years, having an impact from K-12 to lifelong learning. The LinkedIn Relevance group are applying machine learning in their Recommender System theory applied into practice to the domain of Online Education, using algorithms to provide the most accurate course recommendations for each individual with insights drawn from large scale A/B Testing experiments. Next Tuesday January 23 at the Women in Machine Intelligence Dinner in San Francisco, Shivani Rao, Senior Applied Researcher at LinkedIn will speak to our audience about learning to be relevant and enabling life-long learning through AI. We spoke with Shivani in the run up to the event and asked her some questions about her work in AI as well as her current work at LinkedIn.
I began my work in AI by studying Computer Vision and Image Processing, which is now making a comeback with self-driving car companies. Common problems in this domain included scene understanding, image recognition and so on. This was back in 2004-2006 when I completed my Masters Degree. During the PhD program, I spent most of my time working on AI that dealt with textual data, and for applications like search and text modeling. Here at LinkedIn, my work revolves around Recommender Systems--another subsection of AI.
LinkedIn Learning’s relevance group was formed about 6 months after LinkedIn acquired Lynda.com. Lynda.com was acquired by Linkedin because it aligned with a key part of LinkedIn’s mission: to provide economic opportunity for everyone. What better way is there to do achieve that goal, than through education? If we can help members gain the skills they need for their next big role, their career transitions or just learning to stay relevant (via a rich course catalog) then we are one step closer to our mission of providing economic opportunity to everyone via online learning.
We wanted to integrate the learning experience with the rest of our products, and so LinkedIn Learning was born. My work over the past 2 years has revolved around developing recommendation algorithms to provide relevant course suggestions to learners on the platform.
Recommender systems help narrow down the large catalog of items to present only what is relevant to the user. In case of retail and advertising, the goal is to ensure users see portion of the catalog or ads they find most interesting. In case of learning or online education, the goal of the recommender system is to narrow down the courses to select few courses that are relevant to the learner. This not only helps learners stay engaged with the courses, it also helps acquire new learners via showcasing relevant courses to LinkedIn members, so we can acquire new learners on the platform.
Learners evolve how they learn over time. Until recently online education meant taking an online class that would last a semester and mimic an academic course curriculum. The more recent trend in learning has been that learners engage learning content on their hand held devices on a need-to-know basis. We call this micro-learning and the piece of learning content (often a video of duration < 10 mns) is called micro-content. Faced with this new trend, it became important to identify micro-content and recommend them to the learner. Identifying micro-content (videos) in a course that are self-sufficient nuggets of information is an AI problem in itself. Similarly, to identifying what skills a video can teach a learner we use AI algorithms applied to natural language understanding.
AI in online education is in the nascent stages. Several companies have sprung up in the past decade, and have increased the kind of course content available on the internet today. The course content is so diverse and vast that it caters to all kinds of people all over the world. So anybody that wants to learn has now access to it. These are already transforming education in developing countries. But AI has potential to transform the hiring processes. What is missing right now is a way to evaluate if the learner actually learned a skill from an online course, and what level of skill-expertise did he/she gain by taking the course. This is called as knowledge transfer or learner efficacy. Once we bridge this gap, Online Education will be transformed. The skills-gap will reduce, employment will increase and knowledge workers will experience true power w.r.t their careers. If learners are able to take an online course and get their dream role, then we can say that AI has truly transformed education.
Ethical implications of AI is definitely an omniscient problem to address. It worsened by conversations about eliminating the human component required to make AI happen. It is not autonomous-AI but human assisted-AI that will really mitigate these concerns. For example, although Fingerprint recognition technology has advanced to provide great accuracy, a forensic expert still needs to sign off in order for the results to be admissible evidence in court. I think as long as human is in the loop, we may be able ward off the ethical implications of AI.
That is a great question. Industry projects in AI have several moving parts to them, and it is not all modeling and reading papers and innovation. There is grunt work involved, which includes building and maintaining the infrastructure required to make large-scale AI possible. When several strong researchers and engineers crave for modeling-focused projects, there can definitely be a struggle to get the desired project. Now, add the bias that women face into the mix, and the problem gets that much harder. This is not unique to AI, though; women in other non-AI engineering roles face similar challenges.
The other challenge is promoting women to higher ranks, be it management or as an individual contributor. I have thought a lot about this problem. In my opinion, the solution requires changing the way managers and individual contributors are evaluated. They must be evaluated not only on the basis of the work they do, but also how they mentor and sponsor junior women in AI. If it becomes a part of their promotion package, or their compensation, they will be motivated to help grow a junior engineer/scientist into a powerful individual contributor/manager.
If you are in undergrad and thinking of taking AI, then focus heavily on taking mathematics and probability and statistics classes. For those who are just graduating and joining the workforce in the field of AI, find a mentor who is experienced in AI and learn about what makes it stick, what makes it tick, and what results in impact. For those thinking of moving to AI as a new field, I suggest finding a niche in their own field where AI can be applicable. They are domain experts in their area, and they are best suited to find solutions where AI can contribute.Following the Women in Machine Intelligence Dinner, RE•WORK will be hosting the Deep Learning Summit and AI Assistant Summit in San Francisco next week on January 25-26.