At LinkedIn, artificial intelligence is like ‘oxygen’
Deepak Agarwal is LinkedIn’s Vice President of Engineering and heads relevance and artificial intelligence at the tech firm. (Provided by LinkedIn)
Source: Queenie Wong
By Queenie Wong | email@example.com
PUBLISHED: January 6, 2017 at 7:00 am | UPDATED: January 6, 2017 at 10:55 am
SUNNYVALE — Artificial intelligence might conjure up futuristic images from science fiction, but in social media, it’s already being used to recommend the right jobs, tag photos or display news you’re more likely to care about.
Artificial intelligence might conjure up futuristic images from science fiction, but in social media, it’s already being used to recommend the right jobs, tag photos or display news you’re more likely to care about.
To LinkedIn’s Vice President of Engineering Deepak Agarwal, AI is like oxygen for the business-oriented social network.
“Data is our biggest asset. Without AI and machine learning, you’re not going to be able to surface the right insights to the right users and customers,” Agarwal said.
AI is the science and engineering of creating intelligent machines with the ability to achieve goals like humans do, according to John McCarthy, the computer scientist who coined the term “artificial intelligence” in the 1950s.
Agarwal, the head of AI and relevance at LinkedIn, recently sat down with this newspaper to chat about how the tech company — now owned by Microsoft — is using AI to enhance the careers of the site’s more than 467 million members. The interview has been edited for length and clarity.
Q: What did early AI tools look like when you were first beginning your career?
A: Initially, if you looked at the entire internet space, it was mostly search. How to make search scalable and intelligent, how to retrieve the right document for a given query. I remember when I was in Yahoo, we started this project to revamp Yahoo’s front page. We would have an editorial team who decided which news goes to the front page and where. They would look at some dashboards and see, “Oh, this is not doing so great for the last two hours,” and then take (the story) out. That’s when we decided to do this more automatically through machine learning. We started doing news personalization, and Google was doing similar things. Then finally, Facebook, LinkedIn, they all came along, and the work that was done to make advertising, search and news recommendations great extended to social media companies.
Q: What are some of the ways LinkedIn has been harnessing the power of AI to connect people to the right jobs, or recruiters to great potential hires?
A: We use machine learning for almost all our products: the News Feed, a recruiter searching for the right candidate or connecting the candidate to the right job so they can enhance their career. Recently, we revamped our job-matching algorithms and improved performance by 50 percent. Just to give you a magnitude of the improvement, it leads to 2 million more (job) applications than we used to have. If you’re a passive candidate who’s not looking for a job, we’re careful to only surface jobs that are really good and help you get to the next opportunity. If you’re an active candidate, we sometimes take more risk and show you jobs that may or may not be in the ballpark. From the (profile) descriptions, a software engineer at Google may be very different than a software engineer at a startup. Using all the past data about users and what users look like, we are able to teach machines what are the appropriate jobs for a user.
Q: There have been some public blunders surrounding AI, like Google Photos accidentally tagging African Americans as gorillas or Microsoft’s Tay chatbot spouting racist remarks. How do you know if a machine has learned everything you’ve taught it and anticipate future mistakes?
A: To be honest, we have had our share of blunders. In the past, we have recommended executive assistant jobs to CEOs. The thing to realize about artificial intelligence and machine learning algorithms is they’re not perfect. The question is which errors are more costly than others. When you’re constructing an algorithm, you tell them this error is okay, but not that costly. But if you make this error, then it’s a million times more costly than the other error. The machines will try not to make that error more frequently than the others. The programmer himself has to encapsulate that information when they’re creating that program for the machines to do the right job. When mistakes happen, the programmers fix it, and it doesn’t happen again. But given how complex some of these problems are, some of these errors, at least in this moment, are inevitable.
Q: There’s been a lot of hype around chatbots. LinkedIn recently unveiled a chatbot that can schedule meetings. Is there going to be a LinkedIn office assistant in the future?
A: Our strategy is to use the advances that have been made in speech recognition and other areas and then tailor it to what we’re doing. There are products where we’re trying to create something along those lines to help recruiters.
If you join a new company, there are a lot of things you need to do to get comfortable with a new job. Can we create something like an assistant that helps you onboard to the new company? You’re at LinkedIn and you joined a new company. Let’s say Microsoft. The first thing you might want to do is: If you are already connected with a bunch of folks at Microsoft, you might want to reach out and say hello. Should I schedule a lunch meeting with one of them? So the system is proactively trying to help you. Or maybe here are the people that folks in your company tend to follow on LinkedIn, and you don’t follow them. Here are the articles that are trending in your new company.
Q: Your parent company Microsoft along with Amazon, IBM, Google and Facebook recently joined a partnership on AI to benefit people and society. How does LinkedIn approach some of the ethical and privacy concerns that users might have?
A: LinkedIn has always been a member-first company. We as a business survive because of the trust members have placed on us with their information. We never do anything that violates the terms of service or any of the privacy clauses. Moreover, we take it a step further and we make sure that when we’re building any products, we do it when we know it’s going to add value to our members.
Q: Where do you see AI headed in the next five years for social media?
A: We have come a long way. I think some of the advancements I’m seeing in speech recognition, image recognition, natural language understanding is going to play a big role in where social media would be in terms of machine learning and AI. So if we are able to process images, videos and what people are writing and communicating, almost like humans, that opens new possibilities of what to recommend to the user. So the accuracy of these systems is going to be significantly better, and that is going to change the experience we all have on social media.
Birthplace: Calcutta, India
Position: Vice President of Engineering, LinkedIn
Previous jobs: Scientist at Yahoo! Research (2006-2012), Scientist at AT&T Labs (2001-2005)
Education: Ph.D. in Statistics from the University of Connecticut
Residence: Saratoga, California
Family: Wife (Bharati) and two daughters (Naisha and Maliha)
Five facts about Deepak Agarwal
1. I was born and brought up in Calcutta, India, and moved to the United States in 1998. I am fluent in English, Hindi and Bengali. Like all folks from Calcutta, I love “adda” (intellectual conversations about stuff).
2. I follow every major cricket event. When nothing is going on, I like to watch old cricket classics on YouTube.
3. I love cooking new dishes from leftovers. One of my latest creations is a fusion salad with leftover pasta, leftover spicy Indian potato curry and vegetables. My family loves it, but I’m not sure what the rest of the world would think (I haven’t done that experiment, yet).
4. I love watching Bollywood movies, especially the silly ones.
5. I love listening to music, especially music from the 1960s to the 1990s.