Learning

The AI Academy: How We’re Scaling AI Across LinkedIn

AI-Academy-Image

At LinkedIn, we like to say that artificial intelligence (AI) is like oxygen—it’s present in every product that we build and in every experience on our platform. From features like People You May Know to the recommendations for LinkedIn Learning courses, AI helps deliver value to our members across their entire career journey by personalizing their experience. And AI’s impact on the technology we design—and even the process of how we design that technology—is only continuing to grow. As we constantly look to improve LinkedIn, we’re finding more and more ways in which AI can bring additional value to our members and customers.

Another way to say this is that AI is eating software in the same way that software ate the world. But while modern computing resources have enabled an exponential increase in the amount of AI we can implement, the number of people with expertise in AI has not scaled at the same rate. Today, top universities can’t produce graduates with the requisite AI skills quickly enough. Companies around the world compete fiercely for these individuals, without whom they can’t hope to remain competitive; LinkedIn is no exception.

The AI Academy

To address this problem, we’ve created the LinkedIn AI Academy. The goal of this program is to equip employees across the company—in areas like engineering, product management, etc.—with the knowledge they need to optimally deliver impactful AI experiences to our members.  

The AI Academy consists of different courses for different job roles and business needs. Engineers, for example, take a course called AI200, “Building an AI Product from End-to-End.” This consists of five one-day-per-week deep-dive classes, and a subsequent, one-month apprenticeship with the core AI team that takes participants from understanding how to incorporate and maintain an AI system to shipping one for their team. For product managers and company executives, there is a one-day, deep-dive course that focuses on the specific domain knowledge that they’ll need to manage AI products. After completing those courses, participants better understand one of the hardest problems in applied AI: knowing which problems are solvable via AI and which ones aren’t. Beyond these use cases, we envision extending the program to employees that work with populations who use AI, such as sales people who work with technology clients and recruiters who work with LinkedIn Engineering.

We’re currently in the midst of our first cohort of participants from LinkedIn Engineering. While we may adjust our programming as the AI Academy matures, our goal will always be to provide participants with the appropriate balance of classroom and hands-on instruction. Depending on the level of the Academy that participants are in, they may complete the program with an AI product that they’ve fully designed and that is ready to integrate into their team’s production workflow. All instruction is provided by AI experts from LinkedIn’s core AI team.

Class name Class outline
Level 1: Product/Engineering Manager Awareness Designing/Managing a Relevance Project
Level 2: Engineering Awareness Partnering with the Relevance Team
Level 3: Engineering Production Building AI Products In Your Team
Level 4: Becoming an AI Expert A Deeper Dive into the State of the Art

We aren’t trying to turn every engineer into an AI Ph.D with this program—that would be impossible. Instead, we are opening up the AI toolbox to all of our engineers, so that they can more easily incorporate AI into their day-to-day work. We have incredible AI specialists at LinkedIn who design thoughtful and impactful systems. If we can train other software engineers on how to use these, that’s a win for everyone.

AI is an incredibly powerful tool. It is of paramount importance to us that anyone trained to use AI should also be well-trained in how to wield it responsibly. To that end, ethical approaches to AI are an important part of our curriculum. We look at issues like choosing training data without bias and evaluating results against individual subgroups and not just the aggregate. Additionally, we’re taking great care to ensure that any AI work at LinkedIn is still overseen by our specialist teams. The Academy is training engineers on how to use AI tools, but the full understanding of how those tools are composed is beyond the scope of what we can cover. That’s why it’s important that we have the continual guidance of the specialists who built these tools to ensure they’re being used responsibly.

Future plans

For us, success with the AI Academy looks like every engineer, product manager, and other employee at LinkedIn knowing the optimal amount about AI for their job. Our goal is to scale AI across all of LinkedIn. This will give all of our our employees a language in common with our AI specialists, facilitating further collaboration, and will allow us to continue increasing the value we provide our members.

Today, there are also many other resources that our employees can leverage for more general AI education, such as LinkedIn Learning, Microsoft Professional Program for Artificial Intelligence, and Learn with Google AI. While these resources have a more general use case than the AI Academy, we may evaluate their future use for a self-service model for specific use cases and groups of stakeholders.

As we further develop the program, we are exploring making it part of our onboarding process for new employees. This would help us realize our vision of making sure every employee has the access and understanding they need to work with AI in their specific role.

We’re looking forward to continuing to share updates and lessons learned as more cohorts graduate from the AI Academy.

Acknowledgements

The LinkedIn AI Academy would not be possible without the great people behind it. I would like to take this opportunity to thank Badrul Sarwar and Erik Buchanan for stepping up and running the show. I am also grateful for our inaugural engineer-teachers: Apoorv Khandelwal, Jerry Lin, Skylar Payne, David DiCato, Sarah Xing, and Ben McCann. Lastly, I would like thank Deepak Agarwal, Igor Perisic, and Mohak Shroff for their helpful support, feedback and for pushing our team to be ambitious in its goals.