This has been an exciting few weeks for AI – Microsoft invested $1B in OpenAI and Elon Musk wants to connect your brain to a computer via Neuralink.
AI is everywhere. Well, not literally, but it sure seems like it: self-driving cars, AI systems that detect lung cancer, chef bots in restaurant kitchens, poker-playing (and winning) programs, and a recent story about a man who said “I Love You” to Alexa a hundred times a day. Hmmm.
Given the proliferation of Artificial Intelligence in our 21st Century lives, I would like to explore one of my favorite topics – Learning. After all, it’s in the name: Machine Learning. In a series of articles, I will attempt to explain how AI is revolutionizing “human learning” from the inside-out. In this first post, we will explore AI-powered content curation.
At a high level, learning is a choreographed ballet of three different interactions: curation, assessment, and coaching. It all starts with curation (aka, “content”). As a designer, these are the building blocks of the learning hierarchy. Content fleshes out course scaffolding in much the same way as muscles and organs fill out the human skeleton.
From a learner perspective, good relevant content engages, inspires, and skills us. As I’m sure you have noticed, we are all awash in content. Today, the complete collection of human knowledge doubles every 12 months. In the near future (if you believe IBM), the entire universe of content on Planet Earth will double every 12 hours. What!?!
Bottom Line: learning and curation is not a quantity problem, it’s a quality and mapping problem. And like most modern challenges, the answer boils down to two things: speed and scale. Quite simply, how do learning designers find the right content and deliver it to learners in a personalized and prioritized way at the moment of need?
Up until now, this has been primarily a manual process. With the assistance of emerging AI technology, the world of content curation has become a lot more manageable, scalable, and impactful.
Let’s take a closer look at the five steps of AI-powered content curation:
Step 1: Activate AI on Learning Paths
AI-powered content curation starts at the Learning Path. The first order of business is to turn it ON. Effectively, you need to tell the system that you want to bring content into the learner’s carousel in an automated way. And, from a collection of trusted sources. After all, you don’t want to expose learners – or AI – to the complete content tsunami of the Internet.
Why? Because machine learning isn’t good enough yet. We are just starting our ‘crawl-walk-run’ journey. In the crawl stage, artificial intelligence is not able to distinguish content quality as well as human intelligence (more on this later).
So, you need to limit the scope of the search. Effectively, you need to create a secure, trusted sandbox for the machine learning to work in. That’s Step 2.
Step 2: Scan a Library of Trusted Sources
The universe of known content is a great place to play, but only if you are smart enough to discern “knowable” content from “fake news”. In the beginning, less is more. Start with certified trusted sources, such as Skillsoft, Lynda, or Pluralsight. Then, expand to include content-relevant blogs, vlogs, or RSS feeds. Thought-leader Twitter accounts and LinkedIn articles are good. And, if you have them, add internal corporate IP repositories.
While building your library of trusted sources, remember this important rule: the strength of the entire system is only as good as the weakest piece of content. “AI is vital to curation on scale,” said my friend Stephen Walsh (CEO at Anders Pink). “Nobody has time to scan 50, let alone 5,000 content sources every few hours to see what’s new and relevant for them. That’s what machines are for.”
Step 3: Deliver Personalized (and Prioritized) Content to Learners
The real magic in this new system happens when it delivers personalized and prioritized content to each and every learner – one at a time. In Step 3, the system automatically curates specific resources for each learner according to the skills in their Learning Path. In real-time, curator bots personalize the resources using pre-assessment and profile data to find and deliver the videos, activities, articles, and blogs that fill each learner’s skills gaps. The days of re-hashing competencies you have already mastered are gone.
In addition to personalization, AI-powered curation systems can use machine learning to prioritize high-performing content above everything else. Many of today’s platforms present required resources in a carousel UI – similar to Netflix. In this model, only the highest quality content is promoted into the carousel – i.e. the cream rises to the top.
How do AI-powered curation systems determine the “best” content? As with most things, the answer is a four-letter word: Data.
Step 4: Collect and Analyze Crowd-Sourced Usage Data
In his course, “AI for Everyone”, Professor Andrew Ng made the point that the power of AI is relational to the size of the data – the bigger the dataset, the more powerful the AI.
Crowd-sourcing is the best way to fill your data lake quickly and efficiently. As learners engage with your universe of content, the system collects thousands and thousands of records about efficacy (how usage of specific content correlates with outcomes), learner ratings, raw usage, shares and likes of particular resources, and the learner’s level of mastery.
In fact, each learner is surrounded by a “Code Halo” which defines his or her compendium of experiences, choices, and preferences. This is an elegant term coined by Ben Pring at Cognizant’s Center for the Future of Work and a great description of what metadata can do for learners in an AI-powered system.
Step 5: Prioritize Content via Machine Learning
In Step 5, the crowd-sourced learning data, and Code Halo, is packaged up and fed into a series of machine learning algorithms to separate the wheat from the chaff – i.e. only the best resources are elevated into the carousel. And this becomes a continuous self-feeding loop as the prioritized resources are re-curated within the library of trusted sources … enter the learner’s carousel … generate more accurate learner halos … create more effective curations … and the whole systems continues to improve (i.e. “learn”).
Bottom Line: the more learners use the system, the better their curation gets. This is the beauty of machine learning and the power of positive feedback loops – two pillars of AI. So, where do we go from here?
In the ‘crawl’ stage of AI-powered content curation, less than 10% of the resources in a given learner’s carousel are AI curated – all the rest are human curated. Over time, that number will grow as our confidence in the quality of the machine’s curation improves. In the ‘walk’ stage, more than half of a carousel’s resources will be AI curated.
Ultimately, in the ‘run’ stage of AI-powered content curation, entire learning programs will be machine-created thus freeing up humans (instructional designers) to focus on the higher value tasks of designing, scaffolding, and polishing Learning Paths.
Machine learning has the potential to improve human learning, but only if we don’t lose sight of the humanity in the process. Stephen Walsh said it well, “AI alone is not the full answer for curation. Curation is, in my view, an innately human activity. Only humans can truly discern relevance and make sense of it for themselves and their networks. And relevance is subjective and very personal. If we think of curation as seek > sense > share, to use Harold Jarche’s framework: AI can seek content, organize and aggregate it far better than humans. It can be highly effective in learning from my preferences, and those of others like me, and serving up content with a high likelihood of relevance to me and doing it automatically. So, let AI power that aspect of curation.”
Imagine how much easier life will be for all of us when we sign up for the Neuralink operation and combine machine and human learning. Hello Matrix!