How Reinforcement Learning Led to Facebook's Slowdown



When you are on TikTok the machine learning algo matches you instantly with like minded profiles with similar characteristics as your own live. On Facebook you view pictures of profiles who in the past - people also viewed. The key difference is the ability for TikTok to learn about your profile and characters and quickly match you to profile's that you would immediately be captivated by and are similar to.


The ability to match this type of preference, to augment time spent on their platform is almost magical. Like a lure to a fish under the sea the videos that are shown are instantly addicting to watch. Facebook & Instagram show you pictures that based on your previous search history are likely that you would want to see again. In a world that is continuously advancing in machine learning and artificial intelligence best practices, Facebook and similar organizations are already falling behind.


To understand the reason's for Tik Tok's success it is necessary to have an understanding of the difference between supervised learning, unsupervised learning, and reinforcement learning.


Supervised machine learning algorithms can be used to predict either a binary outcome (for example, whether a picture shows a cat or a dog) or a numerical quantity (such as the sales forecast for a particular product).


Unsupervised learning algorithms aim to find “natural” groupings in the data, without labels, and uncover structures that may not be obvious to the observer. Unsupervised learning is useful for gaining insights from social media postings by, say, identifying customer groups and sentiment patterns that can be used to guide. A fashion retailer may use this approach to understand how to segment its customers based on the types of products purchased.


Reinforcement Learning Although they are still relatively underdeveloped, the potential applications of reinforcement learning may be even more impactful than those of supervised and unsupervised learning.


Reinforcement Learning - Software agent can interact with the environment and take actions within it to maximize a predefined reward. By giving the rules of the game or environment to the agent, the software system can quickly learn to maximize rewards and achieve superior performance.




A reinforcement learning software agent can start somewhere and probe the space, using as a guide whether we have improved or worsened our position.

Originally Best Practice was to input vast droves of historical data and build models around common patterns. Now best practice is to have AI learn these models on their own through quick implementation





The easiest way to experience the difference is to use TikTok. TikTok can cluster your preferences live and sync you with videos that are trending. Facebook however, mostly shows you pictures based on what similar past users, like yourself, viewed in the past, relative to TikTok the images and videos will be outdated.


Reinforcement Learning although still relatively underdeveloped, the potential applications of reinforcement learning may be even more impactful than those of supervised and unsupervised learning.


A current state of retailers, with a future state retailers after applying a reinforcement learning conversational ai chatbot on their ecommerce store.



Currently in-store sales and online sales with live person.


Apply conversational AI through an e-commerce platform as online conversational AI chatbot for a desired output – (Higher Sales as an Example) –


Apply continual relearning to uncover new segmentations, customer profiles, & new discounts to increase sales.


TikTok's dominance is a foreshadow to any software, social media, or other type of firm that utilizes the previous two machine learning methodologies. New organizations using reinforcement learning will be able to easily compete and replace them with higher performance, lower costs, and easier implementation efforts.


Sources: Competing in the Age of AI