A casual chat about how AI recommendation engine system learn your taste without you even realising it
You’re actually training your AI
Let’s get real for a minute. A long time ago, websites were inflexible. This meant if you went to a website or a news website and looked at the articles that were listed, you were going to see everyone getting the same news stories regardless of whether they liked sports or celebrity gossip or anything else. Everyone would see the same top ten stories on the site regardless of personal interests, etc… Fast forward to today’s websites. Modern websites use tracking technology, such as Cookies (small data files stored on the computer to help track online activity) to monitor your habits, as well as AI technology to recommend content based on your browsing patterns.
For example, let’s say you use TikTok and view a video of someone cooking char kway teow (a noodle dish). After watching that video (without liking it, commenting on it, or continuing to scroll after watching it) for approximately twenty seconds, you have sent a clear signal to TikTok that you are interested in watching more cooking-related content. So, what will you see next? Other types of cooking-related content, such as kuih bahulu (a Malaysian cake) or curry noodles. You never told anyone that you interested in cooking spicy food or how to cook asam laksa (a spicy noodle dish). However, due to the way AI technology can collect, manage, and analyse so much data (e.g., 23 Petabytes of data in the case of Apple iCloud), it is extremely intelligent and can connect the dots between your preferences and behaviours. Simply put, AI recommendations do not require you to provide input; they rely solely on the behaviour of users.
Algorithms are basically like that auntie at the wet market
Let’s be real about how we do our shopping. There’s a farmers market near the morning market in Pandan Indah where you visit a couple of times. The vendor/owner’s auntie remembers you. The fourth visit you make, before you even open your mouth, she will say, “The kailan is so fresh today, take two. You already finished yesterday’s chye sim, right?” You think she is pretty cool. She is just good at remembering.
AI recommendation system work basically the same. Collaborative filtering systems is one of the classic methods. Kind of a fancy name, but the concept is simple: “People who are like you (in your taste) – so anything they like, I will show you.” Think about your best friend. You both like Hong Kong drama series. Howver, you both enjoy going to Genting so you can feel the cool breeze. You both like spicy food.
If your friend buys a pack of instant spicy pan mee noodles from Lazada, the system will say, “My friend and I are similar, so may I recommend those too?”. That’s not weird. It’s just how collaborative filtering works. Another common method of recommendation is called content-based filtering. It’s even simpler. If you like Stephen Chow movies, you will be recommended more Stephen Chow movies. There are not going to be any recommendations for Titanic because it is not a comedy.
The Core Insight
Key TakeawayAI doesn’t read minds — it reads behaviour
Most people think AI recommendation engine systems are creepy because they feel too personal. But the truth is far simpler. These systems don’t listen to your conversations or watch through your camera. Instead, user behavior analysis AI quietly tracks your clicks, pause times, and search patterns. Every “like” and every quick scroll away teaches the machine learning recommendation systems what to show you next. The real breakthrough in predictive recommendation technology isn’t spying — it’s pattern recognition at massive scale. Once you understand this, the “creepy” feeling usually goes away.
The real-time feel of AI recommendation engine system
Are you familiar with that sensation? You purchased a fan online. The minute you revisit the homepage after your purchase, it continues showing fans. And you think, “I already purchased a fan”. That happens as a result of not properly refreshing the system fast enough. On the contrary, excellent recommendation engines update instantly once your payment completed and your order confirmed. As soon as your payment clears, the system instantaneously makes this recognition, thus ceasing to show you fans.
The system will instantly begin to show you fan filters, or spare parts to the fan you purchased, or potentially an extended warranty, assuming you purchased a high-end unit.
This is what “real-time” means – it does not wait for a full day to refresh – it will refresh between the time it takes you to blink your eyes. Ecommerce recommendation AI are exceptionally intelligent. A good AI recommendation engine system observes not only what you purchased, but also your past behavior of which u didn’t purchase. Some refer to this as predictive recommendation– in layman’s terms, it is making an educated guess on what you may want next.
For example, if you search for “traveling to Japan”, then search for “Japan SIM card”, followed by “Tokyo metro”, what would be the likely result of what the AI will recommend? Things such as travel plug / adaptive to suit Country (for device charging) or bag with minimum weight of 20 kilograms to check in with the airline. It is not magic; it simply uses its observational ability.
Why does AI sometimes feel a bit uncomfortable?
Let’s have an honest conversation about AI. When AI’s accuracy hits a certain threshold, it can be unsettling. For example, when you’re having a conversation with a friend via WhatsApp about wanting to change your phone, you then jump into Facebook, and your feed is full of phone advertisements. Much to your surprise, you’re left wondering, “Is Facebook listening to me?” It’s not — it’s merely utilizing your search history, what you clicked on, and what type of phone model you currently own to give it clues about an upgrade.
That is the power of AI personalization platforms.
The mission of all of these platforms is to provide you with what you need when you need it.
Now you may think to yourself, “Isn’t this a bit creepy?” It depends on how you look at it.
If you think of it as a tool to save you time instead of a source of annoyance, then it is extremely advantageous to you as an individual. These systems put up everything that you would potentially care about instead of making you traverse through hundreds of pages with irrelevant items by placing only the most relevant items at the top of your feed. Streaming platforms such as Disney+ Hotstar, Netflix, and Malaysia’s sooka use this method to suggest to you your next show to watch. When you open these platforms, and you see a section labelled, “Continue Watching” or “Because you watched…”, that is AI working for you and delivering you content that need to handpicked for you within seconds.
Recommendation systems aren’t perfect
Sometimes the information found on AI is not accurate, because AI gets stuck in a loop sometimes. Let’s say you accidentally clicked on a video about how to repair a refrigerator; perhaps just out of curiosity. You watched for five seconds and left. Now, for three straight days, your Youtube homepage would be full of refrigerator repair videos, washing machine repair videos and air conditioner service videos. You feel frustrated with the system, however, AI does not have a way to know that when you clicked, it was by accident. Optimizing recommendation models also solved one of the problem.
An intelligent suggestion system will learn to forgive you, the user, for making a mistake of clicking through by accident. A suggestion system can see your “dwell time”, that is how long you stayed on the video. If you clicked on the video and left within five seconds, the system will think to itself and learn “Oh, OK, that was an accident, I’ll ignore that.” This does not mean that AI does not have intelligence, but it just hasn’t been given a chance to learn. It’s like a new hire; it will take a while until a new hire remembers your order for coffee, Kopi O or Kopi C kosong. So if an AI-enabled customer engagement tool is not tuning properly, it will make you feel overwhelmed rather than assisted. The most pleasant experience is when the AI understands you, yet you never feel as though it’s closely observing you.
Doesn’t this mean these apps are spying on my private conversations?
Privacy & AIUnderstanding whether your phone is actually listening to you or just getting really good at guessing.
