LG seeks patent for AI-Air Conditioner

  • Date

    January 30, 2023

  • Read time

    3 min


Soon your AC might be the latest thing to think, learn, and self-develop, since LG Electronics intends to use AI to keep you cool while saving energy.  LG is the owner of a recently-published patent application, teaching an AI-based AC capable of automatically identifying people and adjusting itself to meet people’s preferences.  

According to an article published by IEEE, energy used on AC has tripled since 1990 and will triple again by 2050.  In 2018 the International Energy Agency calculated that ACs and cooling cause 20% of the electricity used in buildings worldwide, with an increasing trend.  As more people buy ACs in a warming world, there will be an increasing demand on our electricity grids while exacerbating global warming. It’s a vicious cycle that technology can alleviate. It is axiomatic that technology improves in iterations, and while there are plenty of start-ups aimed at new AC technology as aptly reported on September 10, 2022, in the Washington Post, LG proposes one iteration that might pay in dividends in the near-term.  

To that end, the invention is aimed at addressing the increasing energy demand from AC usage while making things more comfortable for people automatically.  

The invention identifies a person through an “image acquisition device” such as the camera and, using this computer vision matches him/her with associated data, including preferred room temperatures and wind speed.   

LG lists the advantages as: 

  1. Reducing power consumption by operating per members’ preferences.   
  2. Improving comfort for the user, though LG appears to hedge on this, stating that it “may have the advantage of” improving the comfort for a particular individual,  
  3. Automatically updating its operation according to the user’s preference, relying on feedback from that user, including via natural language processing.  

The power saving could be particularly interesting. In September 2022, California was again on the brink of blackouts due to the unprecedented heat wave burning up its power grid.  Indeed, Governor Newsom has been beseeching Californians to turn up their thermostats to 78 degrees. Perhaps smart ACs are the first step toward centralized power management, taking into consideration the specific urgency as to the end user such as age or medical conditions, occupancy of the home, or outside temperature. 

What if there are multiple household members in the room?

LG thought of this, describing a father who wants it cooler than the mother, who is concerned about the kids getting cold. The AC can lower the temperature but point toward the father.  But what if there is no room for compromise? It is unclear how the AC will react with conflicting instructions and what other input it should rely on. Using natural language processing, will the AC learn a pecking order? Will it observe sweat or sweaters? Can it lower the temperature if the parents are arguing over who makes dinner?  

The Tech

One challenge is to identify a person from images taken from the side, as it may be difficult to place a camera suitably to acquire clear face photos optimal for ID recognition algorithms. Obscured faces may necessitate analysis of other features, such as gestures, posture, body shape, and gait. The AI algorithm must be general enough to tolerate various situations, such as light conditions, camera positions and shooting angle, while being sensitive to features and appearance of the person that allow for identification. It may happen that, depending of the conditions, distinct personal features may play the main role in member identification. Technically this can be accomplished by appropriate embeddings and clustering.  

What if there is more than one person in the room?  If multiple people can be visible at the same time, perhaps a “key member” needs to be identified. In this case the temporal component needs to be weighed, such as first in the room, less frequent visitors, those standing vs. sitting, etc. It may be beneficial to store settings per group of people observed in the room. It may happen that there is a compromise between co-habitants when they stay together, while their individual settings might be incompatible. To that end, we propose additions to Figure 9 (proposed) of the patent, in the interest of domestic tranquility. 


Want to predict and measure future behaviors with accuracy?

Let’s talk to undertand how you can benefit from our CERN experience with machine learning, AI, data science and algorithms.