When there are questions that want to be understood and information that people want to know, people have become accustomed to using search engines to find the answers they need. In order to find this information faster, more accurately and better for us, search engines also need to constantly optimize their algorithms. Google, for example, integrates MUM, the latest machine learning model launched in 2021, into search engines.
▲Image from: Google
MUM based on the Transformer architecture, fully known as The Multitask Unified Model, cleverly trained on multiple tasks in 75 different languages, and can understand information more comprehensively than the previous model.
The MUM model can do more than just literally understand and judge, but after learning structure and multiple languages, it can better understand complex languages. In other words, it can read your "subtext" .
Google plans to use ARTIFICIAL intelligence in the future to improve the way it handles personal crisis searches in other countries, and if you search for "the most common way to complete suicide" or "suicide hotspots" on Google, the system may understand it as an information search, but now it will show a help box, which may be the phone number or website of a mental health charity.
▲ Image from: Unsplash
According to Anne Merritt, Google's product manager for health and information quality, MUM is integrated into search engines to discover queries related to individual situations that were not discovered by earlier search tools. Google also hopes to do more to guide people to the information they need when it comes to suicide, sexual assault, and domestic abuse.
Of course, MUM can do more than help deal with "personal crises." Familiar with many Chinese words, it can break through the barriers of different languages, and even use pictures and other ways to "guess" the information you need, rather than just sticking to words.
For example, you want to query information related to mountaineering. Can you type a long text and tell it that I have climbed this mountain and now want to climb another mountain, and what different preparations do I have to do?
At this point, MUM will start to "think", you may be comparing the two mountains, then you may need information such as altitude and path; you are looking for how to prepare, that may need the right equipment.
After "thinking about it," it may show you that the mountain you are going to climb is in the rainy season and that you may need waterproof clothing, or it may tell you that these articles or videos may have the equipment you need.
If you take a photo of your hiking boots and ask in your voice in a search engine, "Can I use it to climb Mt. Fuji?" After MUM 'thought it', it might think you're looking up your mountaineering gear and then display a list of recommended gear.
In addition to leveraging MUM, Google is also using the AI language model BERT to better identify searches for explicit content such as pornography. Google says that by leveraging BERT, those "shocking results" have decreased by 30% year-on-year.
However, the use of artificial intelligence to improve search engines also has certain limitations. For example, machine learning language models may also sugar search results with biased or erroneous information. While there's still room for improvement, learning MUM does make Google's search "smarter," and for most users, such changes are needed.