During the latest annual developer event (Google I/O 2021), Google announced the launch of MUM (acronym Multitask Unified Model), the AI-based algorithm capable of understanding and processing information in different formats (text, image or video) and in 75 different languages to provide reliable and detailed answers to complex queries.
Pandu Nayak, head of Search at Mountain View called his creature “A new milestone in artificial intelligence for understanding information.”
In this article, we not only want to explain what Google MUM is and how it works, but also try to understand what the opportunities and, conversely, the risks its introduction may pose to websites and online businesses.
Of course, we also wondered whether, in this new scenario, there will still be room for SEO and how search marketing will evolve.
How does the Multitask Unified Model work?
The principle that makes this technology possible is radically different from the current operating mechanism of search engines.
Indeed, search engines, including Google, deliver results to the user’s query thanks to a three-step paradigm that has remained unchanged until now:
1.Index: this is the preliminary requirement of mapping all web results scanned by the search engine and placing them within the databases.
2.Response: the phase called response is the moment immediately following when the user, after typing his query, clicks on the “search” button: this is the signal by which Google retrieves the index containing all the results, selecting those that according to the algorithm most completely answer the search.
3.Ranking: the positioning (also called ranking) of the results is, finally, the order in which the results are shown on the search engine page, or SERP (from the English Search Engine Results Page).
(If you want to learn more about how indexing and search engine ranking work, check out our seo guide)
With the introduction of the Multitask Unified Model, the current sequential model based on the index-response-positioning paradigm will instead be replaced with a single model:
Google’s current algorithms are also based on Recurrent Neural Network (RNN) models that understand the meaning of each word sequentially. What does it mean?
Take, for example, these two sentences:
- “fertile soil was found for plant growth.”
- “fertile ground has been found for the growth of the economy.”
In both cases, the Recurrent Neural Networks will have to reach down to the last word to understand the actual meaning of “ground.”
The Multitask Unified Model, on the other hand, is based on the Transformer model, which, thanks to the system called “self-attention,” understands the meaning of each term by relating it directly to all other words in the text regardless of position. Therefore, the Transformer model, by directly associating “soil” with “plants” or “economy,” is able to more immediately identify the meaning of the word.
In addition, format multimodality, i.e., the system’s ability to relate images, text and audio, has been introduced with MUM. Indeed, MUM can access information in documents (HTML, PDF, TXT, …) as well as image, video, and audio content, contextualizing and linking content in a way that goes beyond the “artificial” level of language.
The search engine’s priority will then be to provide the information that best fits the search query, rather than simply scanning documents on the topic. In this regard, Google’s search team is also laying the groundwork for breaking down language barriers that hinder the transmission of information. In fact, the algorithm has been tested on 75 languages to develop the ability to learn content from a foreign language and be able to automatically make it accessible even in the language in which the user made the search.
For this reason, Pandu Nayak said that MUM is capable of generating language and not just understanding it.
A striking example was the application of MUM to improve the quality of search results on the topic of vaccines for COVID-19. Artificial intelligence made it possible to identify more than 800 variants of vaccine names in more than 50 languages within seconds. Once validated, the processing results were applied to Google Search so that people could find timely and high-quality information about COVID-19 vaccines around the world.
Wanting to sum it all up in a nutshell, we could say that the end result Mountain View aspires to is to understand even the most complex queries and generate an answer comparable to what could be provided by an expert in the field. All this by gathering information from multiple sources, in different languages and formats, and synthesizing it into a single comprehensive answer.
An intent represented metaphorically with the now famous example of the hiker who wants to climb Mount Fuji, reported by the Google blog.
How will search (and the work of SEOs) change with Google MUM?
It is not yet known to us in what proportion this new technology will change search results. Even less predictable is the impact it will have, in the long term, on the way users themselves will approach searching for information online.
But some reasoning can already be done.
Given the incredible capabilities of MUM (which, according to its creators, is 1,000 times more powerful than the BERT algorithm), the first thing we insiders have asked ourselves is: what will happen to SEO as we know it today?
From the premises made, it indeed seems that the cardinal principle on which search engine optimization (but not only that, Search Ads as well) is based is disappearing: if the user’s language is no longer an issue, if text is no longer the only source that crawlers can draw on, what about keywords?
Actually, if we think about it, the concept of keywords in the narrow sense has long since become obsolete, satisfactorily replaced by the more open “search intent” paradigm.
The advantage, indeed, is that Google will be able to focus more on certain media that were less considered in the past: videos, images or podcasts. It is also likely that some “cornerstone principles” of SEO will even be strengthened: among them, the authority of backlinks. We can also expect an increase in the relevance of structured data as a ranking factor.
For webmasters, SEO consultants, and all professionals working to provide high-quality content and an optimal user experience, MUM will not bring disruptive changes for the time being.
Will online businesses be affected by this update?
A better understanding of search results across different media and languages will undoubtedly have a twofold implication: it will increase the potential target audience that a site can reach, and with that, competition will also increase.
In the future we will see our content compared with much more multimedia information in different languages, which means there will be much more competition than before.
Those who might be most affected are publishing projects: in addition to the increase in the number of sites with quality, perhaps local content (I’m thinking of special recipe blogs, travel blogs, comparators of products and services), the main competitor of these sites might become Google itself. Indeed, MUM’s artificial intelligence will give a strong boost to the process of transforming Google from a search engine to an answer engine, and this will potentially lead to further erosion of traffic.
For many ecommerce businesses, however, often the most overwhelming barrier to entry into other markets is not so much language or modal, but the limitations imposed by logistics. Despite this, some niche but economically strong businesses, if they can take advantage of the opportunity to present above-average quality content, may be able to emerge even without a localized site.
Why Google MUM is not only interesting to insiders
Talking between coffees about this news, more than one team member remarked how this algorithm somehow represented the emblem of many of Google’s famous (in a manner of speaking) “ten commandments.” Just to name a few:
- User first: the rest comes later.
From the beginning, Google’s professed priority has been to provide the best possible user experience, in this case, the most relevant and complete answer.
Putting the user at the center must therefore be a priority, not just to position oneself online.
- The need for information crosses all boundaries.
The Multitask Unified Model is, to date, the most significant expression of Mountain View’s stated mission: to facilitate access to information for people around the world and of all languages.
Facilitating this understanding process, with high-quality content and the use of appropriate html tags, will help your business get to the right people and in the best way possible.
- Excelling is not enough.
As Google says, we must remember that being the best is not the end goal, but a starting point. Setting ambitious goals and still expending all the energy necessary to try to achieve them leads you to go farther than expected, to try to introduce substantial changes, intercept latent needs, and always seek new areas in which to make a difference. - It is better to do one thing, but do it flawlessly.
A digital consulting agency is the right ally with which to prepare to best cope with change and optimize your online presence. Indeed, our goal is to be highly competent in all areas of digital marketing and to stay ahead of the curve by constantly improving.
