Blake Lemoine, a software engineer at Google, recently made a remarkable statement. He claimed that the corporation’s conversational AI bot LaMDA (Language Model for Dialogue Applications) had obtained a consciousness. Lemoine noted that the chatbot speaks its rights and perceives itself as a person. In response, company management suspended him from work - reported by The Washington Post.
“If I didn’t know exactly what it was, which is this computer program we built recently, I’d think it was a 7-year-old, 8-year-old kid that happens to know physics,” said Lemoine.
In fact, the capability of machines is becoming more advanced year after year, and Natural Language Processing (NLP) demonstrates prominent results. However, ones can protest in response: Then why do software like Siri or Alexa still barely cope with the most primitive tasks in languages?
For us, people, the language seems to be an ordinary and natural process. However, it doesn’t change the fact that natural languages are a very complicated system. Previously, there was a dominant idea that only humans could use logic, reasoning, and intuition to understand, use, and translate languages. However, human logic and intuition can be modeled mathematically and programmed. When computers became more advanced, people were trying to make them understand human speech. That is how the history of NLP went a long way from the virtual psychiatrist ELIZA in 1964 to the first automatically deciphers ancient language machine in 2010 and chatbots on almost every website.
Nevertheless, despite the long history of research, machines still have a range of serious limitations and barriers in terms of NLP. Machines can hear and read what we write, but they still do not completely understand what we mean since they don’t know the whole picture of the world. This is one of the problems of the century in artificial intelligence technology.
In the 80s-90s, at the beginning of the mass scale of the Internet, Web 1.0 users could only read the content. With the appearance of Web 2.0, there was the possibility to interact with text (Read-Write). That is why NLP technologies became especially useful and widespread in the second generation of the Internet. Thank to NLP, certain processes were facilitated, such as spam detection, chatbots, virtual assistance, and many others. Although the ability of machines to communicate on the human level remains relatively low, there are some quite interesting achievements:
Despite the fact that voice assistant was not developed to the level of the sophisticated interlocutor, they excellently perform their functions by assisting the user in different tasks.
Every day Google processes more than 3.5 billion searches. There are several advanced NLP models that are used to process the queries, including the most famous one, BERT (Bidirectional Encoder Representations from Transformers).
An example of how BERT improves the query’s understanding: When the user was quire “2019 brazil traveler to the USA needs a visa”, it was not clear to the computer. It could be a Brazilian citizen who is trying to get a visa to the US or an American to Brazil. Previously computers were giving results according to the keywords. In opposite, BERT takes into account every word in the sentence. In this exact context, “to” means a destination.
As was mentioned, Web 2.0 allowed users to create text in the Web space. Hence, there was a high demand for text correction. Previously, users were relying on the built-in corrector in the Office 360 software that was pointing out mistakes. However, the mistake-detection technology in that software was quite vulnerable.
The new generation software that is AI-driven, such as Grammarly, uses NLP to help correct errors and make suggestions for simplifying complex writing or completing and clarifying sentences. They are more advanced and precise than the error-detection technologies of the previous generation.
Today there are a lot of websites that offer on their web pages chatbot assistant. Same as with voice-controlled assistants, they are not fully advanced and, in many cases, use simple keyword search and decision tree logic. However, it helps the users at a certain level and facilitates the work of the support team.
It may seem that NLP development doesn’t achieve the expectational level as it could be and has a range of serious limitations. Hence, in Web 3.0, we must deal with all those flaws of NLP we have today.
There is no doubt that NLP will be a critical and foundational component of Web 3.0. AI-driven speech recognition and analytics technology can be used in many fields, such as enabling voice-based user interactions, command-driven hands-free navigation of virtual worlds, connection and interaction with virtual AI entities, and many others. Those are the examples of how NLP technologies will be used in web 3.0 and what are the objectives we need to achieve in this new internet era.
When we are talking about navigating in the metaverse, we might think about handheld controllers, gestures, eye-tracking, or voice control. In fact, voice operation would bring the user experience to a new level. Meanwhile, NLP technologies would help generate audio responses with linguistic nuances and voice modulation.
Speaking about voice operation in metaverse, it is important to mention development of the voice operation in games. The representatives of such in-game technology were games Seaman, Mass Effect 3, Bot Colony, Nevermind and others. Their NLP technologies were not that precisive. However, they can be considerate as the inspiration of the voice control in metaverse.
Voice operation is already in the development process by Meta. The company launched a Voice SDK that allows VR developers to create virtual environments using voice commands.
In Web 3.0, the users would expect assistance in more daily tasks than in Web 2.0. Hence the virtual assistance would need to be upgraded to a higher level.
For the development of such a voice assistant, Meta took responsibility. Meta claims that their voice assistant will take presence in the metaverse, so it will probably have a virtual form of an avatar. Moreover, Meta wants its voice assistant to be more sophisticated, unlike the existing voice assistant in terms of understanding the context of the request.
“To support true world creation and exploration, we need to advance beyond the current state of the art for smart assistants,” said Zuckerberg.
The difference between the Virtual assistant and AI companion is that the first one is designed to serve the user, while the AI companion is designed to be a virtual friend who knows your interest, can maintain a dialog, and give advice.
Web 2.0 offered us different AI chatbots. New generations of AI chatbots go further. One of the most prominent examples is the app Replika. After creating an avatar of its virtual companion that can also be visible via AR, users can start a conversation. AI companion, unlike many other AI chatbots, is capable of memorizing information about users, such as name, date of birth, interest, hobbies, memories, opinions, and others. Also, for the paid subscription, the AI companion can also be a romantic partner.
Developing technologies of virtual assistants and AI companions is crucial for the Web 3.0:
If we manage to master those, it will pave the path to mastering other AI bots that will be responsible for the various jobs, such as psychologists or consultants.
Trade is one of the areas where AI-driven chatbots that will have the role of seller/consultant will be required. Trade, in general, has a great potential in the metaverse. That is how AI chatbots in this field will upgrade the purchasing process and improve the buyer experience.
In the 2000s, the developers were trying to bring to the life idea of social media with a built-in translator. Users could talk to different people around the world in their own languages, and everyone will understand each other thanks to the built-in translator. For some reason, those social networks didn’t gain popularity. Moreover, translation technology at that time was not that well developed and often provided awkward results.
Metaverse will be more than just social media, rather also a place for business, networking, education, shopping, sports, and so on. In that case, the automatic translation would be very useful. However, to achieve this, we need to make sure that AI-powered translators cope with their job precisely, fast, and flexibly.
Meta is already working on such a translator. This year the company announced they are developing an AI-powered c for languages in their metaverse.
Another prominent company that reached certain achievements in the real-time AI-powered translations is Unbabel. The software combines the speed and efficiency of machine translation with the accuracy and empathy of a global community of native-speaking translators. The merits of the company was noted by Windows, they said:
“We’ve seen CSAT scores jump as much as 10 points, and in one instance, we increased issue resolution by 20 percent.”
Those are just a few examples of how NLP technologies can be applied in Web 3.0, which will be used provided that the NLP technology is advancing fast enough. In fact, the development of NLP in the context of metaverse remains one of the underrated topics. Nevertheless, it doesn’t change the fact that the ability of machines to perceive text remains one of the prior tasks for AI today.