Artificial Intelligence that not only imitates human thought and action but surpasses them? The concept of “Superintelligence” inspires enthusiasm and stirs up fears at the same time. The field of Conversational AI, in particular, can make a contribution to the development of Superintelligence.
However, we are not there yet. We still have a long way to go before we reach the age of Superintelligence – and thus realise the complete automation of many operational processes. In this article, we examine the connections and interactions between Conversational AI and the age of Superintelligence.
To do this, we look at the following questions in more detail:
What will Conversational AI look like in the age of Superintelligence?
Why is the structuring of knowledge in the form of a Knowledge Graph the best prerequisite for achieving Superintelligence?
How does Onlim help to make companies ready for the age of Superintelligence today?
First and foremost, let’s have a look at what Superintelligence actually is.
What does “Superintelligence” actually mean?
The term Superintelligence is used in scientific contexts but also in colloquiallanguage to describe an Artificial Intelligence (whether in beings or machines) that is so sophisticated that it is already superior to human intelligence and capabilities. In this sense, it is “super”, i.e. “above” humans.
While we are talking about extremely powerful AI, don’t think of Superintelligence as an AI that develops a life of its own and opposes humans as its inventor. As you might have seen it in some horror scenarios described by some books and films.
At Onlim, we see Superintelligence as the ultimate goal to which all current efforts by companies, theoreticians and researchers in the development of Artificial Intelligence are dedicated. From a business perspective, the focus is on harnessing benefits such as automation, cost efficiency and productivity gains while maintaining or improving service quality and customer satisfaction.
Superintelligence in Conversational AI
Conversational AI is an important sub-area of Artificial Intelligence. Conversational AI makes it possible for consumers to request facts, receive information and initiate transactions such as purchases and orders in natural language-based dialogues with conversational assistants. This can happen through any output channel, such as chatbots, voice assistants or intelligent search. When we talk about Superintelligence regarding Conversational AI, the goal is to achieve a similar level of quality and information talking to a conversational assistant as in conversation with other people.
The basic prerequisite for any Artificial Intelligence is a sufficiently large stock of data and information that algorithms can evaluate and gain insights from by means of targeted analysis. The processing and structuring of data can take place in different forms, whereby the Knowledge Graph approach is the most superior method. You’ll find out why in this article.
Putting an end to dialogue trees – The enormous limitations of static dialogue management!
In the last five years, we have seen an enormous development in the field of Conversational AI. In the beginning, in 2016, there were mostly quite simple chatbots. They became especially popular on Facebook where people used them intensively. However, these chatbots could only answer very basic questions. They were only equipped with question/answer options in the form of dialogue trees, which restricted conversational abilities.
Basically, a dialogue tree works like this: You start with a question and draw all possible answers to this question in a path diagram. Then, for each possible answer, you define the necessary follow-up questions and add possible answers to them.
The problems of this approach are manifold. Most importantly, such a tedious dialogue tree, executed down to the last detail, is very rigid. It is exactly the opposite of Superintelligence, of which characteristics are its ability to respond dynamically and in a flexible way to the dialogue and upcoming questions of the user.
Conversations limited by dialogue trees are a huge frustration for customers because they do not receive personalised and relevant information or results. As a result, companies may lose orders, contracts and customers (and possibly damage their brand image).
Why does Super Intelligence require a Knowledge Graph-based NLU setup & training?
Modelling data on the basis of a Knowledge Graph is a technically superior approach. Management consultancy GARTNER agrees, seeing “graph technology” as a prerequisite for Super Intelligence.
A Knowledge Graph consists of entities that we link to each other by nodes and edges. This linkage makes it possible to map the context of the information in such a way that the conversational assistant can recognise and assign the intention of the user. Basically, the Knowledge Graph maps and connects data in a kind of superior dialogue tree. Thereby so-called Natural Language Generation (NLG) carries out the formulation of the data into natural language-based sentences.
In the context of Natural Language Understanding (NLU), a Conversational Virtual Assistant (CVA) can simply jump to the relevant node in the Knowledge Graph and retrieve the required information. This enables dynamic dialogues between the user and the conversational assistant, and relevant and meaningful answers are provided – without the need to create exhaustively formulated dialogue trees.
Scalability and language understanding as obstacles on the road to Superintelligence
On the road to Superintelligence, Conversational AI has to overcome two major challenges: a) scalability and b) Natural Language Understanding (NLU), especially semantics.
The scalability of Conversational AI is necessary so that a large number of requests (by a large number of users) can be answered simultaneously and the company’s resources are relieved. Scalability is only given if the course of the conversation can be dealt with flexibly and in a dynamic way. This is why the creation and use of a Knowledge Graph is an essential prerequisite.
Let’s illustrate this with the example of buying a bicycle in a sports shop or a specialised online shop.
Knowledge Graph example:
The customer goes to a bike shop and wants to buy a green bike. He asks:
“What price range is the bike in?” 🡪 Criterion of price range
“For which gender?” 🡪 Criterion of gender “E.g. unisex, women, men”
“In which colour?” 🡪 Criterion colour
This guidance behind a dialogue comes dynamically from the content mapped in the Knowledge Graph. This is only possible because we specified in the Knowledge Graph that a bicycle has a price, target group, colour, etc. as ”specifications”.
Based on this modelling, the user can have a dynamic dialogue. There will be no dead ends, just informative results, such as the suggestion of suitable products (corresponding to the desired criteria).
Language understanding is only sufficiently given if the context and meaning of what is said or written can be correctly understood through semantics. This ensures that the Conversational AI can ask the correct questions at the right time, or present the correct output, e.g. information or result. This results in what we call “Knowledge-Driven Dialogues“.
Knowledge-Driven Dialogues as an early stage of Superintelligence
Dialogue tree-based, static dialogues are the past. Knowledge-driven, i.e. knowledge-based dialogues are the future – and the basis for Superintelligence.
A dialogue is knowledge-based the chatbot is enabled to ask questions based on existing background knowledge of a specific topic and can therefore guide the flow of the conversation. It retrieves this background knowledge from a well-structured Knowledge Graph and therefore enables a dynamic, meaningful dialogue. The conversational assistant can create and conduct an unlimited number of knowledge-based dialogues.
Thus, we create a dialogue qualitatively close to a conversation between people. The assistant is able to provide extensive knowledge in the form of natural language-based answers to relevant questions and topics. Therefore, informative answers and the change of the topic of a conversation depending on the course of the interaction imitate an interpersonal conversation.
If users perceive a chatbot as more intelligent or smarter, this also improves brand perception, service experience and user engagement, which in turn encourages purchase transactions.
Example - Search for a specific type of bicycle in an online bicycle shop:
Static dialogue guidance in a chatbot for product search can lead to existing products not being displayed because they are excluded too early by the order of the information requested. You may lose a customer or purchase as a result. As a customer, the chatbot might then guide you through an entire questionnaire (chatbot collects all info before querying the product database), only so that it can inform you at the end: “I am sorry, but I have no matches for you.”
In a Knowledge-Driven Dialogue, the user can start a product search with one criterion and the chatbot can suggest all relevant or in-stock relevant results. It knows the availability and stock data, so the assistant can steer the dialogue in the direction that will benefit the customer. e.g. “I want a red bike.” → if there is no red bicycle in stock, the chatbot provides this information after the first request. The customer is informed and time is saved.
The next part is then to assemble these answers in such a way that they provide the best possible answers for every respective channel. Via the Onlim platform, the knowledge modelled in the Knowledge Graph can be used as a basis for providing it in natural language-based form via chat widgets, voice assistants or search windows. Companies are thus ready for the vision of the future: to make all knowledge available in natural language-based form – an end to classic keyword-based search queries.
Linking Data & Dialogues with Insights & Analytics
Finally, it’s necessary to link the existing data and dialogues with insights and analytics. With each dialogue – whether it was a request to a voice assistant, a search query or text-based question in a chatbot – you can data, which tells you more about customer needs and their search for information.
How Onlim chatbots learn
The Onlim platform enables its users to analyse and evaluate this data. This way companies can gain valuable insights to optimise their data quality and complement its scope. For example, if you recognise that certain topics are always queried in combination, you will connect the relevant data in the Knowledge Graph. The structure of dynamic intents comes primarily from these insights; accordingly, training data is expanded and intents are set and expanded in the Knowledge Graph in order to be able to conduct more dialogues, and in a more flexible way.
It’s no different than in a brick-and-mortar shop:If you as a salesperson know that you don’t have a red bicycle in your stock, you will have a different conversation than if you know that you have one available. Guiding users through the dialogue with the help of the Knowledge Graph can immediately let the user if a product is not available and end the dialogue if no other options suit the user.
Using a dialogue tree, however, the chatbot would go on and ask more questions, only to let the user know at the end that there is no suitable bicycle in stock.
Onlim provides detailed insights and analytics in the Conversational AI platform. This allows customers to analyse and optimise their data and conversational assistant. We support the customer during every step of the process, from onboarding, to setup and ongoing operations. The maintenance of the data or the Knowledge Graph can also be handed over to us, whereby further maintenance is important so that the customer can continuously optimise its data quality.
Outlook: Superintelligence market-ready in 5 years – will you be ready?
5 years ago we found ourselves at the pioneering stage of the dialogue tree-based chatbots popular on Facebook. And it will take another 5 years or so before we reach the stage of Superintelligence. In about 10 years from now, we will interact with Virtual Conversational Assistants (VCAs) in everyday life. It will be natural without a second thought, even in virtual spaces.
What does this mean for companies?
First and foremost, that time is of the essence. Organisations need to take the first steps today. That way they create the foundations to be among the leading and most innovative companies in 5 years. Those who will not have sophisticated conversational assistants in use and have missed the innovation leap along the way will disappear from the market in 10-15 years.
Looking at this short time frame, the question is: How can companies begin to create the prerequisite in their operations, IT and processes to get ready for the age of Superintelligence? The answer is – the sooner you start, the better. Start with small steps. Find a use case to familiarise yourself with the technology, its benefits as well as challenges.
Think about the daily challenges in your operation. What could be improved by a conversational assistant? It could be a simple chatbot to start with. But make sure to initiate the development of the Knowledge Graph required for that use case. You can expand it incrementally; how to do this is up to you.
Once a company gets a better feeling for what potential the technology offers, it will also identify which use cases could (or could not) add the most value. The generated data as well as the fact that the company has to deal with this innovation will support this process. Last but not least, you must think of Conversational AI as an integrated part of the company. It has to become a part of market or customer communication.