What is Artificial Knowledge (AK) ?

The best way of describing AK is first to understand what knowledge is. We all know intuitively what we would classify as such, but experience problems whenever trying to define it. Historical battles or cultural movements would be considered knowledge, but what about news of Lady Gaga’s meat dress or Donald Trump’s tweets? Would classifying such information as knowledge only depend on if I am one of their fans or not?

To understand what is knowledge, we must resort to philosophy, and its knowledge related discipline called epistemology. Through debates that go way back to the Greeks Plato and Aristotle, classical epistemology teaches us that knowledge is a collection of information that passed the process of a justified true belief (JTB). Let me give an example. A person would need to believe that Trump’s tweets are correct, that they are in fact true and not fake news, and that he or she is justified in believing the tweets to be right. Since the 1963 essay of the American philosopher Edmond Gettier,¹ we know that JTB tests aren’t sufficient and must add other requirements to take into account biased and time-based beliefs. Therefore, it must additionally be justified by a community of agents agreeing that the right methods and conditions were applied, when proving that the belief was correct. It must also pass the test of time and be meaningful.

By applying these rules and as, forty years from now, nobody is likely to care if Lady Gaga wore a dress made out of flesh or if Trump twitted false affirmations, we cannot consider such news as knowledge.

Constructivism explains what the sources of knowledge are

So how do humans gather knowledge? Most of us heard during high school about Rationalism and Empiricism, two philosophical movements that are now outmoded because they failed to integrate new findings coming from neurosciences and cognitive theories. For instance, the role of experience in shaping our neural circuitries and through this process enabling intelligent thoughts has been clearly identified, contradicting Rationalism’s posit that knowledge derives from reason alone. Empiricism, which denies the notion of knowledge acquired at birth, also fails to integrate factors such as heredity, which neurosciences and IQ tests prove to have a significant impact on.

In his book “Critique of pure reason,” Immanuel Kant shows that understanding the environment requires both experience and a priori concepts (5 Great philosophical works you should read).¹ Our universally shared mind structure shapes our experience and the way we project ourselves in the world, according to notions such as time and space, as well as categories (e.g., cause and effect, substance, unity). Moreover, our mind orders its experience in its own way, according to specific innate patterns, which don’t reflect necessarily the reality. Thus, and though there cannot be any knowledge without sensations, these sensory inputs cannot provide it all by themselves either. Knowledge is possible because it isn’t so much how things really are but rather how they appear to us. To use an image, reason provides the structure of what we know, while the senses supply its content. As reality is incoherent and unverifiable because depending on our biased perception, there cannot be any claim to universalism or objective truth, but there can be a representation of reality based on a constructed model of this world. In other words, Kant’s view is compatible with Constructivism, a movement which pictures knowledge as contingent on convention, human perception, and social experiences.

Knowledge originates from propositions

How does this structure allow our human mind to abstract from previously acquired knowledge and why don’t we restart the knowledge acquisition process all over again, when confronted with a different situation? Well, we achieve this by classifying things of this world or events in categories, which can be seen as pointers in this knowledge process. Kant created this notion of Categories (the word is even his invention) to deal with the idea of Objects (e.g., cat, tree, or freedom) and how we understand what they are. By grouping Objects and Concepts by commonalities, this categorization process allows us to recognize new Objects as members of an existing category and to focus on relevant factors, within the observed environment.

With Kant, for the first time, someone reasoned that the issue with knowledge wasn’t so much to ask what a cat was, rather than asking if a cat is an animal that stands on its four legs and hunts mice. It changed the perspective of understanding knowledge by disserting about the ‘knowledge that’ rather than the ‘knowledge of,’ which is better associated with intelligence. In this context, to understand knowledge is thus to focus on propositions, which brings us to linguistics, rather than to the Object itself.²

 

Categorization creates general knowledge

It isn’t so much through the perceptual reasoning required to form these empirical concepts that we create categories but through the act of imaging an Object (like the picture of a cat) that we understand what this Object is and can recreate it. By focusing on the interpretation of a series of propositions, we acquire the knowledge base, which has been personally, socially, or scientifically structured. This categorization gives us the prism through which we filter and interpret the world, providing us, individually or collectively, with the connections between justification, beliefs, and truth. To make an analogy with the digital world, these Categories are not only the rules that we infer and apply to the links between different data or information, in a kind of relational database (i.e., Objects related to one another) but they also provide the connections inferred between different relational databases, in order to create a grand repository. In other words, Categories glue together data, information, and knowledge.

Understanding what a cat is

To understand what this glue is, I’ll now refer to this categorization process, by introducing the notion of Cognitive Type. Umberto Eco developed a semantic/semiotic model, which explains how information on an Object or Concept, already “imaged” through the cognitive process, can be organized around semiotic practices, culturally acquired. An individual defines the content that best describes what the Object is, according to these practices. They include, besides the obvious verbal language and visual imagery, all representations by our inner self to “talk about” this object. According to Eco, there are four different types of object information: iconic, propositional, narrative, and affective. Eco’s semiotic model describes the general way we categorize things.

Description of Artificial Knowledge

AK follows the same rules that we just presented, but from a computing perspective. It doesn’t include the machine learning process, associated with Artificial Intelligence, or all the other processes linked to deep learning and different algorithms that create content. In fact, it must be seen as the end result of these AI processes, together with its storing and retrieving. Therefore, AK involves search engines.

Nowadays, search engines such as Google, are trying to organize the world’s information in a meaningful way to provide the content that the readers are looking for. In the past, these search engines had to rely on the words within the pages, matching these words with the reader’s exact spelling, but this isn’t necessary anymore, thanks to semantic networks.

New technologies and web structures are enabling the match between the query intent and the suitability of the identified web pages or contents. In the past, HTML used a simplified list of semantic tags: <h1>;

 

As a result, content will more and more be categorized at a finer granularity level, enabling constantly more relevant queries and extracting knowledge more efficiently. In other words, these new web technologies are increasingly solving methodologically for machines, the issues related to the “Knowledge that.” They not only give them access to all of humanity’s knowledge base but also provide them with this richer contextualization, giving them the propositions, which enable them to understand what objects are.

It means that for sophisticated systems such as IBM’s Watson, machines will in the future, read and interpret two million pages in less time than the current three seconds. If we add to this incredible performance the capacity to generate structured information that systems based on the Internet of Things (IoT) with modern software architectures (e.g., SOA 2.0) have now, we could claim that we’ve already reached Strong AK (by analogy to strong and weak AI). However, this is another issue that you can discover in my book “What’s on their mind? Biological and Artificial Intelligence.”

 

 

¹ Edmond Gettier, “Is Justified True Belief Knowledge?” Vol. 23, No. 6, O. U. press, Ed., 1963, pp. 121-123.

² U. Eco, “Kant and the platypus.” 1997

³ Serge Van Themsche, “What’s on their mind? Biological and Artificial Intelligence.” April 25, 2018