The PLATO Connection (Knowledge Representation in AI)

Ram approached the large stone building; the evening wind was chilly, and he only wore a light jacket. Ram hurried in and reached the main lobby of his university, as he walked through the corridor passing one big pillar at a time, he started to remember how he had enjoyed working hard to get into the university. The university was everything and more than he had hoped for, but things were not the same anymore he was struggling to complete his work and assignments, Ram thought to himself maybe the pressure of academics have got to me, I have lost my mojo along the way.

Ram was behind on his paper on Knowledge Representation in AI, his friend convinced him to meet Prof. Marko, to get some insights. Prof. was known for his warm personality and innovative approaches; Ram was not in a mood for a motivational talk, but he decided to meet Prof. anyway as he had a lesser interest in working on his paper.

Prof. Marko sat in a large empty classroom reading a book when Ram found him. Prof. noticed Ram and greed him warmly, good evening Ram, putting down the book in his hand, your friend told me that you were going to meet me, so how can I help you? Ram noticed the title on the book put down by the Prof. it read Philosophy, Ram felt more uncomfortable and spoke, frankly I don’t know how you could help me, I am right now behind on my paper on Knowledge Representation in AI. Prof. looked at him and spoke, so how far have you got across with that paper then? Ram said in frustration I cannot even get started. Prof. smiled and said maybe you are looking at the wrong place. Ram protested I do not think I Philosophy is what I need right now, pointing at the book.

Prof. whimsically, on the contrary, Ram Philosophy may exactly be what will help you make progress on the paper. Ram demand how so?  Prof. continued what do you know about the ideas presented by Plato? Ram almost in disagreement, Prof. I need to write a paper on Knowledge Representation in AI, not about ideas with no practical use. Prof. said undeterred, Ideas from Plato can be considered as a key to the concept of Knowledge Representation in AI. Ram with a renewed interest, I did not know that, please tell me more. Prof. said one of the key ideas he presented is as follows,

Ram in a poker face, Prof. I will not pretend that I understood that, nor do I see how this can be used in knowledge representation. 

Prof. moved briskly towards the blackboard and drew a representation. He spoke pointing to the diagram, let us understand about the World Model,

There is a real-world, which is rich very rich and complex, Imagine a sunrise or standing on a traffic signal, the world has such richness and complexity that it cannot be completely comprehended, but to function and navigate through this complex world as humans we use or sense organs and perception to build a world of ideas, and Intelligent system does something very similar, it uses various sensors to capture the information about the real world and creates a representation of the real world which we can call the World Model.  

Prof. moved to the other edge of the board and drew another diagram, and said now consider this more familiar block representation of the AI system,

The system gathers various information about the real world through the sensors at its disposal and processes this information to create a World Model, which is a representation of the real world in the digital domain, World Model is used understand the current situation and generate behaviors in order the meet the assigned goals, which are translated in the real world through actuators, robots, voice assistants, etc. Better the World Model can be represented better will be the process of decision making and action. This brings us to the importance and motivation of Knowledge Representation in AI systems.

Ram was now beginning to see how Prof. Marko was leading him in the right direction, Ram spoke with enthusiasm, yes, I see how one could start approaching Knowledge Representation in AI, but what is the Plato connection?

Prof. Smiled and continued, Plato was one of the earliest contributes to the field of Epistemology (Theory of Knowledge), this can be considered as the basis or our Knowledge Representation and Knowledge Management in AI today.

Among the top problems that an AI system must solve is that of classification, imagine we are designing an automated driving system, the system must now be able to make out the difference between the road and the curb for starters, it is not good enough if it can do this classification on paths it has already traveled but it must also be able to do this in a new path where it has never traveled before. 

Plato explores how humans are normally able to do this, say we teach a child to recognize a dog, once the child learns about the dog it can recognize other dogs with ease, even different breeds. Plato says this the humans can do because once they learn about a dog, they form a mental image or an Ideal form of the dog in the world of ideas. Every time a new dog comes by it only an imperfect copy of the Ideal form in the world of ideas and hence can be recognized (or classified) as a dog.

This gives us valuable insights into how we can design and build an AI system that learns from a data set and can perform similar classification on data sets that it has not trained before. Knowledge Representation is at the core of this function.

Ram now fully in sync, yes Prof. I see how this is all adding up, so theories of Epistemology were some of the basis of how we have modeled AI systems and how Knowledge Representation or Models is central to this kind of Intelligent systems. So where do we go from here?

Prof. taking over again, there are 2 additional important assertions that Plato makes in this regard. One, he says rather proves using a mathematical example that actionable knowledge is not attained through senses alone but by reasoning, therefore even when 2 men may witness the same event and learn or conclude different things. This means that we are nowhere near the pinnacle of Knowledge Representation or World Modelling, changing the reasoning will create different and better models that will generate better decisions so this field of AI is wide open for more exploration and development of better models.

Ram, wow this is exciting what is the 2nd assertion that Plato makes?

Prof. seizing the moment, The Second important assertation that he makes is “If particulars have to have meaning there must be universal”, let me break it up for you, If a dog has to be recognized as the dog then there must be an Ideal dog (or a universal dog) in the world of ideas, without this Ideal dog one may not be able to recognize or deal with the dog. Let us understand the profoundness of this argument and its impact on AI. It is one of the key survival instincts of humans to classify unknowns into a known template so that we could respond appropriately. Just consider how we can easily classify both and oncoming car or a charging bull at threats. We can do this by generalization that is being able to apply what I have learned in one experience to a new situation. This again is a topic of wide research in AI systems we have got to a point where we some fairly smart systems that can solve specific problems or narrow AI, there is still a long way to go before we can develop a truly Generalized AI, and guess what the road to Generalized AI passes through Knowledge Representation and Modelling.

Ram was raring to go now, Thank you so much Prof. I am so thankful that you helped me out here, I feel my energies flowing back, I am all set to work on my paper and make a difference.

Prof. Marko, picking up his book and preparing to leave, I am happy I could help you, Ram, I would be keen to look at your paper when you are done. Saying that he walked away into the large corridor with a smile.