Symbolic Artificial Intelligence
you:- So what is The difference between AI and ML
Internet:- ML is a subfield of AI
you:- okay then what comes in AI but not in ML
Internet:- Vision, NLP, Robotics ….etc
you:- But doesn’t that require Ml algorithms??
Internet:- Yes
you:- *Visible Confusion*……. 😶 😶
This is generally the scenario when you come across an article/blog/video on the Internet which not only confuses you more but makes you question the information you already possess about AI. It is hard to find such articles that fully answer the question and make your life less miserable. So here we’ll try to discuss it way simple and see it bit by bit but you have to bear some history (*don’t worry we’ll keep it short*). So what’s all this fuss about, well you see in the present time Machine Learning has advanced so much that when someone new enters in the field of AI all they hear is ML ML ML… so much that difference between them gets fuzzy, and while it is true that ML is awesome, there are other mind-blowing fields too. Have you ever heard of “SYMBOLIC AI” or “GOFAI” or “GOFR” (Let me tell you this technology is powerful enough to create bare bones of “CHITI THE ROBOT” or “JARVIS” which ever you prefer) .while ML is the tech that solves subproblems like NLP, vision, robotics, autonomous vehicles…and much more.
Let start with the basics (here comes the history):-
What is Artificial Intelligence?
Its a term coined in an article of the 1950s when a brilliant (& lazy I guess) mind thought of a technology that can make decisions on its own (just like a human)with the data provided to it and the very first program based on AI was written in 1951. (Machine Learning wasn’t even born yet) This technology progressed a lot after 1980 but soon this giant fell. This period wasn’t good for AI and many even called it as “AI winter”
What is Machine Learning?
It’s a technology that is used to search for a pattern in the given set of data and then make predictions for new but relevant data from the pattern it has found. (Yes you might be confused but bare with me it’ll make sense after some time).It was first coined in 1959 ( nearly a decade after AI) and gradually started its dominance in the field of data analytics. Also its most popular approach to AI.
So What’s the Difference?
The Main Goal of an AI model is to mimic human intelligence as closely as possible or beyond it while on the other hand The main Goal of an ML model is more towards recognizing a pattern in the data set. Yes an ML model that can not make decisions is still an ML model if it can recognize the pattern or structure of data.
But what’s the use if it can not make decisions right??
Let’s take an example
Suppose you are a foodie and today you want to enjoy your dinner at a nice restaurant which you haven’t visited yet so you ask your google assistant (or some similar bot) to show you top restaurant in the neighborhood. It will show you a list of restaurants which are famous around you. Did it decided for you?? , No cause final decision still depends on you. So can we say it is an ML model, Yes because it still made the list based on various parameters like how many people liked it, does it maintain cleanliness, is service and food good there, is it well rated, etc. Is it an AI, well answer it yourself does it mimic a human brain somehow no right. But now you may argue that a google assistant is an AI well yes because it does much more than selecting a restaurant for you which differentiates it from a normal ML model. (Note we aren’t referring google assistant whole as an ML model we are referring to that particular feature which shows the list of restaurants. Google assistant is, in fact, an AI which comes under the category of “Narrow Artificial Intelligence”)
So yes an ML model can exist without becoming an AI. But is it’s reverse possible, Can an AI exists without an ML model. Before continuing try to think about an example of it .
Well, I hope you made that effort and might have found it but still I’m gonna write about it anyway so hang in there. For those who weren’t able to figure it out the answer is
*drum rolls*
Affirmative (*Of course right😂 bummer*)
Well, Yes, it is possible to build an AI without ML but how right, The whole decision-making process needs to find a pattern in the data on which decisions can be made which is what ML does so how can we build an AI that can decide without it. Well answer to that question is “Symbolic Artificial Intelligence” (check here for more information) Hero Of this story, whose legends are spreading in the valley till this day (*Okay enough*).
Symbolic Artificial Intelligence
So what is it and how it is different than ML, well the whole concept of Symbolic Artificial Intelligence (also known as “Good Old Fashion Artificial Intelligence” or “GOFAI” *believe I’m no making stuff up this is what its called*) revolves around the Symbols and their relation. Well what are those?? A Symbol can be referred to as a real-world entity(constants) and a Relation(predicates) defines how these Symbols relate to each other.
Let’s clear it more with an example, suppose you defined a relation like between A, B, C entities
Father(A) →B //Father of A is B
Father(B) →C //Father of B is C
GrandFather(A) →C //GrandFather of A is C
Now you introduce a new entity D like
Sister(A) →D //Sister of A is D
Than by logic we can say that Father(D) →B and GrandFather(D) →C
That’s it .it’s that simple (well until more complicated relations are made and it’s very helpful if you are familiar with terms predicates and constants if not than focus on the above analogy). This is how the first AI model was invented. If you are interested watch a GOFAI playing GridWorld.
Advantages:-
-Unlike ML, here the whole prediction process is Interpretable.
-Decides with the concept level through reasoning.
-Well established algorithms developed back then.
If it’s this advantageous than why does this has to pass through such a rough phase like AI winter. It’s actually due to its one disadvantage i.e
-All knowledge and Base relations need to be handcrafted by the experts
Since its handcrafted with more data it gets hard to construct such models. This is the reason why ML took over this position since once ML models are constructed they can train automatically without any human dependencies and it is advantageous in modern days since there are lots and lots of data.
But still, after this rough phase, it’s not yet forgotten it is still used by some companies and also used in the research field of “General Artificial Intelligence” which aims to create a human-like Artificial Intelligence like Terminator (*just kidding… or am I*)
Summary:- saw some history, gave general definition to AI and ML, saw how they can exist independently and an alternative way to approach AI.
Conclusion:- Artificial Intelligence is a huge field and frankly we have just peeped through a hole. Machine learning is Cool (it is awesome actually) but it’s not the only field so don’t narrow down AI only to ML.
