Devdutt Shenoi 16 September 2017

Artificial Intelligence

This is a new section of the BEARly Speaking blog where I condense a topic, according my understanding of it. Be sure to read my other write-ups. You are free to comment your thoughts in the Disqus chat below or just email me devdutt@outlook.in
Artificial, as in Man made?

No, technically Artificial Intelligence is built on man-made machines, but these thought-like processes are infact the manifestations of a program which runs through a lot of data and finds similarity in patterns(infact, click this link to know more myths). Yes, this is a process very similar to the one we humans use to define stuff(e.g. kids find a pattern which they use to make demands), but this makes use of a lot of past knowledge and that is where one-shot learning comes into use, so a true AI is not completely human programmed and neither will it be completely human independent.

Define AI, DeepLearning and Machine Learning

Basically, they are all the same. AI is infact a power-set of all the jargon I threw up there. Deep Learning(DL) is a sub-set of AI which deals with the field of data science and finding patterns in large data, it also happens to be the sub-set of Machine Learning(ML) which deals with development of programs for computers to learn. All of this teaching and learning process is from patterns which are inherent in data. So, in an example:

a learning agent 'AlphaGo' => played the game Go, thought to it by a Machine Learning Algorithm, specifically a Neural Network(NN) which so happens to be a Deep Learning method.
This all comes within AI, don't worry, it gets easy.

Deep Mind by Google
Is Artificial General Intelligence, dangerous

This is a very dangerous topic, it is like saying that Trains are monsterous or the Titanic is unsinkable. We can only let time define the path we take. While Elon Musk and Stephen Hawking might be catious, people like Mark Zuckerberg and myself(wink-wink) woud love to program true AI.

Jarvis by Mark Zuckerberg
What will happen to the work economy

Yes, many people will lose their jobs, but is it not true that machines abolished manual-scavenging? Similarly, we will reach a pinnacle of humanity, if we let machines intelligently make decisions in their own narrow fields of work(see, I don't profess a world with AGI). We can let a thermostat define the right temperature according to the comfort of the human, but not let a computer define the right time for Nuclear Annhialation (I am singing "Give peace a chance" AFK).

So will computer programmers be affected

Okay, we are very self-less people, we coders, we made sure that if computers can learn, they can learn how to code by themselves. While it might not be able for machines to inherently understand complex topics, let alone the stuff of PhD Thesis(as in RNNs and CNNs), the jobs of testers and debuggers are already being handed down to machines, if you have not already noticed. So, if you are ready to innovate and show your creativity is worth your salary, you don't lose your job.

Where are we headed with Intelligent Agents?

Yeah, Siri and Google Now are here to stay, they are like the ancestors to a world of "HER" like PAs -who will remove the use of human connections and bonds forever- I'm joking alright, this is all hala-ballu. Humans have always been more or less of two type, the social and the anti-social, this is Psychology, and I didn't do Med-School, don't ask me.

What is it all about CNNs and RNNs in Machine Learning

These are the methods in which data is processed to make out patterns, these NNs are either Forward-Feed(FF) or Circular(Ci). Convolutional Neural Networks(CNNs) are straight FF-NNs which don't have processes passing through them in loops, whereas Recurrent Neural Networks(RNNs) are Ci-NNs which make use of loop structures to make out patterns. CNNs have been successfully used in image recognition, RNNs are more preferred for handwriting or speech analysis.

From my Reading list

These are some of the articles which I would like to point to, while I don't personally vouch for the content of these, they have certainly helped me a lot.

  1. How do I learn Machine Learning? (Quora)
  2. Machine Learning in a Year
  3. Learning how to code a Neural-Network
  4. Learning Machine Learning and NLP from 187 Quora Questions
  5. Parsey McParseface by Google
  6. Inside OpenAI
  7. Let your Code learn from Text
  8. Andy Rubin on AI
  9. DeepMind : Demis Hassabis
  10. GoodAI : Mareka Rosa