Lets talk Artificial Intelligence…

Winter is coming and so is AI, In fact AI has been here for quite a well now.
The history of Artificial Intelligence (AI) began in antiquity, with myths, stories and rumors of artificial beings endowed with intelligence.
Although AI did play an crucial role in shaping the World War II but what truly sparked AI was the question proposed by Alan Turing in 1950, “Can a machine imitate human intelligence ?”. In his seminal paper, “Computer Machinery and Intelligence”, he described this with the help of a game named as “imitation game”.

AI has been changing the way we work, we live and is changing the society. There has been a lot of buzzword, lot of good hype but also sometimes unnecessary hype.
Lets try to demystify the AI with the help of following diagram,

• AI is actually two separate ideas. Almost all the progress we are seeing in the AI today is Artificial Narrow Intelligence. These are AIs that do one thing such as a smart speaker or a self-driving car or AI to do web search or AI applications in farming or in a factory. These types of AI are one trick ponies but when you find the appropriate trick, this can be incredibly valuable.
• AI also refers to a second concept of AGI or Artificial General Intelligence. That is the goal to build AI. They can do anything a human can do or may be a superintelligent system that can do even more things than any human can do.
• Lot of progress on ANI has happened but equivalent progress has not been made in the AGI as a result of which people are afraid that robots might replace the humans soon.
  If you are planning to build an AI system for your projects, make sure that thay are technically feasible as well as valuable to you or your business or other organization.

Machine Learning

Rise of AI has been largely driven by one tool in AI called Machine Learning. There are majorly two type of learning.
1. Supervised Learning
2. Unsupervised Learning

• Supervised Learning : This is the most commonly used type of Machine learning. This is learning that happens via Input to Output mapping. This Type of AI learns A to B or input to output mappings.
• e.g. in the input is an audio clip, and the AI’s job is to output the text transcript then this is speech recognition.
  - More the data we have, better results we can get while processing an input, with help of neural networks the level of performance increases.
• Data varies from business to business. More data is usually better than less data but no one can guarantee that gigabytes/terrabytes of data available will be valuable, therefore avoid rushing to your AI team with huge chunks of data. So before Overinvesting make sure that you have some discussions with your AI team because Data is messy. Infact Garbage data, incorrect labels, Missing Values, Multiple types of data etc add more trouble than value.
• Unsupervised Learning : Unsupervised learning is different than Supervised learning, In here we only have the input data and no corresponding output data/variables. The Goal of Unsupervised learning is to model the underlying structure or distribution in the data in order to learn more about the data.
  In Unsupervised learning there is no correct answers instead the algorithms are left to their own devises to discover and present the interesting structure in the data. UnSupervised learning can further be grouped into Clustering and Association problems.

Lets talk about various commonly used Terminologies in Artificial Intelligence :

1. Machine Learning : A system or a piece of software that any time of day, anytime in night you can automatically input A and it results in Output B. Machine Learning often results in AI system. An AI system running, serving dozens/hundred/millions of users is usually a machine learning system.

2. Data Science : A team that analyze your dataset in order to gain insights. So if a team with input A comes with conclusion saying that if do things in an XYZ fashion then it might help us in getting a better deal. Output of a data science project is a set of insights that can help you make business decisions.

Comparing Machine Learning and Data Science : As stated by Arthur Samuel, “Machine Learning is the field of study that gives computers the ability to learn without being explicitly programmed.”
A machine learning project will often results in a piece of software that runs, that outputs B given A. In contrast, data science is the size of extracting knowledge and insights from data. So, the output of a data science project is often a slide deck, the PowerPoint presentation that summarizes conclusions for executives to take business actions or that summarizes conclusions for a product team to decide how to improve a website

3. Deep Learning : lets say you want to predict the housing prices for an area. You have input like size of house and number of rooms. One way to do this is to feed this input to a magical BOX and have the price as output. The big magical thing in middle is call neural network(Artificial neural network).
Neural network takes the input(size and number of bedrooms) and then gives an output B which is the estimated price of the house. This is one of the interesting topic that is worth covering in an another article.

AI has few other useful terminology like neural networks etc. There are lot of other buzzwords like Unsupervised Learning, reinforcement learning, graphical models, planning, knowledge graph etc. We will try to cover them in other articles.

What AI can and cannot do ?

This required to understand that what AI can do and what it cannot. People some times infact most of the time has inflated expectations from the AI.

We should keep in mind that people, media etc mostly reports the success stories. They only broadcast stories of the projects that were able to implement the AI in there projects but they often miss to share information on the failures. Failure are the important part of any project as it helps in moving towards the right directions and taking calculative risks.

• for e.g. if a customer in e-commerce website raises a request of refund then
  - What AI can do is take refund request as input and forward this refund request to refund/shipping order department so that they take care of this.
  - What AI cannot do (atleast today is) to generate a dynamic response, empathize with the customer. Its very difficult to generate a dynamic output for every single possible type of email/message that you might receive.
  - Possible Cons of the above are :
  1. You may end just generating the same very simple response like “Thank you for your mail”, no matter what the customer is sending you.
  2. AI generating gibberish. This is a hard enough problem which may not be resolved even with 10,000 and 100,000 email examples.

Conclusion

There are no hard and fast rules for what AI can and cannot do, But it takes engineering teams to spend few weeks doing some work on determining if the project is feasible or not. AI is the new technology in the industry but it is not magic and it can’t do everything under the sun.