Intelligence based learning - AI versus Humans

There was a era where computer programs would take hours to compile and run, slowly it changed into minutes, seconds and now milliseconds or sometime nanoseconds. Thanks to Artificial intelligence, technology is unfolding in a way that applications are delivered to market in rapid time. From writing code, deploying it and taking customer feedback to the current era of taking initial feedback, delivering fast and rapidly iterating, machine and deep learning have been creating waves.

In a nutshell machine learning, is learning from behaviors of a machine and predicting the future outcomes. We did a poll recently on what are the most commonly used applications developed using machine learning

Even though poll numbers are not critical, poll result does give a indication that :

a. Machine / Deep Learning helps get our work done fast - Chatbots e.g.

b. Tracks data in real time over a long period of time and gives meaningful analytics that we can interpret for the better - Wearables e.g.

c. Ensures that our purchasing patterns are also based on history and our categories of interest - Recommendation systems e.g.

d. Ability of institutions to grant loans e.g. is not just based on how much we earn but much we have been able to sustain over a long period of time - Credit Score e.g.

Does this mean human inputs are not needed?

Once we create a system that knows history, patterns will this run like a machine and give better outputs?

What are the adversities of such a system?

Lets take couple of examples :

Example 1 - We need a to aggregate ratings from a e-commerce website (for a chain of products) and provide the information to company and also give predictions of where this product would be w.r.t buying trends / customer satisfactions in the next few years.

We obviously can write a lot of code but such a code is hard to maintain. We need to build a machine learning model that can read every rating interpret details provided as reviews and then predict sales for this model in next year. Below is a analogy along with this example on how a machine learning model is built :

As you can see above there is a lot of effort involved to cook a great dish and yet there is a possibility your best dish may not be liked by all. Same goes with building a machine learning model, it needs a lot of training, feedback and even if it has great accuracy and reliability it still needs to be retuned as data keeps varying, who knows one customer gives one rating but has only positive things to say about the product.

For Tech Savvy - We call this predictive analytics and as we have labelled data, we can use supervised learning techniques to build the model

Example 2 - We need to classify images for a website that has fun online quizzes for kids. Assume there different breeds of cats and dogs and kids need to find whether its a cat or dog and machine lets them know if they classified image correctly.

Imagine there are 10000 images already present in system of cats and dogs, writing code to classify these images may not be fruitful. We can perhaps use a deep learning algorithm to build a model that would be trained, tested.

For Tech Savvy - If there are 10000 images we can use approximately 8000 images to train the model and 2000 to test it and use Convolutional Neural Network Algorithm for building the model

There is a lot of intelligence that can be tapped into from building these models. however there is a lot of dependency on data, correct data, amount of data etc. Note that anything and everything needs human intervention. A doctor cannot just use a algorithm to predict if the patient has a disease, he can however use the technology to make informed decisions but his experience has to come to the fore. Building brains on top of machine is certainly exciting and a lot more problems would be solved, but the seed of solving such problems lies with the humans