Artificial Intelligence & Its Allies
Views expressed by the author are his own and done in his personal capacity. These are no way representative and/or reflective of the organization he is associated with.
Background
Recently I was watching one of the interviews of Humanoid Robot “Sophia” and got impressed with the way she was conversing, taking pauses at appropriate places, and providing logical responses to the questions. We can find more about her on the internet, but it was really amazing to see her body language and spontaneity in picking the questions and coming out with the responses in such a natural way, and we can never feel it as artificial. So how is it all possible? What are the tools and technologies involved at the back which are helping in achieving this feat is something which intrigued me to probe further? So overall we are trying to get little more into the details of Artificial Intelligence (AI), various terminologies associated and the way they are all connected to get into a holistic perspective. Al has undergone a major transformation in the recent past and we often see some of these terms like Artificial Intelligence, Machine Learning, Neural Networks, Deep Learning, etc. getting used interchangeably during conversations. The effort here is to segregate these topics individually highlighting their inherent capabilities.
I know this is an audacious attempt as most of these topics are widely in discussion and there might have been many articles already available, so let me see how much I will succeed in bringing my perception to the floor.
Let me start with the question of whether these topics relate to Science or Technology or both. Inherently science is something which is very fundamental in nature and has to be atomic through which other layers of technology can be built. The technologies thus built can help in arriving at the final solutions which should make sense from a business standpoint, or from a human consumption standpoint. Taking the same analogy one step further, we can map Artificial Intelligence as the technology built on top of computer science to mimic the human behavior from a cognitive standpoint, and this can have a huge impact both at the individual and the organizational level, and we are already seeing the results of this through the way the business got transformed in the last 5 years comparatively. The intent of this conversation is to provide a quick snapshot of the tenacity of this journey along with all the variants possible.
It is a no-brainer to understand the fact that all these revolutions (s) are becoming a reality only due to the economic ease we are seeing with respect to the underlying operational parameters like processing power, storage, and network bandwidth. Some of you may argue that Artificial intelligence is not new and was existing earlier too, and I agree with you all with a caveat that it was not effective due to the technical limitations with respect to these parameters mentioned (processing power, storage, and networking). However, with all the developments seen in the recent past, it is imperative now that sky is the limit for its accomplishments and “Sophia” just stands as one of the living testimonial supporting this.
Overview
Traditionally AI started as a rules-based concept which can be driven by complex decision-based trees, etc. However as the complexity got increased, the technical limitations with respect to processing and other parameters made it little unviable for larger complex problems. Now with all the recent advances in technology, we started seeing more and more growth in this area for the past few years. Engaging Robots in industries is pretty common these days and we are also talking about the lighter version of robots (called as Cobots) collaborating with human beings picking some of these helper functions. Unlike Robots, Cobots are easy to program and can be deployed quickly, and they can be a good bet when safety is a concern in some of these hazardous/lethal tasks. The developments we see with respect to Humanoid Robots, Driverless cars, and many other use cases are just a few of the examples and there are many more coming soon.
Machine Learning: With the advent of social and other data-centric initiatives, the volume of data has grown exponentially in the past few years and this has given some more opportunities and room for the AI to evolve. Machine Learning is one such opportunity which works more on the data and can help in predicting things in advance for better planning, preparation, and recommendations to the customers. There are many examples and use cases, and few of them listed below:
- predicting the time it takes to reach our destination when on a drive,
- predicting the right inventory required for a grocery store during peak times,
- predictive scaling of capacity in cloud subscriptions,
- predicting the weather forecast,
- predicting and recommending complementary items to the consumer
and the list goes on.
The data is playing a key role here, and this is more of an experience or knowledge-driven approach where we are leveraging the huge data available to come out with the possible predictions and recommendations. This may not be 100% accurate but we eventually get to that state with more data getting included. Similar to human beings, machines here are trying to learn based on the experience and making themselves more matured (learning from their mistakes) as they get into the future.
Neural Networks — Though it is listed as a separate item, it is a subset of Machine learning and the reason for its special recognition is due to its nature of handling with complex use cases like image recognition, voice recognition which cannot be possible with the regular Machine learning algorithms due to more no. of features and being computationally intensive. The option here is to take the analogy from the typical brain functioning and work on breaking the complexity into three layers (input layer, hidden/processing layer and output layer) for arriving at a final prediction.
For example, if we want to recognize an image (say a truck), there are multiple facets to it starting from its front, rear, doors, height, length, wheel size, mirror size, etc. and all these inputs have to be processed in the processing (hidden) layer to come to a conclusion on labeling it as a truck. Though it sounds simple and implicit for a human being, there is big math involved here as each component involved will have a lot of pixels associated, and all these have to be consolidated, processed to conclude it as a truck which is a mammoth exercise in itself. Here we are trying to take the Artificial intelligence one step closer to the way the real intelligence (our brain) works.
Also, with the regular machine learning algorithms, we can come to saturation on the performance of the algorithm, and there cannot be any improvement in spite of adding more data points. However, with Neural Networks, more data can always help in coming out with a better prediction.
Deep Learning — After seeing Neural Networks, we may be wondering about what else is required as we tried to mimic the human brain to predict some of the complex things. However we just touched a small portion, and there are many more exciting things possible and the answer for all that is Deep Learning. Let’s take an example of driving a car, here we are engaged in multi-tasking and will be attentive and watchful with respect to the on-going vehicles, pedestrians, traffic signals, and many more. It is a simple demonstration of our physical and cognitive abilities blending together in performing that task. This cannot be done with one hidden/processing layer but there need to be many parallel hidden layers working in tandem to achieve the desired outcome. Thus Deep Learning is an extension to the Neural Networks dealing with multiple interim layers handling multiple aspects in parallel. Also, the beauty of this model is that it is self-sufficient and works on improvements without waiting for human intervention to correct the model.
The best example to consider here is the “Driverless car” which has to process many things in parallel to get the car going without compromising on the safety of the passenger. The other common example we see on Deep Learning is about coloring the black and white images, there is a need for multiple interim layers as there may be many different objects, formatting, coloring involved within the same image.
Thus, Machine learning can become aggressive as we move towards Neural Networks and Deep Learning targeting more and more complex problems.
Final Comments
Overall AI has many flavors including physical, cognitive and it is working with many other technologies like Big Data, and the Internet of Things to extract the true potential. Baring some of the exclusive real-world situations like dealing with Natural Disasters, etc. AI can go hand in hand with human intellect in creating a better tomorrow. Any doubts?
