AI denotes a broad umbrella of hardware and algorithms encompassing machine learning as its subset, which, in turn, has deep learning as its subset.
In 2016, Sundar Pichai declared that the next 10 years would see a world ‘that is AI first.’
In simple words, AI is the simulation of human Intelligence processes demonstrated by machines in contrast to the natural intelligence displayed by humans. These processes include learning (acquiring data), reasoning (processing data to reach conclusions) and self-correction.
FROM ROBOTS TO MACHINE LEARNING…
For long, machines have been built which can handle a fixed sequence of operations with known inputs. Designated as robots, everything is standardised. Consider a pick-and-place robot. It is designed to move a known component from point A to point B. The same part comes to the same ‘A’ location, and the robot pushes it to the same position ‘B.’ Nothing much ever changes.
Not so for a machine with AI capability. Take a machine, which makes a medical diagnosis. Inputs will be from different people, the process of analysing the information may follow a standard sequence but could vary depending on changes in data. So, a machine with AI can take in data, analyse through algorithms, (a set of instructions) and produce a result. In the above case, the result would be a ‘prediction’ of a diagnosis. It would next need to be improved upon by the doctor’s ‘experience.’ How do we build this into the machine? That activity is called ‘machine learning’ (ML). Initially, the algorithm is fed examples of data whose results are known. In the medical diagnosis case, the logic of the doctor arriving at a decision is supplied. The algorithm notes the difference between its ‘prediction’ results and ‘correct’ results and tunes weighing of inputs to improve the accuracy of ‘predictions’ until they are optimized.
THE AMAZON INFO
When you buy a product from Amazon, you get lots of additional information on similar items. Based on your history of purchases, a suggestion of things is listed of what you are likely to be interested in. As you buy more items, this list gets modified.
A traditional machine-learning model works well so long as input data comprises features directly relatable to the desired output as in the earlier example of Amazon purchase. However, when the number of features to be considered becomes large, relating all the input features becomes difficult. In the above Amazon experience, if one also wants to take into account the sex, age, income bracket, family size, location, etc., of the buyer, the features to be considered become very large. Feeding logic to arrive at a decision becomes complicated. So came the next development: to let the system study the pattern of the data and work out the logic and arrive at the result. This is ‘deep learning’ (DL).
Our own brains learn to do difficult things through practice and feedback. A child sees the picture of an apple, points to the mother a tomato in the shop as apple, the mother says “no,” then the kid recognizes the fruit like an apple and receives confirmation from the mother. ‘Deep learning’ uses the same approach.
In the case of ML, the algorithm is told how to make an accurate prediction whereas in DL the algorithm is made to learn that through its own data processing. To enable this, a layered structure of computing systems called ‘artificial neural network’ is engineered, the design of which is similar to the biological neural network of the brain.
To sum up, AI denotes a broad umbrella of hardware and algorithms encompassing ML as its subset, which in turn, has DL as its subset.
What has given impetus to AI in recent years are the advances in the fields of specialized hardware, such as improved algorithms and availability of massive data from various sources. This has enabled AI to find application in many areas like games. Some examples are in the fields of reasoning (games), knowledge (a medical diagnosis), planning (demand forecasting), communication (real-time translation of spoken languages), perception autonomous vehicles)…
For now, we may not know where AI will take us. But in the coming days, all of us will be impacted by AI.