How to create an AI solution: Some top minds explain

How to create an AI solution: Some top minds explain thumbnail

Synopsis

Machines are getting smarter and smarter as Artificial Intelligence experts make them work for different things in your life– from driving your car, to teaching you languages., to even choosing this article for you to read. They can get it right, they can get it wrong, They can be useful mostly or irritating at times. But you just want more of them.

iStock
Machines are getting smarter and smarter as Artificial Intelligence experts make them work for different things in your life.

Machines are getting smarter and smarter as Artificial Intelligence experts make them work for different things in your life– from driving your car, to teaching you languages., to even choosing this article for you to read. They can get it right, they can get it wrong, They can be useful mostly or irritating at times. But you just want more of them.

How are these AI solutions created? TOI spoke to some top AI practitioners to understand how it works.

Srikanth Velamakanni, co-founder & CEO of Fractal Analytics, said it has to begin with framing the problem well. “In autonomous driving, you can frame the problem by saying that at any point of time, the car must accelerate, decelerate, turn the wheel left or right, and so on. And I need to make sure these decisions are made every few milliseconds, and these decisions are made in a way that human beings are not hurt, and that all traffic rules are being complied with,” he says.

Shridhar Samantaroy, architect – decision analytics at Pegasystems said most people believe that AI algorithms make machines smarter, but it is in fact the quality of data that determines how efficient the AI can be.

Prithika Priyanshi, data scientist senior principal at Accenture Applied Intelligence, recalled building an AI solution to find customers for a lending business. “If the customer needs a consumer loan to get married, we expect s/he will go to a wedding website for clothes, to a website to compare loan interest rates, and may physically visit a jewellery store. So, we could get the web browsing data. We could get geo visit data from vendors who capture your offline footprint based on your apps,” she said

Samantaroy then spoke of how a data scientist spends a lot of time cleaning the data. “They check to see if there are discrepancies in the data. If data is not properly represented, then there are statistical techniques to make the data more representative. Say, you need data about Patna, but there’s little data about it. Then you could take data from a city close in demographics to Patna and see if that can be extended to Patna,” he said.

After the data is ready, the algorithm is built, Priyanshi said. She added that there are several types of algorithms that cater to various needs. Abhilash John, director of engineering at Byju’s, said there are lots of pre-existing algorithms you can choose from. “But often you might have to tweak these algorithms, sometimes a programmer might have to write the algorithm from scratch,” he said.

Velamakanni added, “There’s a lot of engineering, and innovation required. Similarly, how you combine algorithms, how you put it on data, that’s key to the AI solution.”

Samantaroy said you need a background in mathematics, and preferably a good understanding of statistics, to understand data, clean data and find patterns in the data. Building the model, he said, requires knowledge of programming and algorithm complexities. And to frame the problem well, you need to know the business well.

(Catch all the Business News, Breaking News Events and Latest News Updates on The Economic Times.)

Download The Economic Times News App to get Daily Market Updates & Live Business News.

ETPrime stories of the day

Leave a comment

Your email address will not be published. Required fields are marked *