Lata Iyer, VP of Research at Rakuten India, opened her session “AI for Human Empowerment”, at The Rising 2022 with an interesting anecdote about Alan Turing. “When Alan Turing found the first Turing machine, the whole program was to demystify and decrypt the Enigma machine. And he actually reduced wartime, because a lot of these messages could have taken 3-4 years to decode,” she said. The Turing machine, in a lot of ways, contributed to the growth of AI. “In the last two years, meeting everyone from home conveniently or ordering in food would not have been possible without AI,” she said.
Watch all the recorded sessions of Rising 2022 here>>
How businesses can work on a problem
Lata discussed the importance of starting with a problem and seeing it through production. She then spoke about the challenges of working on a real-life project. “When you work with a large data set and put it for production, the challenge is that you go through the CI-CD pipeline, you go into the deployment stage, and once it has gone into production, you have to do continuous monitoring,” she said.
The transformation brought by AI has been so pervasive that it is deeply influencing user experience and how humans interact with brands and technologies. More and more new algorithms will be developed to handle complex data, learn deeper with the lesser size of annotated data sets. Most of the NLP tasks will be supported by large pertained transformer BERT/BART models along with Neurosymbolic AI study to bring more context-awareness supporting reasoning layer. This shall reduce the effort of training text or images through all possible orientations, explaining causation effects and reasoning beyond learning.
At Rakuten, the DNA platform helps understand customer data. The platform collects data from all places. Rakuten uses a neural network technique to understand a customer’s touch points. “Having all kinds of data studies across all the different paradigms gives you a higher performance,” Lata said. She also recommended organisations to do moonshot exercises and patent and publish their ideas. Patenting and publishing “protects your intellectual aspects and helps you find out whether you are ahead of the curve,” she said.
Power of data
Further, Lata spoke about the power of data. “You have your data in your control. You can monetise this data externally,” she said. Multimodal data enriches customer recommendation because it allows for better customer understanding and personalisation. Lata illustrated it with an example. “On one hand, I have my customers whose purchase behaviour I know,” she said. This behaviour includes information about their social circle and preferences through chats. When this information is expanded to billions of customers, it is no longer a linear problem, and the massive amounts of data allow for the right prediction and better output.
“If somebody promotes something to me, they would know it based on my persona. They could also suggest something for my community because they would be able to track all that information for me,” she said.
Rakuten has prepared a graph network model with 5000 customer attributes. The company further used this data in their email commerce platform to optimise the banner better.
While the opportunities of AI are great, there are risks that need to be addressed. Datasets and algorithms can reflect or reinforce gender, racial or ideological biases, she said. When the datasets (fed by humans) that AI rely on are incomplete or biased, they may lead to biased AI conclusions. Moving forward, we need to have a very good understanding of international standards, technical specifications and requirements which will help to build AI technologies and solutions that perform well which are reliable and transparent especially in the highly regulated medical and finance areas, she concluded.