Desperately needed: An Indian AI giant

We urgently need an Indian company of large size, like TCS or Infosys, or even Flipkart, focusing on Indian AI.

What is Indian AI?

Indian AI is an existing collection of data sets, pre-trained models and networks, cloud hosted APIs and available AI related services that can be used to create India specific AI products. Let’s see the individual aspects of this idea.

Indian data

AI today is a game of data – not just any data, but humongous data. Machine Learning models (including Deep Learning networks) are trained using large data sets. Here are some examples:

Images: Companies like Google and Facebook have a huge pool of images. They have trained Deep Learning (DL) neural networks with these images. The trained networks are being used to recognized objects from images.

Speech: Google, Apple, Amazon and others have built trained DL networks using a lot of speech data. These networks are available either through APIs like Alexa and Google Speech or programs like Apple’s Siri.

Text: Over the last many years both companies and universities have built very large databases of words and phrases in English and a few other languages. These are used in any processing of natural language.    

An important aspect of the data to remember is that it has to be ‘labeled’. An image of a cat, for example, has to be tagged as ‘cat’. You can imagine that labeling of data, especially in things like speech, requires huge efforts.

So where does Indian data come in? For that, we have to understand how AI application building works. Companies like Google and Amazon have created the trained models, as we have seen. These giants then make them available either as free and open source software, or as paid APIs. There are even (at least partly) free APIs, such as Google Speech. The startups then use the models to develop actual solutions that we read about every day. Of course the big companies also make products, but their products are more generic in nature, or help their own business. It is the smaller and energetic tech startups that can create a far bigger number of solutions, but they rely on this foundation.

So what’s about Indian data again? Startups looking to create India focused solutions do not find models trained with Indian voices, Indian languages or faces. It is beyond a startup’s capability to acquire and label large dataset that are required for training. There are so many languages, accents, foods, actors, books that need to be trained. This has to be done by some giant Indian AI company.

Indian cloud

AI products need cloud for two reasons. They need to be hosted on a cloud that can provide resources like GPU. The products also need to call APIs of various types, as we have seen. These APIs are themselves hosted in datacenters of the provider companies.

Currently, the prices of both these recourses are unaffordable for the Indian business. Both the hosting and APIs are provided companies with bases in North America and Europe. The pricing is usually in line with the business conditions in these developed lands. If a startup wants to build a chatbot using IBM’s Watson or Amazon’s Alexa, it will most probably find the cost prohibitive for Indian customers. Or if another startup builds a Deep Learning solution that uses GPU on AWS or Azure, the monthly bill will be too high to get a good margin.

We need an Indian AI giant to provide these services at Indian cost. Cloud is central to the AI industry and getting one at the right price will push proliferation of AI in India forward.

Indian talent

Today, there is an enormous interest about AI in the colleges. In my company Cere Labs, we get a flood of applications for internships and jobs. Many colleges are thinking about starting courses in Data Science (which, though not strictly AI, is an important link to it), Machine Learning and Deep Learning. Robotics courses are already popular in many institutes.

Tasks like machine learning require very talented engineers and statisticians. I know many students who have what it takes, but many of those will be going abroad for post-graduation and later job. However, given good opportunities, a lot of them will stay. Unfortunately, startups in India are not considered as good opportunities, especially by the families. A large Indian AI company will help to retain this talent which will power our AI ecosystem for years to come.

Who can do it?

India boasts of a number of large home grown IT services giants such as TCS, Infosys, Wipro and so on. We also have the new consumer internet cash-rich startups like Flipkart and Ola Cabs. They are all engaged in AI research and development. It is (at least it seems) easy for them to start working on becoming an Indian Google or Facebook for AI. A lot of research on Indian languages has been done in universities such as the IITs and public research institutes such as C-DAC. Monetization of this research is required to make it useful.

There are also many funds started by visionaries and ex-captains of our industry. There is an ample space for a non-profit organization such as Elon Musk’s OpenAI, backed up by one of these funds.

Conclusion

India is becoming a bigger and bigger market for technology. In the past, we never made operating systems, did not create a browser or a mobile technology, though we were a big market for all of these. AI may just be our chance. We desperately need an Indian AI giant.