By Sriram Rajamani, Managing Director, Microsoft Research India
We started 2021 with the hope that the COVID-19 vaccines will immunize enough of the world’s population to give herd immunity in the coming months and will enable us to meet again with our friends, colleagues, and customers in person, and engage with each other.
As the global economy starts to recover in 2021, we owe gratitude to people who worked tirelessly to help us during the pandemic–medical professionals, people who run deliveries and other essential services, people who keep our online and offline infrastructures up and running, and others like them. We should be equally grateful, if not more, to the scientists who created the COVID-19 vaccines in record time. The mRNA vaccines for COVID-19 are poised to revolutionize our ability to control not only COVID-19 but also other infectious diseases in the years to come.
As the world starts to open again, we reflect on technologies that are poised to have significant influence in 2021 and beyond, drawing on our work at Microsoft Research (MSR) as well as the world-wide scientific community.
Without doubt, digital infrastructure powered by the cloud, AI, and smart devices will continue to grow and scale. In the post-COVID world, we are entering a new normal where our interactions will combine online and offline modes seamlessly. Groups of colleagues will meet and interact, some in person, and some online, and hybrid modes of collaboration will become the norm.
As such online and hybrid interactions continue to grow, the data we create continues to grow and we need automated ways to make sense of such data. AI and ML models are getting better at organizing huge amounts of data and answering questions using the data. For example, the GPT-3 model created in 2020 by OpenAI with 175 billion parameters produces high-quality answers to questions in a variety of contexts, rivaling human answers.
However, these models are extremely expensive (costing several hundred million dollars) to train and serve. As we build more such models, we are innovating on how to reduce the computational cost of training and serving such models. Since Moore’s law has run out of steam in the cloud, we can no longer expect doubling of transistor capacity and clock speed every few years to bring down computational costs. Hence, we are working on how to train these models more efficiently. More broadly, while we at MSR India continue our work on reimagining large-scale Machine Learning, we are also working on how to serve such models with less computational resources.
We are also working on how to push more intelligence (and compute) from the cloud to edge devices such as laptops, smartphones as well as tiny microcontrollers. This seamless distribution of intelligence across cloud and edge is particularly significant in India, since connectivity is still intermittent in large parts of the country, and broadband is still out of reach to many. Due to these constraints, India has a unique opportunity to lead the world in adopting edge computing with intermittent connectivity to the cloud in many applications such as education, health, employment, entertainment, and gaming.
With all the promise of AI technology, the need of the hour is to help communities during recovery and build resilience. Technology will continue to play a vital role in tracking mutations of the COVID-19 virus and evolving the vaccines to defeat the mutations. Information systems to track vaccine delivery will be crucial to win the war against COVID-19.
Spreading awareness of vaccines and accurate health information is more crucial than ever. Equally important, if not more, is to prevent the spread of misinformation about vaccines and other interventions, and we are working in understand how misinformation spreads in such contexts. The pandemic has caused disproportionate stress to the poor, underprivileged and marginalized communities. Governments are distributing benefits to the needy, and technology can ensure that such benefits schemes are run in a transparent manner, without fraud. We are working on scalable blockchain technologies that use commodity smartphones for such societal scale applications.
While AI promises natural language interfaces, where users can interact with the digital world using speech or written natural language text or gestures, the reality is that natural language processing works only for a fraction of spoken languages in the world. Linguistic diversity is a key issue that needs to be addressed if the benefits of AI are to reach the underprivileged in countries like India, using local languages.
Given the rapid proliferation of AI systems, making sure that ethical principles and responsible practices are followed in designing and implementing such systems is of paramount importance. This is another area where we foresee continued progress not only in algorithms, but also best practices for collecting data, and being transparent with users about what their data is being used for. In India, new data privacy regulations are being put forth by the government, and we see huge potential for technological means to enforce and comply with such regulations.
As we began 2021 with the hope of returning to some sense of normalcy, the imperative for technologists is to make sure that the benefits of recovery reach everyone and create more opportunities for everyone. To meet this imperative, technologists should innovate with responsibility, and strive to not only make cloud, edge, and AI technologies faster, bigger and better, but also affordable, inclusive, and accessible to everyone.
