Q: What prompted Intel’s focus towards digital technologies like AI in healthcare?
In India, we have 0.8 doctors which means not even one doctor for 1000 people. There is a shortage of doctors and hospital beds. Having said that, what I’m excited about is the point A doctor for 1000 people and the expertise they have gathered by looking at the variety and volume of patients is amazing.
We have a very precious group of doctors, so can we look at being an expert system, using artificial intelligence, and make the doctors available for every single person that we have in our country? Let’s start with not just doubling our doctor capacity through technology, but quadrupling.
Now AI thrives on data, volume and variety. We need a collection of data, contextualizing and annotating data. I believe we have people who have the capability to drive this kind of data cleaning and annotating of data.
Then we have companies like ours, who can come up with the training of models and build an inference model which can then be implemented anywhere, it can be provided as a service. People will put in reports and it comes back with 99% accuracy of the diagnostics.
Many people in rural India are suffering because they don’t have access to doctors for simple healthcare issues like fever and headache. These models we are building through AI should answer the diagnosis and treatment. And if the symptoms are alarming, this tool can give an alert to a doctor. This way it will be an augmentation to the doctor’s bandwidth and be able to provide a doctor when it is required.
Q: What will be Intel’s role in the recently announced partnership?
The vision through this All.ai 2020 and the research centre announced in Hyderabad- INAI is that we want a disease-free India. We will approach in a stepwise manner, but to start with, we are looking at diagnostics. We have collected data over the last two years in partnership with IIIT Hyderabad, Public health foundation of India and Government of Telangana.
We are trying to build a standard platform for AI with open access. We are now focussing and collecting data for healthcare, smart mobility, and education. For health right now we have breast cancer, retinopathy, bone degeneration and slowly, we’ll build more. This will not only have a standard platform of input-output access or data, but there will also be annotated data for multiple segments and an open and secure data exchange platform.
Third, we will build some foundational AI algorithms and tools. Example: the breast cancer detection algorithm will be built on multiple breast cancer scans. So, it will be trained on multiple breast cancers and if a new scan is put in, it will be able to tell with 99 plus per cent accuracy, whether the person has breast cancer or not. The target is to reduce false positives and false negatives. False negatives are worrisome as that would mean you are told you don’t have breast cancer and turns out you have.
So, these may be the three collaterals that will be generated. This will be based on a very robust data governance and privacy security framework. The third layer will be access to startups, innovators, researchers and enterprises, to drive economic growth and value for human good.
We are all coming together as there’ll be a requirement for a lot of policies around safety, security around bias, so, the government has to play a huge role. There is a role for a company like ours because there is a need for CPU, GPU, FPGA, AI accelerators connectivity, ethernet, optical fibre, etc.
Q: What will be the crucial aspects for the hospitals to grip this digitisation?
Right now hospitals have begun digitizing data, it is still not done, diagnostics will be the second step, but hospitals alone will not be able to drive diagnostics, they have to partner with technology companies, academia etc and that is the last gap that we are trying to fill in.
Data is one part of it, whether its raw, unstructured data or curated data. The next part of it is IP (the detection algorithm) and third is technology. Then, there is a policy requirement. So, hospital by itself will not be able to drive this entire digitization and technology backbone into healthcare. We have to come together otherwise, yes, we can do it, but the pace will be like 20 years.
Q: Do you think standardising healthcare data will be a challenge?
I will be happy to tell you that in healthcare data, there’s much more standardization. It has a lot more standardization than legal cases and the output English which is up to a lot of interpretation, but when numbers are concerned it is easier and more standardized.
Once we build this platform, even the data exchange will be standardized. so if you go to different pathology labs, this platform will be able to access both and come up with a solution to your healthcare progress. Our platform will not only be looking at treatment, but also the management of health. This platform will be able to help you manage that data and you don’t have to be necessarily sick to leverage it.
Q: What are the latest innovations through AI at Intel?
The most recent is what we call as matrix architecture, which is the AI accelerator. Now this compute is meant to do the best performance on AI, whether it is used in cloud, car, or healthcare. We also have an AI compute which is so small that it can be leveraged on a watch or your glasses.
We have the ability to connect those multiples refrigerators of computing called servers or cluster of servers, through capability like ethernet, or optical connect. Then in networking, we have something called Smart NIC (network interface card).
Now not only are we putting the building blocks together, but we will also partner with the verticals to optimize for their use cases.
Lastly, I want to tell you that to do genome sequencing we need trillion operations. In the year 2000, 1 trillion floating-point operations would cost $50 million. So, eight times of those compute means 8 trillion and floating-point operations can be done in $600. One person’s genome sequencing could be done in a hundred dollars. That’s the kind of economics that the scaling of compute has followed through innovation and through the years of working together.