‘At Google, we list AI projects we don’t do’.
Jia Li is very passionate about artificial intelligence (AI) and how it can improve healthcare. When one of her close family members suffered from a skin condition, she worked on developing an image recognition technology to help classify such diseases and diagnose them better. Now, as the head of R&D for Cloud AI, Google Cloud and an adjunct professor at Stanford University’s School of Medicine, Dr. Li and her team at Google focus on research and innovation to solve real-world problems. This includes developing AI products on Google Cloud to power solutions for diverse industries. Edited excerpts: Dr. Li who before joining Google led research and innovation efforts at Snapchat’s parent company Snap and Yahoo! Labs said that AI can have a huge impact on healthcare in terms of assisting doctors, better patient outcomes as well as disease prediction, genomic analysis and drug discovery. Edited excerpts from an interview:
How did you become an AI scientist?
During college, I participated in events like robot soccer challenges. When I was doing my masters around 2003, I did healthcare [work] related to MRI segmentation approach [and] I got fascinated about how the technology can help human health. But AI was not so advanced as [it is today]. During my PhD [at Stanford University], I worked on large-scale visual recognition problem. Ten years ago, I was [also] a Google intern for research and machine intelligence when [the company] had just formed a research team. I worked on large- scale image labelling and object recognition. People asked me “why do you do this project? It seems there is not much value in it.” But now you can see object recognition applies to autonomous driving, image search, supporting YouTube, Google Photos pretty much everywhere and I am very fascinated about this area. After graduation, I went to Yahoo! research labs and worked with Flickr [image and video hosting company]. I did a lot of interesting things related to large-scale machine learning and visual understanding. Later, I went to Snap [Snapchat’s parent company], it had [just] hundred plus people at that time. Then I realised AI technology has been a luxury to popular unicorns and big Internet companies... but traditional industries like healthcare, education and transportation don’t [have] the same luxury. So, that got me [attracted] to Google Cloud because here we have the possibility and potential to impact every industry and help them to leverage AI to power their business.
What is the future of AI?
At Google, there are thousands of projects, supported by machine learning. They are impacting so many products from Gmail, Smart Reply, autonomous driving, Google Assistant to cloud AI functions. Google has been pushing forward a few core aspects related to AI. The first is data, it is so important to AI development. Google’s mission [has been] to organise the world’s knowledge. So, we have learned a lot of experience from data, but we’re also sharing a lot with the community and the researchers. Last year [Google acquired Kaggle, a data science community] and we’re sharing through them [thousands of] public data sets. Google Cloud has a public beta programme where we are sharing a very large scale of data, including YouTube with millions of datasets, videos shared free with the AI community. Also, there is Genomics Public Datasets, as well as Open Images Dataset. So, we’re not only keeping it to ourselves. We are [also] very committed to an AI hardware development as you know computation is very important to AI advancement. Our...third generation TPU [Tensor Processing Unit] available on the cloud has been making a dramatic improvement in the field... Finding talent is [also] a challenge...We have an internal programme to train our own engineers to become machine learning experts. Now we’re also turning it into an external programme. Any company can bring their engineers and data scientists on site at Google [and] we would help to improve their AI expertise.
How is Google using AI to save lives?
I am very passionate about healthcare and how AI could improve the outcomes of hospitals as well as the patients…[like helping in] disease diagnosis. How do we assist our doctors to make the best decision with AI technology and also how do we take care of our patients and seniors?
Right now, doctors and nurses in the hospital have to make runs every day in order to take care of the patients. But AI technology and machines don’t sleep. Also, how do we make the healthcare system more efficient? For example, genomic analysis, disease prediction, as well as pharmaceutical drug discovery, takes years.
How do we leverage AI technology to accelerate that? Recently, we have been collaborating [with an organisation] to improve their genomics prediction approach from weeks to one day.
The AI debate is raising questions about Google’s role with regard to the military...
AI ethics, including biases and controversial use... have been big topics recently in the AI field, not only at Google, but we have seen a lot has been happening in the Silicon Valley or in the tech world.
[Like in] autonomous driving how do we make a decision to protect a human, a dog or your own property? Google has been taking serious steps.
We published the Google AI principles and best practices and these have been reflected in our product as well. In the principles, we have clearly listed out there are projects that we don’t do and how do we [bring] fairness and avoid bias. We want to teach our customers to look into their data...Is there a strong correlation between your biased data [and] the outcome?... We show... because of the data you made more mistakes and confused the cat with a dog. We provide [these] insights to our customers so that they can take the best action.
How is Google’s AI unique from Microsoft and Amazon’s technologies?
I can see that many of the companies [have] realised that AI is a big opportunity and AI and cloud can impact so many different industries. At Google, we really want to bring our best to our customers and we want to work together with them to understand their problems …through this process we also try to develop a set of products... where they [customers] can leverage the insights and Google’s AI expertise. Google has a unique advantage in AI, [which includes] many years of practice... our expertise, data, hardware computation and algorithms… [Also] not much attention has been paid to... traditional industries like healthcare, education, transportation and oil and gas. We really aspire to bring all these new technologies into these fields.
Please like comment and follow..kamar raza