Bhat's Notes

Coverage by Bhat Dittakavi of Variance.AI on “Future of AI” at GES on 30th November 2017

Moderator: Rama Kalyani Akkiraju, IBM, Watson

Nivruti Rai, Country General Manager, Intel

Elizabeth Gore, Chairman of Alice, Dell

Shubha Nadar, Sr. Director, Data Science, Salesforce

Nivruti: Proud to be on an all-women panel of AI. Bank of America uses chat bot for customer care. One old Japanese woman I know uses chat bot as her companion. Leading cause of death for people aged between 18-24 is road accidents. AI can be used to minimise this. I also would like to avoid human made errors in healthcare using AI. Today we talk about how AI impacts society and industry. I am into human computer interactions. We want to understand sentiments, emotions and personality traits. 

Elizabeth: Use AI to predict barriers for founders based on location, stage and many other parameters. We coded Alice to help them get over those barriers. All of us have unconscious bias. I am excited about diversity and inclusion that AI can take into account unlike the biased humans.

Rama: That means you are using AI to help founders make more AI. Nice.

Shubha: We are at a great time of confluence of data, computing power and hardware innovations. Transformational diagnostics, personalized medicine and access of machine learning to hundred of thousands of businesses that don’t have access to AI for example using Salesforce.

Nivruti: Three key things in AI.

Datasets: Huge sets of data of objects. Image dataset, auto auto dataset and so on.

Model Building: Frameworks. Fitting known variables to unknown variables. Math, programs, stats and memory.

Deployment: Where can we apply these models  built?

Stack: Hardware, math libraries, framework and platform and then applications on top.

Shubha: Explicability is the key.

Elizabeth: Entrepreneurs of AI succeed only when they understand culture and humanity. ATM doesn’t discriminate you when you withdraw money. It is not about the teller or VC on the other side. I still think technology is as dangerous as the people behind it. Technology and values have to stay together. Technology is used to solve human problems.

Shubha: Biases are big concern. The data may have biases. A VC built a prediction model on “which founders would succeed”. As all the data fed is of male founders, model thought success is associated with male founders. This is a clear bias. We must make sure our input is not biased. We may not be aware. Zip code could be proxy for race. But we shouldn’t drop those fiends that may induce bias. We need to include such fields so we measure the bias.

Rama: What do you mean by data management model? Do you know any models emerging around data?

Nivruti: Amount of relevant data we can generate in India is crucial. If all the cars are connected, we can track the road bumps and the times when the wipers get on and so on. Such data can be used for insights. Data comes from many sources and we need to make “data sense”. Ownership of data is key too. In Machine learning we have two  modes of learning: Supervised learning and unsupervised learning.

Elizabeth: AI engines have personalities as humans behind them have personalities. We did empathy test on engineers. Engineers must have empathy so machines get into them. I believe empathy and understanding culture is embedded in code.

Nivruti: MobileEye’s Founder said once “It has to learn the small wink kind of negotiation skills and nuances we humans do.”

Rama: Transparency. Do you need transparent models?

Shubha: It depends on the use case. Loan approvals and founder assessment don’t need deep learning. Hence explainability is better. Deep learning is important but don’t forget cheap learning.

Elizabeth: We allowed the chat bot to ask the discrimination questions as we trained. We need this to guide those founders to appropriate funding and government initiatives of set aside. Intent is the key. There is no right answer.

This is about data collection transparency.

Rama: AI in society. Read Dan Brown’s Origin. The climax is about man machine symbiosis.

Nivruti: Jobs go and jobs come. That is historical trend. Reskilling is important. There are different jobs in 1) datasets 2) model building 3) deployment.

Descriptive to diagnostics to predictive to cognitive. As Bill gates said tax the AI tasks to put that money for reskilling humans. There are jobs around deployment as a robot in China has become racist and a robot in Saudi made things interesting and so on. People in deployment jobs can help here.

Shubha: I ask the goverment to bring in training programs for citizens. We have, at Salesforce,  gamified learning platforms for Technology.

Elizabeth: I challenge the citizens to think about whom you elect as they make a lot of impact on you. If the elected officials don’t have technology know-how we are self-creating trouble. If not, testing them is about damage control.

We talk alot about blue collar jobs but let us talk about new color jobs. -CEO of IBM

Disruption is good. AI transformation is good.

Nivruti: Government can open data.

Rama: Our grandkids will be driven to school by self driving cars. Corruption will be minimized. Old people have happy companions in AI. What is your take on future of AI? Rapid fire.

Nivruti: Women shall contribute to AI.

Elizabeth: Values to AI.

Shubha: Education. My daughter can never learn to drive. Weird.

Shubha: Personally identifiable information. Though often useless for machine learning. We don’t share the data of different customers.

Nivruti: Blockchain is needed.

Nivruti: Fear is justified. No unmake. AI is additional intelligence for humans.

There are both sides of views to each technology and whatever be the technology, it leads to job creation. Look at the history. Jobs will be there in India for the globe.

Rama: AI, industry, Healthcare, city, infrastructure, opportunity.

Shubha: Jobs are always there until we are done with solving all the problems.

Most of the jobs of the world are from small and medium size enterprises. So no threat of AI. SMEs can manage things better using AI as robots can simplify their tasks.