Coverage by Bhat Dittakavi of Variance.AI on “AI is the new electricity” by Andrew NG at NASSCOM ILF 2018
Andrew NG is the sole reason why I have participated in NASSCOM India Leadership Summit. I resonate with him that AI is not a threat for humanity and artificial general intelligence is probably hundreds of years away.
Andrew is world-class professor, disruptor, researcher, practitioner, entrepreneur and fund manager. As a world-class professor, he taught AI courses at Stanford University. As a disruptor, he pioneered free online courses through Coursera as an alternative to expensive classroom teaching. As a top class researcher, he has got more than 100 research papers published. As a practitioner, he sphere-headed AI research at tech giants such as Google and Baidu. As an investment manager, he raised $175 million to invest in AI startups.
Unlike other speakers who used a slide deck, Andrew delivered his keynote using a white board. It is very natural for his teaching instincts. He turned the venue full of 2000 delegates into a classroom session.
Electricity transformed the world hundred years ago. AI has reached a point similar to that. It will transform every major industry. There are lots of opportunities in the world. Even though AI brings tremendous economic value, 99% of this value comes from input to output mapping. We call it supervised learning.
- Input is email and output is spam or not. Input is loan application and output is will the applicant repay the loan or not.
- Input is an ad and output is to see whether you click the ad or not. (Online advertising)
- Input is audio clip and output is text transcript.
- Input is English text and output is Hindi text.
- Input is radar reading and output is position of the car.
So much value is created by input-output mappings. At Landing.ai, we input a manufactured part and our output is scratch or dent. Amazon’s Alexa is reliable now and hence it is getting adopted. Customer adoption is the key.
Data versus Performance
Take data on X axis. Data has grown over time on X axis. Rise of IoT creates more and more data. Take performance on Y axis. Small neural networks have outsmarted Traditional AI. To get high-performance, you need lot of data that is big data and you need to have reasonably deep neural networks. You could notice performance difference between traditional AI and deep neural networks as data grows in size.
Rule of thumb: Anything a typical person can do with less than a second of mental thought can be done by AI now.
AI is shaping the corporate strategy. Five years from today many CEOs will regret for not coming up with AI strategy in advance. IT revolution is first and then comes AI revolution. Why? IT creates data and AI uses that data to create insights.
Education adopted digital first. Then Health-tech adopted digital. Then fin-tech adopted digital.
Pattern: Blue River, known for see and spray agricultural machines, was sold at $300M to John Deere. Founders are students of mine and they went around and took a bunch of photos of cabbages. Then they built computer vision to assess these cabbages. These machines go up and down day and night and they accumulated a large data of pictures of these farms. They have the largest data set of plants in the world. It makes them very indispensable.
What makes an internet company
Shopping mall + website != Internet company ( here != means “not equal to” )
Just because a shopping mall has a website through which it does ecommerce doesn’t mean it becomes an Internet company. In order to be called an Internet company, the following three characteristics are needed.
-Pervasive A/B testing
-Short shipping time
-Distributed decision making
In internet companies, decision making is shifted from CEOs to engineers and PMs.
What makes an AI company
Traditional Tech Company + Neural Networks != AI company ( here != means “not equal to” )
Just because a traditional tech company is doing neural networks doesn’t mean it b comes an AI company. The following three characteristics make it an AI company.
-Unified data warehouse (this is the key. If we have 50 databases and we need 50 VPs to approve, we can’t have insights that are driven by multiple variables coming from these databases)
-New job descriptions (What worked in traditional tech companies won’t work in AI companies. We need to switch gears and create new job descriptions as old job descriptions won’t work anymore)
-Lifelong learning is the key. (e.g. Coursera)
-Build centralized AI teams
-Employee training and development through AI
-Curate content than creating content
Q) Is AI an existential threat?
Andrew) Narrow intelligence isn’t a threat. Progress in Artificial General Intelligence is very slow and hence there is no threat.