AI What it means to business leaders by @MohanSawhney at TiE Chennai Charter Member Meeting
Launching a course on AI for business transformation next week and is open to all. Focusing on the now what and so what of AI
Leaders should know the right questions of the data scientists. This is a C Suite conversation especially due to the accountability & governance rests with you all
AI - a Program that can sense, reason, act and adapt. ML a subset of AI - algos whose perf improves as they are exposed to more data over time. DL (deep learning) subset of ML where multilayered neural networks are used to mimic the human brain
First questions to ask is where is the data, how are you collecting the data, what’s the training data
Speech recognition, image processing/machine Vision, text is mostly using deep learning. If you search google photos for images of cats it’s using deep learning
Supervised, unsupervised and reinforcement learning - clustering is unsupervised. Supervised is where you provide a training data which are also labeled so that the machine can learn.
Then you take real data and do the prediction using the model you have trained. The model is going to need continuous refresh
What’s new - data explosion especially unstructured data, cloud computing, open source, improved algorithms, sensors
Jio is collecting 150 billion data points everyday (Mohan is on the board of Rel Jio)
Compute capacity is now 300000X of what we had 7 years ago. Tensorflow.. open source is also growing exponentially. When all of these exponential things come together it’s a massive thing
Talks about Tesla’s autopilot system and how advanced it is. Says in India it will take time - you have to model a cow’s trajectory (humorous)
AI changes the nature of our work so it’s a big leadership CxO level challenge
We need data scintists, IT, Domain experts, Partner ecosystem led by a Business Leader
Business Problem - how and where do I find the most promising use cases for AI in my business
Data - how do I get the data? Is it legal to collect it (GDPR etc) what about the Technolgy & Organization
Shows an AI Canvas - 7 elements - business problem, business value top 2 boxes. Objective Function (what is the dependent variable & predictor variables), Modeling Approach, Model Training, Customer Value, Data Strategy - nice thinking tool
Gives the use case example of predicting employee attrition. Shows all the feature variables (feature engineering)
Shows the filled out AI Canvas for the problem
Needs cross functional collaboration to pull it off. 3 models - CoE model (example firm Stitch Fix), hub & spoke (Anthem Healthcare) and federated (Microsoft)
Over time the org should move from IT led to AI led, use 3rd party tools to start and then as we get deeper then we may build our own custom tools.
Chief Analytics Officer is a new CxO role he mentioned. Standardize & centralize is the mantra. Uber has a platform called Michelangelo - it has the tools, data, models... you can create a hypothesis, assemble a model and put it in production. Democratizing AI.
Why AI initiatives fail? 87% of data science projects never make it to production - venture beat July 2019
If you don’t have a single unique Cust ID and a data lake, you may not want to start as it may not be effective. Solving the wrong or incorrectly chosen business problem - use AI Canvas and MVP approach (don’t do perfecting the PoC approach)
Don’t chase the biggest prize (in terms of biz value) because it might take too much time.
Says shipping is more suitable to autonomous systems. You can eliminate a whole lot of people. 92% of marine errors are made by human beings. That can be eliminated by autonomous systems.
Gives example of the bot gone rogue Microsoft tay. Another example - A predictive policing system keeps sending police to the same area over and over
Another example - Amazon’s recruiting tool eliminated women resumes because the training data based on past hires didn’t have enough women resumes
Moving towards Explainable AI so that the decisions made by AI can be reviewed and corrected. But this introduces additional perf considerations.
Gives example of Espoo, Finland city as a service AI based platform
Key takeaways 
