Coverage by Bhat Dittakavi of Variance.AI on “AI disruption in Education” at CIE IIIT Events room
Abhilash: I met Dr.Shailesh Kumar at Google as intern.
Dr. Shailesh
First leap: Abacus to calculator
Second leap: Calculator to computer
Third leap: Computer to AI
Evolution of intelligent machines
Counting and taking averages all is business intelligence. The way we can do similar Web pages, outliers and so on. This is unsupervised learning. Then there is structure on data. Can we find cause?. Can we do something about it? Can we convert prediction and a series of actions like a thinking machine. Not just one but a series of actions.
Once machine learns to think, can machines teach us how to learn?
A teaching machine
Knows and learns how to solve problems in alternate ways to help students learn to think and not just learn to solve.
Understand student’s proficiency and temperament to adapt to curriculum and what is being taught.
How do machine understand a pully diagram? Understand images, formulae, handwriting and so on.
Reason using knowledge, exploration, optimization and learning.
Personalise using solution, content, problem and solution.
Communicate using emotion, gestures, conversation and speech.
Do all the above by teaching machines.
Purpose is “learn to think, not learn to solve”.
“Do exactly as I say” led to 100000 IIIT graduates but no Nobel winners.
Education is not the learning of facts, but the training of the mind to think.  -Albert Einstein
And once we could build a machine that thinks, surely we could build machines to teach humans how to think and how to learn.  -Audrey Watters
How to get machines learn to think?
What is thinking?
Say finding route on a map is thinking where map itself is knowledge. Knowing that rules is knowledge and playing the game is thinking. Knowing Grammer is knowledge and knowing what to say is thinking.
In how many ways you can simplify a polynomial multiplication?
Many. Simplest solution and shortest path!
Knowledge gives options. Learning helps you choose the best option.
Adversarial environment needed for infinite learning.
Deeper step-level understanding
If we teach students through error correction and scoring, student doesn’t know how he got the correct answer.
Parents can’t solve student’s problems. Peers laugh. Can we have a system to help students.
Root cause analysis and even Bayesian reasoning!
If ads and news can be personalised, why not the curriculum?
How do you choose which book or YouTube to choose from 1000s? Content personalisation.
Can student pick his teacher?
Can machine match learning style with teaching style?
Problem adaptation
Think of this as a state machine. You go from one state to the next state.
Experience of a teacher versus experience of a teacher at planet scale! This is where big data helps.
How do you build a machine that captures 100 billion math steps of students data per day? The biggest big data problem!
Using our math app, third leap, We don’t solve problems but we solve templates.
Goal is fixed and the effort is different as against today’s teaching of “effort is fixed and goals are different”.
Real-time, adaptive and personalised using mathematical reasoning engine.
Q) Challenges from machine learning?
Profiling of the student.
Q) Any conditional probability as part of adaptive learning?
Yes. Build profiles that are probabilistic in nature.


Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s

%d bloggers like this: