Debunking
AI
Nishan Pantha / @nishparadox
GitHub/ NISH1001
nishanpantha.com.np

What This Talk About

AI from a non-technicalperspective

Myths and Hoaxes

Current Situation

But first...

what is Artificial Intelligence?

Building systems that can do intelligent things

Intelligence is

Perpective

Psychological

Ambiguous

Map

Translate

Alpha Go

Youtube's Recommendation

Facebook's News Feed

Photo Manipulation

Drones

Text Generation

Face Recognition

Other

  • Spam Filter in Email
  • Google Search
  • Weather Prediction
  • Games

How does a Machine Learn?

Machine Learning :D

one of the mechanisms to AI

AI Hierarchy

Building systems that learn from experience

Different methods used to LEARN

Machine Learning

Supervised Learning

- Train the system

- Student learning from a teacher

Cat vs Dog

OCR

RIP Tay Tweets

Hot Dog or Not?

Supervised Learning

Unsupervised Learning

- Let the system figure out the patterns itself

- Students learning on their own

Usage

- Topic modelling

- Unusal activity detection

- Grouping similar texts

Reinforcement Learning

- Rewards based learning

I don't know how to act in this environment, can you find a good behavior and meanwhile I'll give you feedback.

Myths

AI is all about making machines that can think

Not Even Close

AI won’t be bound by human ethics

AI will spin out of control

Autonomous Systems only do good for what they are trained for

AI will be a series of sudden breakthroughs

Slow and Steady

News Media Be Like

News Media Be Like

Artifically Intelligent?

Perhaps, NO

Should We Fear?

No. But Yes in some ways due to social media.

AI has become a misnomer

Where Are We?

Artificial Super Intelligence

Perhaps, Never...

Artificial General Intelligence

Not quite sure we will make it there...

Artificial Narrow Intelligence

Definitely, here...

Reality

Still a baby

Data. Data. Data.

This...

Precise Segmentation

Real Time Object Detection

Baidu’s AI Mimicks Voice

Aid in Tumor Diagnosis

Aid in Surgery

Software 2.0

-Andrej Karpathy

What should we fear?

This?

NO!

Faking using Deep Learning

Personalized Influences

Adverserial Attacks

Confusing Computer Vision

Experimentation

Learn. Share.