At IBM Think 2018, Shantenu Agarwal, Lead Offering Manager for Visual Recognition and Donna Dillenburger, IBM Fellow, spoke on using AI for Good, Addressing Global Challenges with Visual Recognition. They highlighted a variety of ways in which IBM teams use visual recognition to combat issues around sustainability and fraud.
AI and Social Good
Statistic: 1.2 trillion photos were taken in 2017
Statistic: 85% were taken using smartphones
There are a lot of global issue like no poverty, quality education, zero hunger, decent work, economic growth, etc.
- You can teach Watson to be a Forestry expert, clean water expert, etc.
- Watson with clean water can do image inspection without having to drive around everywhere.
- Watson can be an expert in nutrition. It’s looks at the food you eat and tell you about it. Take a photo and get a result.
What is Watson Visual Recognition
Visual Recognition focuses on 4 categories:
- Identify objects and people
- Categorizes the object for easy organization
- Assesses the image for better problem solving
- BMW assesses car parts
- One insurance company assesses damages in a car
- Recommendation: what should you do to fix it.
Examples
Environment
One company built an app in Europe to report trash bins overflowing. Citizens submit photos in the app. Watson reviews and recommends the right action.
Another company has a damaged pipes app. They take pictures and then allow Watson to assess.
bioBotGuard by Blue Digital applied drones to farming. The drones take pictures and then provides an assessment to the farmer.
Another app in a hackathon uses Watson to identify fake fish. Too many restaurants use different fish than what was advertised.
Migration Paths and Human Trafficking
How can you identify movement and whether it’s illicit?
Wildfire prevention: Use visual recognition to recognize where brush and trees need to be cleared
Watson Verifier
It’s a cell phone. Attached is an optical device. Verifier takes visual recognition AI models and lets you classify small things:
- Drug, win, are luggage
- currencies
- manufactured parts
- DNA identification, biological an
- Detect water pollutants
- Plants, seeds, fruit verification
- Oils (motor, olive, palm, etc.
Demo time
- Used her iPhone live.
- There is an actual optical device and put the app on
- She held has motor oil, extra virgin olive oil (diluted with regular olive oil)
- took an image and put the data on the block chain
- in the app, labeled the scan
- captured the very up close image
- captured the characteristics
- then sent it to the blockchain
- showed the actual block chain record
- Then acted as a consumer who wanted to check the olive oil
- take the verifier
- Get the block chain record (or scan it on a label???)
- Capture the image
- human eye would say it’s the same. Watson said that without a doubt, it wasn’t 100% virgin olive oil
What about drugs?
- Debra showed a video of using the Verifier with Asprin
- One for Bayer
- one for generic brand
- The Watson graphs were a little different
- The optical signature had some overlap but also some unique elements
- Watson found differences in the drug imprint
- Bayer had deeper “aspirin” imprints.
Then looked at Ibuprophen
- Advil vs generic
- Under the Verifier, Watson was able to recognize the difference.
Labels?
Watson is able to analyze the paper weave and the color spectrum.
Metals?
Watson can do checks of steel to verify the grade of steel, copper, etc. The Verifier can see different striations in the metal. The Verifier was able to tell the difference between copper and brass in the demo.
DNA and Biologicals?
At 1 micron the Verifier can identify bacteria, plant cells, human hair, and even gmo molecules.
Bottom Line
There are some incredible possibilities for visual recognition across a wide range of industries. They can provide a lot of value. Frankly, I was blown away by what the Verifier could see and use Watson to recognize.
