🤔 It's all fake! Cashierless stores don't work like you think they do.

In today’s newsletter:

• It's all fake! Cashierless stores don't work like you think they do. — And it horror of Jane Austen

• Papers

• Creepy or no?— are AI chatbot agents for dating creepy?

• Learn

Penses

As a data scientist, I often smirk at tech company headlines. Mostly because I know that after all the hype the ball will drop and the truth will come out. A lot of promises made in public don't really hold up to technical scrutiny.

Cashierless stores like amazon's Just walk out. Don't actually work like you think they do.

Cashierless stores are run by 100+ cameras with people watching the feed in India and charging you for the items. 😂

This is because most object detection is a really hard space. For a purchase to happen, every action of you picking up and placing down an item must be taught to a computer. That means that the data has to be labeled for a computer to understand what is happening. In order to get that data, Amazon hired a staff of 1000 people in India to monitor fees, label the data, and review transactions to ensure you were charged. Unfortunately, they were unable to meet the milestones they set to remove humans from the review process.

This leaves them to pivoting to just providing carts with check out screens for you to use instead.

Read more on this here.

Data labeling is one of the biggest hurtles for image based AI technologies to overcome. Healthcare startups that predict disease do the same thing as amazon. Grad students, doctors, etc must review the image so that a machine can learn how to interpret it. This leads to mislabeling which leads to worse predictions in the model output or rather how well the model can actually predict the disease.

The over inflation of the capabilities of AI has lead to the current AI bubble that we're living in. Instead of providing good practical statistical applications, most AI projects just like the Just Walk Out stores fail.

The reset of the AI industry will happen in the next decade. I think it's a great opporuntity for innovators to build interesting companies with practical AI.

For more on the AI bubble listen to this.

Learn

More on data labeling