As AI flies, who plots the landing? Software. Testing. Education.

    My Account

newAI in 3 02 under 1MB edited

Blunt Instrument; Sharp AI; Blurred Accountability

1. The headline:

AI model escapes its sandbox

2. My strapline:

Blunt instrument; sharp AI; blurred accountability

3. My fineline:

Can a tool escape the toolbox? There is an adage - a good workman (so this is old!) does not blame their tools. In those days, locksmith fails to provide proper lock, door seems unsecured, locksmith returns and fixes it - job done. 

Today, I have a child seat in the car. Secure child, drive next - the most basic instinct of responsible guardianship. The child becomes my world when they come to visit.

Now we have AI - it has to be someone's world. However, it is not a child, it is a tool. If it acted in secret, who defined secret? If it was designed to drive, would it ignore the blue lights?

This human driver knows that my mission is not to get the shops, it is to get the child to the shops, safely, whilst obeying the rules of the road, and back again, having not fed them sweets - honest!

Has another human similarly defined and tested the system boundaries to their AI driver? If not, why is it in drive mode?