I thought of the same reply but after giving it some thought...

I thought of the same reply but after giving it some thought based upon what I understand from all my readings I think it comes down to purpose and training.

If you had a reason to have AKIDA recognise a Rozella from a Budgie or a Zebra Finch from a Wedgetail eagle then it can be trained to do this in the initial data set. If later you had a problem with Crows you could then just add a Crow photo as per the demonstration and AKIDA will recognise a Crow from all of the other already trained birds.

Herein lies one of the revolutionary aspects of AKIDA as it is impossible to add another bird with one shot learning with any other system.

To make it clear you would have to add multiple pictures of the Crow to the original data set and then retrain these systems from scratch all over again.

An example .where this could prove useful is airports. Birds such as Seagulls are a problem at Mascot. A drone system is set up to react to the presence of Seagulls and eliminate the threat. To do this using deep learning the drone camera is trained for hours on Seagulls. The drones work brilliantly.

Then one day as a result of the drought in Central Australia thousands of white Cockatoos descend on Sydney and the word goes around that Mascot is a great place and hundreds turn up. The drone not being trained on Cockatoos does not recognise them as birds because of their funny wobbly walk.

All international flights are grounded while the drones are taken out of service to be retrained from scratch.

If the drones cameras were serviced by AKIDA one drone would be shown a picture of a Cockatoo it would learn to recognise a Cockatoo from one shot and then share this learning instantly with all the drones.

Crisis averted.