Quantum Neural Network that solves 10000 piece jigsaws from working memory in a human like way is a serious step toward thinking machines

Anyone interested in machine learning (ML) or Artificial Intelligence (AI) will have been seeing more and more talk of ML as it relates to Quantum Computing (QC). This is often called Quantum Machine Learning (QML). A lot of the talk seems like hype, but today Toridion Project a QML research company from the UK and Thailand demonstrated how QML can do things that no classical algorithm could hope to achieve by solving an arbitrary 10,000 tile picture puzzle by using short term working memory Neuroplasticity – a characteristic of human brains, not classical computers. The result is proof undeniable that quantum approaches to ML and AI have the capability to perform tasks that no classical approach can achieve, and outperform humans as the problem size increases.

A serious step towards truly thinking machines

Toridion have done a lot of talking about how approaches to ML and AI need to change if we are in with a chance of making computers think and process information in a human like way, but todays demonstration summarized below leaves you in doubt:

“If you think quantum doesn't have a role to play in AI, you are gravely mistaken”

The demonstration of QML based computer memory that can reliably reconfigure itself and solve problems with trillions of trillions of trillions of permutations in mere minutes is a serious step on the road to creating realistic thinking machines.

Solving a 10,000 peice jigsaw using Quantum Machine Learning

In a controlled demonstration, a QML algorithm was trained on a photograph of Barack Obama producing a tiny 16 byte quantum memory engram. The QML was then shown a version of the image that had been randomized using 10,000 2x2 pixel tiles – much like the 15 squares puzzles you played with as a child.

The QML algorithm was able to solve the puzzle and re arrange all the pieces to recreate the original photograph in under 20 minutes.

How many ways are there to place 10,000 tiles in a puzzle?

The basic formula to calculate the answer is 10000! , the “!” stands for “factorial” which means:

10,000!= 10,000 X 9999 X 9998 X 9997 ... X3, X2,X1

The answer is a LOT. There are more ways than atoms in the universe, but to give you an idea the approximate number for only 1000 tiles is shown below :-

Summary

We have demonstrated that QML can do something that really no classical algorithm can do, something that is much more a characteristic of human memory function than any computer.

The analogy is equivalent to a human being looking at the photograph on the box of a jigsaw for a few minutes then trying to complete the 10,000 piece jigsaw by never looking at the box again.

Caveat:

We have compared this demonstration to human problem solving rather than classical computing because as this test stands it would be impossible for a classical computer algorithm to compete in let alone win this challenge. Because the test requires the function of short term memory that does not hold a complete picture of the answer, a classical algorithm would not have a finished picture with which to compare its random attempts.

The best you could hope for on an equivalent classical approach is to perhaps hash the original image and then try to randomly arrange the pieces of the randomized image and see if the answer you have produces the same hash you saved. This strategy is however likely to never succeed with 10,000 pieces, as we have already shown the number of ways to put the pieces is many times more than atoms in the universe.