Proposal: for a random number generator.

A piece of radio active material is suspended inside a sphere.

the inside surface of the sphere is covered with a material that can detect impacts upon its surface.

Upon initialisation of the device, two impacts are recorded and the time between those impacts is recorded

upon the third impact a device records the co-ordinates of the impact upon the inside surface of the sphere.

The two numbers making up the coordinates are multiplied to give a number.

The resulting number is multiplied by the difference in the time interval between the first and second impact, thereby giving a random number

This number would be unpredictable as there would be no constant seed number, and each subsequent number would be unconnected to the previous number.

There would be no way of knowing (by looking at a list of results) which numbers were used to produce them.

Weaknesses that I can see include the following

The values produced would be constrained within parameters allowed by the following:
The number of possible co-ordinates available (or measurable) by the equipment used.
The minimum time period allowed by the time difference detection device.