# The precise relationship between energy & information that form the foundation of the post AI economy.

*Information and energy have an equivalence and their relationship in semiconductors, as defined by physics, is about both unit conversion and the difference in the effective number system used in nature and computers. The advent of AI makes this supremely important as the requirements for compute and energy intersect.*

## Information and energy equivalence.

When mass is converted into energy you get a nuclear explosion. when you convert information into energy you get heat, and all modern computers do this as they delete information (change it into heat) as the logic gates that process information run irreversibly.

Energy and information equivalence is just as profound and just as simple in mathematical expression as energy and mass equivalence (E∝m). E ∝ temperature x number of bits. Instead of c², the factor of proportionality as defined by Landauer is kln(2) and as we see below this is really just a unit conversion.

## Information converting into energy in computers

Energy and bits are equivalent and when you reduce the number of bits by making calculations irreversible as all current computers do, but don’t have to, you inevitably get heat that you can potentially eliminate if you make things reversible. In practice this would mean logic gates that have three input and output wires (instead of the current 2 in and 1 out*) so that you know the input from the output.

## Information and heat entropy

Instead of energy, we will measure entropy, which is the unusable energy per unit temperature and where two different types of entropy (heat and information entropy) that were previously thought to be merely analogous,, are essentially the same.

## Heat entropy.

Heat entropy uses the symbol S (from Sadi Carnot). Entropy itself comes from”τροπή” (trope), meaning transformation, to highlight its parallels with the transformational nature of energy flows.

The heat entropy of a physical system is proportional to the number of equally likely microstate arrangements (W, from Wahrscheinlichkeit,” the German word for “likelihood”) per macrostate. i.e. the logarithm of the arrangements, where the base is e so it is the natural logarithm, ln.

i.e. S is proportional to lnW

A macrostate is a measurable, overall characteristic of a system, like temperature or pressure and a microstate can be a particular arrangement of things like position and velocity, spin or whatever of particles. An analogy would be the number of different photographs of systems particles that would equal the same temperature overall.

This proportionality is between temperature and energy and is called k.

So: S = klnW

Planck chose it, despite being on Boltzmann’s tombstone, he never actually used the constant that is named after him. Its numerical value depends on the choice of units for energy and temperature and so if ‘natural’ Planck units are used where the macroscopic temperature is related to the energy of atomic particles, k disappears. and S = ln W.

## Information Entropy

Shannon defined information entropy, H (possibly a tribute to Ralph Hartley, who laid the groundwork for measuring information).

The information entropy of a variable X is defined by the following formula which looks much more complicated than Boltzmann’s one for heat because it allows for the variable to be not just a number but, say, letters where each letter has a different probability of occuring (like in English). It also allows for the log to be in any base, when for computers it is base 2 unlike Boltzmann’s for heat which is in base e.

The information entropy (potential information) in a truly random sequence is the number of digits (which is the logarithm of the sequence in the base of the number system used).

In computing, this base is 2, leading to entropy measured in binary ‘bits’.

So for a given number of potential states n (e.g. 2 if there is a 1 or a 0 or 4 is there is a 00, 01, 10 or 11) information entropy for computers is:

*i.e. in a random sequence where 1011101 and 1100011 are equally likely, the entropy is the number of digits (log base 2 of the number of different 7 digit base 2 numbers) — bits, a*nd in this instance, H (X) = number of 1s or 0s.

## Bringing heat and information energy back together.

So if information entropy is the log, base 2 of the number of arrangements of ones or zeros and heat entropy is k times the log base e of the number of possible arrangements of components that determine, say, the temperature, then converting from one to the other means just converting the base and multiplying by k for standard units of heat.

Going back to energy from entropy,

**1 bit converts into kTln(2) joules of heat for a given temperature.**

*except for NOT gates which have one input and one output.