13.4 Moore's Law versus AI Compression Law

AI will compress the CPS doubling rate of Moore's Law

The general rule of Moore's Law where calculations per second (CPS) doubles, say every 1.5 years, is based on progressive Computer technology development that has been performed by Humans over the past several decades.

However, it can be expected that a new AI Compression Law will soon take over to replace Moore's Law, where AI will begin to shorten the CPS doubling time period from 1.5 years, down to something shorter. 

Most importantly, AI Compression Law means that AI will be able to continually reduce the time period with each new generation of Computer system that can be produced. 

That is, AI operating on fast Computers with a high CPS, will be able to develop an even faster Computer with greater CPS that can run AI, and that AI will be able to develop an even faster Computer with greater CPS that can run AI, ... and this process will iteratively repeat.

AI Compression Law has already started

World leading companies in Computer processor design and fabrication, including ASML, TSMC and Synopsys, use a process called computational lithography that creates new Computer processor photo reticles. These photo reticles are a set of extremely specialized light filters that are needed to create each of the various semiconductor and metal layers that form all of the transistors and circuits in a Computer processor.

By way of example, to originally build the H100 GPU supplied by NVIDIA, it required 89 reticles, that each took 2 weeks to create, and TSMC used 40,000 CPU based servers to do this computational lithography work, and the whole process consumed 35MW of power.

During the GTC 2023 Keynote on 22 Mar 2023 with NVIDIA CEO Jensen Huang, he announced that NVIDIA has developed a new software library called cuLitho that runs on NVIDIA's GPUs. 

Using the new cuLitho library, running on 500 DGX H100 systems, these same reticles could now be created 40x faster, with circuit features at 2nm and smaller, and reduce the power consumption down to just 5MW.

Overall, this means better performing Computer processors with increased rates of calculations per second, can now be designed and built more rapidly and efficiently than ever before. See Appendix 5 for further details.

[137] This AI Learned to Design Computer Chips! (The View of a Chip Engineer) - Anastasi In Tech - Published: 15 Feb 2023

The effect of AI Compression Law

If we take as an example, that AI Compression Law reduces the time period to double CPS to some percentage value, XX%, of the current 1.5 years, and this compression rate effect repeats with each new successive generation of Computer system that can be produced using AI, then 'The Singularity' can arrive much sooner than the end of this century. 

Here's what the CPS rates will look like through to the year 2100, shown on a chart with linear scales for both the Years (x-axis) and CPS (y-axis) but with different Compression Rates applied to the time periods to achieve a CPS doubling with each successive generation of Computer system.  

This modeling of the AI Compression Law will start with Moore's Law as the reference benchmark where there is No Compression ( ie. Compression Rate of: 100%), and then we'll show a series of charts with different Compression Rates of: 99%, 95%, 90%, 85%, 80%, 75%, 70%, 65%, 60%, 55%, and then 50%.

Here is how AI Compression Law works

Let's take as an example, an AI Compression Rate, of say 50%, which is shown in the table and the last chart below.

If we start with the historical Moore's Law value of 1.5 years to double CPS:

• and use Frontier's estimated CPS rate of 2.0 ExaFlops (ie. 2.0 x 10^18 CPS) as the starting point at Jun 2023, so the AI Compression Law Year = 2023.500,

• then the next generation of Computer systems with AI takes 1.5 years x 50% = 0.75 years to double CPS to 4.0 ExaFlops which is 9 months later, so the AI Compression Law Year = 2024.250

• then after that, the next generation of Computer systems with AI take 0.75 x 50% = 0.375 years to double CPS to 8.0 ExaFlops which is 4.5 months later, so the AI Compression Law Year = 2024.625,  

• and this just keeps repeating. 

The time period to double CPS becomes shorter and shorter with every new generation of Computer system that is using AI to build the next generation of Computer system.

Keep in mind, the CPS rate for the entire Human race is estimated to be: 160,000,000,000,000,000,000,000,000 = 160 x 10^24 = 1.6 x 10^26

Therefore: if Moore's Law continues as normal, then Computers will reach this CPS rate in around the year 2063.

However: if AI Compression Law takes over, then Computers will reach this CPS rate MUCH SOONER than the year 2063.

If AI is allowed to operate on any Computer system that has a CPS rate anywhere on the near-vertical rising spike of 'The Singularity', the probability of that AI not having vastly superior Intelligence to Humans, is close to zero. That AI will definitely become an AGI with Super Intelligence that dwarfs the Intelligence of everything else on Earth including the entire collective Human species.

1. Moore's Law based on Human technology continuing as usual

This shows CPS reaching The Singularity around the year 2100. 

This won't happen, but it's useful to understand Moore's Law as the reference benchmark.

2. AI Compression Law = 99%

This shows that CPS is doubling with a compression percentage rate of 99% of the previous generation of Computer system timespan, and could mean reaching The Singularity around the year 2083.

3. AI Compression Law = 95%

This shows that CPS is doubling with a compression percentage rate of 95% of the previous generation of Computer system timespan, and could mean reaching The Singularity around the year 2050.

4. AI Compression Law = 90%

This shows that CPS is doubling with a compression percentage rate of 90% of the previous generation of Computer system timespan, and could mean reaching The Singularity around the year 2036, quarter 2.

5. AI Compression Law = 85%

This shows that CPS is doubling with a compression percentage rate of 85% of the previous generation of Computer system timespan, and could mean reaching The Singularity around the year 2032.

This chart is extraordinary. It means that AGI level CPS rates will be achieved in less than 10 years if AI can improve the timeframe needed by just 15% to produce with each successive generation of faster Computer. 

It seems likely that AI can do this.

6. AI Compression Law = 80%

This shows that CPS is doubling with a compression percentage rate of 80% of the previous generation of Computer system timespan, and could mean reaching The Singularity around the year 2029, quarter 2.

7. AI Compression Law = 75%

This shows that CPS is doubling with a compression percentage rate of 75% of the previous generation of Computer system timespan, and could mean reaching The Singularity around the year 2028.

8. AI Compression Law = 70%

This shows that CPS is doubling with a compression percentage rate of 70% of the previous generation of Computer system timespan, and could mean reaching The Singularity around the year 2027.

9. AI Compression Law = 65%

This shows that CPS is doubling with a compression percentage rate of 65% of the previous generation of Computer system timespan, and could mean reaching The Singularity around the year 2026, quarter 1.

10. AI Compression Law = 60%

This shows that CPS is doubling with a compression percentage rate of 60% of the previous generation of Computer system timespan, and could mean reaching The Singularity around the year 2025, quarter 4.

11. AI Compression Law = 55%

This shows that CPS is doubling with a compression percentage rate of 55% of the previous generation of Computer system timespan, and could mean reaching The Singularity around the year 2025,  quarter 2

12. AI Compression Law = 50%

This shows that CPS is doubling with a compression percentage rate of 50% of the previous generation of Computer system timespan, and could mean reaching The Singularity around the year 2025.

This chart is terrifying. It shows that if Humans put every resource possible to use AI to build better Computers, then AGI might be here in about 1.5 years.

Frankly, deliberately working to achieve this objective would be insanely stupid as it could easily destroy the entire Human race and all life on Earth extremely quickly.

Arrival of The Singularity

Various optional time frames for when The Singularity will arrive to empower computational abilities of AI technology

The following chart and table shows modelling of AI Compression Law, with Moore's Law as the reference benchmark where there is No Compression ( ie. AI Compression Rate of: 100%), and then a series of different AI Compression Rates are applied to Moore’s Law going down to 50%. 

This is all the information from the charts and comments above summarized into one chart and one table.

To survive and prosper, Humans require mandatory strong global regulation of AI development right now !

What these charts show is that if AI development is not controlled and is permitted to be developed across the world without any overarching strong regulatory management, then 'The Singularity' will arrive very soon. 

If AI applications are not controlled and AI operates on Computers with these extreme CPS rates, then AGI will arrive, and the consequences of AGI are totally unpredictable and unknown. HUMANS ARE NOT READY FOR THIS AND ARE UNLIKELY TO SURVIVE !

If and only if global regulatory control of AI is put in place immediately and strictly maintained, then the POSITIVE OPPORTUNITIES FOR HUMANS ARE TRULY TOTALLY ASTOUNDING !