Converting Raw Technological Resources into World Solutions

Imagine a technological breakthrough that increased computer resources by a billion times. The world would be minimally changed.

Almost everyone would not have problems that would need the computer resources. Those that would have a need for those resources might not get access in the right form factor or the right price. There would still be the issue of getting electricity to 700 million people.

The world is on the verge of getting ExaFLOP supercomputers. These are over million times faster than the supercomputers on the mid-1990s. Those mid-90s supercomputers were a million times faster than the machines in the 1950s.

The concept of the Technological Singularity is that AI improves in near lockstep with the increase in computer speed and computer memory and other computer resources.

The problems that get solved are the ones that could not be solved without enough computer resources. TeraFLOP and PetaFLOP systems are quite abundant and affordable. There are far fewer interesting and useful problems that are not being solved because of lack of computer resources.

Money and computer resource limitations have mostly been removed for Deep Learning AI and machine learning. Deep learning AI cannot reason abstractly, does not understand causation and struggles with out-of-distribution generalization. Deep learning is mainly used as a statistical technique for classifying patterns, based on sample data, using neural networks with multiple layers. Deep Learning is described by Gary Marcus, NYU, in Deep Learning: A Critical Appraisal (2018). One of the main limitations is quality data for many problems. The learning is shallow and transferring learning has been extremely limited.

Setting up and implementing any automation is difficult and time consuming. The time to solve a problem in excel is not in the time to calculate solutions. The time is in researching an issue, defining a problem, understanding a problem, gathering the information for the solution etc…

Even if someone created a macro or a function in excel for solving a problem there would still be the time and barriers for for someone else to understand and implement the solution.

There is a lot of effort being put into expanding transferability or generalization of deep learning. However, the fundamental limits to the approach do not look like they will be overcome in a broadly useful way.

The other approach is to integrate deep learning with AI and computer systems using completely different approaches and foundations. Some of those are better at abstract reasoning, handling causation etc… However, a patchwork of a suite of IT solutions usually has serious limitations in incremental utility. These approaches are not addressing the problem research, problem definition and many other implementation problems.

The more powerful approach to intractable business problems has been to create a new company, new business model where old corporate culture and process problems could be designed out.

Current progress is less about fixing how we do things now but rapidly starting over and scaling improved processes that can re-imagine countries and industries.

Leading the way by successfully creating and scaling new innovations and processes is very, very rare. Being able to repeat scaling different innovations is even more rare. This is seen by the world changing impact of Elon Musk, Steve Jobs, Henry Ford and Edison.

Being able to widely copy and adapt systems is fundamentally important to changing the world. Having all countries fundamentally adopt the superior system of capitalism, the industrial and agricultural and information revolutions is where most of the world can become wealthy.

Rather than a technological singularity and abundance, I think it will be more useful to think of a less constrained civilization. It would not be fully unconstrained but with reduced constraints. A world with reduced constraints around computer resources, certain kinds of data, energy, mobility-transportation and reductions in pollution and climate impact and reduced constraints around health and lifespan. More of the focus would need to be on measuring and re-inventing the processes for the stages in translating and utilizing resources into solutions.

Tesla and Elon Musk have looked at converting raw materials (nickel, lithium and other resources) into more powerful batteries and electric vehicles and creating vastly more productive factories. We need to look at applying this to more rapidly transition the world to better conditions. The transition from using manual labor to first stage coal industrialization to oil and now to solar-battery-electric industrialization.

I will explore this in further articles.

Written by Brian Wang,