Why AI is hard

Artificial intelligence was one of the first buzzwords I can remember from the previous century. It promised a future with intelligent computers or devices which could understand you and act autonomously.

Up until now we still do not use AI-enabled devices in our daily life. Why is that?

Our brain has 10 billion brain cells. Each brain cell has 10.000 connections to other brain cells. Neurons activate pathways to brain cells. The activation of brain cells along those pathways pass along the message of the neuron to other brain cells. Some pathways end and the message is not forwarded. The combination of brain cells and pathways between brain cells is how the brain processes and stores information.

To make AI happen we should be at least be able to mimic the brain cells and pathways functions. There are two options available; hardware and software emulation.

You could compare the brain (without memory) with a CPU. Current top-of-the-line computer processors contain 2–3 billion transistors. That comes close but each transistor has only a few pathways and is in the end only a transistor. If you would emulate the pathways the current CPU need to contain at least 50.000 more transistors than they today. Based on Moore’s law (CPU transistor count doubles every 2 years) we can have to wait until 2025 to get to this point.

From a memory perspective, you need to store the values of each of the pathways of the brain cells. This would require 10B * 10k * 1 byte = ~90TB of data. The biggest individual memory units currently available are 12GB. Applying the law of Moore again and we reach 90TB memory units in 2023.

With current technology, we are able to emulate the brain function of an insect. Hence you see all those insect-like autonomous robots.

Based on these calculations it looks major progress can be made in AI in 15 years from now because the necessary computer hardware is available to be able to emulate the brain functions. Of course, emulation is only the first step. The emulated brain also needs to be programmed to work. Much like any brain is programmed before birth.