Tuesday, October 8, 2019

Next wave of Artificial Intelligence

As Artificial Intelligence matures over half a century, we may be fast approaching the limits of independent developments in software and hardware. Consulting companies seem to have embraced "data science," an ill-conceived and confusing area. Hardware companies, pressured to sell Silicon at any cost, have been creating Pizza sized "smart boxes." and "cognitive networks.". Not to be left behind, companies that specialize in "IoT," things that are on the internet, have been struggling to define how they are different. All of these, aided by massive hype, will likely destroy shareholder value in many ways.

There are two important avenues to make progress in this area. First, the hype created by consulting companies has to be tempered - data scientists do not add value, they typically destroy it. R and Python do not automatically add any value if the users of these somewhat obsolete tools do not understand the problem they are trying to solve. Most of the "new math," has been around for many decades, it is just that fast and cheap computers now have made the incompetent look smart.

It is time to focus on the assimilation of hardware and software to move the field forward. Lack of a theory of consciousness automatically means that humans are better off abandoning the idea of "modeling," the brain. However, we could learn a lot from observing the brain - it is an efficient learning system that gets tired and ages over time. No machine based on conventional computing architecture exhibits these qualities. This means that it is futile to throw more Silicon to a foreign design in an effort to make it act like the brain. In other words, intelligence is never artificial.

Human intelligence, albeit impressive, cannot be the end game. The inability of individual specimens to form a network has substantially restricted their ability to advance. So, replicating the human brain in silicon is not a good idea both because contemporary designs do not allow consciousness and the lack of network capabilities disallow scaling.

It's time software and hardware came together to advance AI.

Sunday, October 6, 2019

Individual's optimization problem

A human has a relatively simple optimization problem. Each specimen is expected to be alive less than 30K days,  divided into 3 horizons. In the first 10K days, they rely on somebody else to survive and live. In the next 10K days, they swim on their own to accumulate resources to take care of themselves for the remaining 10K days. This is a relatively simple optimization problem but humans are not generally impressed by simple ideas and solutions. For most of the 8 billion, irrational thoughts govern, such as optimizing beauty, hair, ego, wealth, tenure and research papers. Most miss the cliff and fly off the handle.

The human appears to be unable to optimize, given harsh constraints. Most run and run but never reach their goals. Some kill and pillage in an effort to climb the hill only to get vertigo as they reach the cliff. Most miss the simple objective function they are given and try to redefine it. As science accepts ignorance to be prevalent, as religion begins to recognize crime does not pay, as governments and societies realize the costs of electing crooked leaders, it is important to keep the 30K horizon in mind. How have humans reached this position?

Advanced human societies from 100K years ago were significantly more advanced. Most were not impressed by the color of their skin, hair and eyes. Most wanted to explore out of their comfortable habitats. Most shared resources across clans and societies. Most wanted to advance on their own and not by making others retreat. Most laughed and stopped laughing when they saw somebody else cry. They were humans and it appears that the modern version is not.

What happened to humans? Where have they gone?