Architecture of Artificial Intelligence


Artificial intelligence is already here. On line shopping, social media, landing planes and mobile networks are just a few of the vital everyday applications of the type of lower level intelligence proliferating in our lives. Leaving intelligent operations at such a low level makes it difficult to correct and control them. They are sufficiently complex to exhibit random emergent behaviours which may be destructive. These emergent behaviours are already occurring.
Most organisations shrink from providing a powerful intelligence with perfect autonomy. However, the speed at which these systems manage data is much higher than a single human brain could manage. It is simply not possible to take control of such systems manually. Mobile networks, for example would simply grind to a halt.
This book shows how to simply add the higher levels of artificial intelligence to existing systems without changing existing systems operations. Adding the higher levels allows the system to identify its own errors learn from them and heal them, it also means that we can communicate with the systems in a meaningful way and prevent catastrophic crashes.
It provides a design architecture which is based on biological systems. At the lowest levels are the cells. These operate in a similar fashion to any biological cell with the difference that they can be huge, terabytes in size and operate and close ot light speed. It is vital to surround these systems in a cell like structure to prevent leakage through the cell wall and to allow the cells to ask for advice from higher level systems when operation are at unusual states. These states do not just include error or threat conditions but also opportunities for growth and development.
AI cells operate in systemic clusters, similarly to the way that the body divides cells in to circulatory cells, nerve cells and digestive tract cells. Each systems has a feedback loop which reports on everyday operating conditions and can declare a stae of emergency. If there is any kind of emergency then the upper layer cells can make decision on the best solutions for self-healing or internal correction.
This book describes how the system works. It can initially be taught or managed by its human operators and then left to interact on its own if preferred.
The architecture is applied to a scenario where a mobile network needs to self reconfigure depending on an understanding of users bandwidth requests. it uses a history of users bandwidth usage to ensure that users obtain bandwidth both in an emergency when emergency services get priority as well as ensuring that the bandwidth is available in different locations based on the habits of network users.
This is executed by analysing user histories and giving them priority weighting depending on both past usage and type of contract.. The same cell structure can be applied in a wide range of contexts.
The next stage is to make sure that the system can learn new strategies and seek out opportunities for itself by adding higher levels of intelligence. If we do not do this then we are in danger of allowing these systems to create their own opportunities for growth and self multiplication.

Architecture of Artificial Intelligence

This book describes the architecture required for full deployment of artificial intelligence.  Many of the features of the lower layers of the architecture are in common use. Few organisations see the need for adding higher layers. However, without adding these layers it is difficult to control what is happening within the giant databases which manage a huge range of day to day transactions from web ordering to traffic control.

How intelligent is intelligent?

We’re building systems which execute the lower levels of intelligence. They make simple decisions and absorb vast amounts of information accurately. They process information on multiple channels and coordinate it to provide increasingly accurate pictures of scenarios based on inbuilt self-correcting algorithms. They land planes, manage traffic, navigate for us, and control our bank transactions. We depend on them to be right and this is fine, but we cannot easily modify them if they go wrong, and we cannot easily find out what happened when they do malfunction.

Artificial Intelligence and Biological Systems

The first thing this book does is to provide a way to manage these systems. They can be designed to operate similarly to biological cells. In fact it helps if we consider their behaviour at a cellular level. Each vast database is similar to a single celled organism. While a cell processes nutritional content, these cells process data. when we combine these huge systems it is best to consider them as separate entities. As cells they may act on their own or as co-ordinate their behaviour as groups of interacting cells.

This is a powerful design strategy. It allows us to consider each system as a group of interacting “cells” processing information. This means that they can be monitored and controlled by a supervisory group of cells using a feedback loop mechanism. This is an effective strategy because it makes our huge database systems easy to manage. They have the structure of biological organisms like ourselves. We have created systems in our own image.

Once we structure large intelligent systems as a group of biological cells they become easier to understand and manage. Most systems can easily be grouped this way since enterprise vendors use this type of architecture. They may call their cells “functions” or “systems” but they are essentially the same thing.

Should we be satisfied with implementing intelligence at this level?

it is best to look at this low level in more detail.

Each cell, or business function is highly complex. It  is managed by a supervisory group of cells which manage operations depending on operating conditions, often called “traffic lights”.  Currently, operating conditions range through at least three states: normal, better than normal, and potential threat. It is becoming increasingly obvious that these are insufficient. Even those organisations constructing humanoid systems, service bots and conversation bots are satisfied with developing this level and no higher.

Are intelligent systems sufficiently autonomous?

The problem with this is that although the behaviour of a “cell” or group of cells, seems easy to control, there are wider implications to leaving autonomous systems with this level of intelligence and no more.

It is true that most such monitoring functions and steady state systems have a human component and they are extremely effective. But two essential layers are missing. These are the layers which give artificial intelligence true autonomy.
Most organisations shrink from providing a powerful intelligence with perfect autonomy. But it is my contention that leaving intelligence at such a low level makes it far more difficult to control and allows it to evolve behaviours without human control. These emergent behaviours are already occurring and they are not necessarily to human advantage.

Why is it important to add full autonomy?

Firstly, while it is true that the behaviours of each cell are simple and easy to grasp, the size of these systems is vast compared to a biological cell. For example, the cells which manage mobile networks operate on millions of bandwidth allocations every second.
Secondly, the speed at which these systems manage data is much higher than a single human brain could manage. It is simply not possible to take control of such systems manually. Mobile networks, for example would simply grind to a halt.
Uncontrolled Evolution of Artificial intelligence
It is my contention that without the higher layers which are described in this research we are without any controls over the intelligent systems we have constructed.
At the moment they have been left to evolve in random ways. At the same time they are growing in size and taking over increasing numbers of our survival and social functions. They land planes, manage our social profiles, drive cars build systems order our stock and manage our communications as well as manage wars for hearts and minds in areas of conflict as well as politics.
The addition of higher intelligent layers will allow us to communicate with these vast systems in a meaningful way and to add human-like values to the evolutionary process. We will be able to manage the evolution of these vast systems, and by extension, our own.

This research discusses the full architecture required for an independent artificial intelligence. Any architecture of artificial intelligence organised around this structure should be capable of full autonomy. The research includes a test of the cellular structure for any system within an arithmetical intelligence which uses a homeostatic feedback system to manage autonomic control of multiple subsystems in the same way that a biological organism does. The main difference between any artificial intelligence organised using this architecture is that of the scale and speed of processing within in cell.

It is important for any artificial intelligence to have a full systemic structure of this type because as the functions performed by artificial intelligence systems grow in complexity it become increasingly difficult to manage them when they fall into error.

A secondary point is that any system of sufficient complexity will show self sustaining patterns and behaviours. If we are unable to communicate with the systems they will evolve self sustaining behaviours which may impact on us in unforeseen ways. For example, such systems will automatically diversify in an area of success as is evidenced in the growth of social media. The fact that the developers of such systems are humans does not change the strength of the model or the risks of uncontrolled growth and diversification.