Overview of AI Logic Layer
The ethical foundation – AI logic layer, provides for the following components.
- Neural Network
- Machine Learning Algorithms
- Ethics Processing Unit
- Big Data Integration Module
When a request is submitted to the AI, it will flow through the necessary machine learning algorithms and big data integration, processed by the neural network and then feed into our ethics processor. The ethics processor will then evaluate the output against the pillars (universal truths/human rights layer) and the bedrock (3 laws of robotics) A decision would be made to continue or to reject the request and provide reasons why.
Components of AI Logic Layer
Neural Network
Here lies the brains, our neural network. In order to make our architecture flexible, the ethical foundation would ship with a neural network already in place. However, to give vendors flexibility to build there own AI systems, this layer would allow for modifications. Vendors would be able to tweak the existing neural network provided, or the framework allows for them to remove it all together and add in their own neural network. To understand more on neural networks, you can read my article titled AI Technical Flow Part 2 . Additionally, here’s an article by IBM titled What Is A Neural Network.
If you’re thinking a vendor specific neural network can bypass the ethical foundation we’ve built here. Think again. This will be covered under the ethical processing unit component.
Machine Learning Algorithms and Learning Unit
The ML, or machine learning algorithms and leaning unit can also be found in our logic layer. Like the neural network itself, these components ship with the AI Ethical Foundation. To once again provide for vendor specific implementations, it can be modified or even replace. Different AI systems can have specifics roles. Medical, law, environment, political, etc. each requiring specific types of data and learning methods. There are a variety of ML types. Here’s what Geeks for Geeks had to say in their article Types of Machine Learning. The field of machine learning continues to evolve, so additional training methods are being added. We’ll introduce one of these newer approaches when we cover our next component, the Ethical Processing Unit
The Ethical Processing Unit
The Ethical Processing Unit is the key to our entire AI Ethical Foundation. When a request comes through we need to determine if it passes our ethics. To do this prior to the AI proceeding, like checking the requirements first, it leaves a gap, allowing the AI to implement as it see’s fit. However, it’s through implementation itself where ethics can be violated. To avoid this pitfall or gap, the request, through implementation is processed. Only then is it fed into our ethical processing unit. This processing unit will leverage the pillars of universal truths/human rights and then flow into the bedrock, the 3 laws of robotics. Once complete the request or project is either approved or it’s rejected. If rejected a full explanation is provided.
The ethical processing unit is itself an AI and is fixed. This simply means a vendor cannot change, remove or otherwise tamper with this layer. In addition, this layer will itself execute a newer machine learning type, adversarial learning. This process allows for continuous improvement of our ethics processor, finding and compensating for ever changing attacks against the pillars or 3 laws of robotics. IBM discusses this approach in their article titled, What Is Adversarial Machine Learning.
Big Data Integration Module
The last component in our logic layer is the big data integration module. Just like the neural network, machine learning algorithms and processing, the big data integration module is shipped with the AI Ethical Foundation, but is flexible. Given the needs of different vendor implementations, this module can be tweak, changed, or completely swapped out. Big data, like machine learning covers a large area around data. Dataversity explains it nicely in their article titled Big Data Integration 101: The What, Why and How.
Summary
The objective of the last 4 articles has been to provide a roadmap to implementing ethics in our AI foundation. Previous articles discussed governance around ethics in AI. This proposed architecture provides instant verification, real-time to proposed AI projects. To implement governance from some external source would be time consuming. Developers would spend a large amount of time following external verification procedures instead of doing what they do best, develop.
No system is perfect. But this provides a start. This provides a way to burn in, if you will, ethics. To ensure the projects undertaken benefit all peoples, to solve a need.
