As someone who has worked on and experimented with artificial intelligence for over 40 years, I should be very excited to see a field I’ve dedicated much of my life to becoming part of everyday life, but I am not.
In fact, I feel the opposite of excitement and I actively oppose the development and deployment of most of what is currently referred to as “AI”.
Why is this? There are numerous reasons, far too many to meaningfully cover in a single blog post. Right now I can’t recommend and I won’t voluntarily contribute to any of the popular, contemporary AI systems.
But could anything change my mind? I tried to imagine the conditions under which I could work with and endorse technologies which utilize these techniques. Here’s what I’ve come up with so far:
The process, logic, training and execution of models, decisions and data must be completely transparent to anyone subjected to their output. The source code for any software and design files for any hardware must be available for analysis and any dependencies or intermediaries of the systems processing must be published without restrictions. No black boxes, no trade secrets, no redactions.
This documentation must be so complete that anyone can recreate the systems output in a reliable, repeatable fashion given the same training and input data.
Consent
Anyone impacted by the use of the system must opt-in or otherwise consent to its effects before the system is used on them.
Any input used to train or operate the system must receive active consent for its use by the creator of the input (simply acting within the limits of a pre-existing license is insufficient: the use of the input as part of an artificial intelligence system must be granted explicitly before the input is used).
Quality
These systems must be used only when they are capable of producing a higher-quality result than natural intelligence. This includes improving the quality of life for humans or other natural intelligences.
Using artificial intelligence to simply make an existing process cheaper is not acceptable.
Enlightenment
The use of these systems must result in elevating the knowledge or happiness of human life. Any application that results in a reduction of natural intelligence (deskilling, loss of livelihood, reduced cognitive function, reduced creativity, reduced happiness, etc.) is considered unethical by these standards.
A Start
This list may not encompass all unethical applications for AI, and nefarious uses may find ways to maneuver around these protections. But at the moment I’m not away of any example of an “AI enabled” application or platform that can meet these standards let alone something more selective.
So until then this will serve as a starting point and a reference for when people claim that I am a Luddite or an “AI doomer”; it’s quite the contrary, I just believe it must be done the right way.