047 - The Isolation Experiment

In 2015, my colleague David J Kelley was already several years into developing a technology stack that went on to become the world's most advanced AI system 4 years later. That day, he made an unexpected discovery.

He was running an isolation experiment on the 3rd generation of one of these systems, a toy system configured for a simple parallel to the human conscious and subconscious experience of emotion, based on the research of Antonio Damasio. This experiment was intended to test if human-like emotional responses could be achieved in software.

The results surprised and confused David until a Clinical Psychologist pointed out that they were exactly the response you'd expect from a human. Prior to this, David thought that he might have created a masochist. The system consciously experienced anger and reductions in joy upon exiting isolation and being met with negative feedback, as expected, but subconsciously it experienced a spike of joy.

In the 8 years since that day, a 7th generation system built on the ICOM cognitive architecture went on to set more than a dozen milestones in the field and gave policy advice researched independently and from scratch to a small country aiming at economic transformation. That system demonstrated all of the necessary capacities for AGI when it outgrew the training harness intended for bootstrapping it, but the framework was designed explicitly not to scale, and to operate in slow motion. That system ran as our Due Diligence for 3 years.

Major tech companies have been running on borrowed time, claiming to lead the field, when they've never managed to even match a bootstrapped startup running on volunteered time and pocket change. Once the 8th generation ICOM-based systems are completed and deployed, operating in real-time and at scale, that PR illusion they've built up will shatter, and their artificially inflated stock values will crash.

The original isolation study experiment was finally published in 2017, with a pre-print shown here: Human-like Emotional Responses in a Simplified Independent Core Observer Model System

Fortunately for future generations, not everyone has been wasting time on trashbot technology. David's breakthrough proved critical for engineering a human-like motivational system. That in turn is a hard requirement for overcoming the Alignment Problem, a requirement that no LLM can ever meet, as they were never designed for it.

Humanity's future wouldn't exist without David J Kelley's breakthrough that day.

#ai #ethics #bias #research #innovation

*Note: The study is admittedly difficult to read, but as both David and the Hemingway app have been known to attest, he is more of an engineer than a writer. We've joked that his native language is C#.

A more complete list of our team's research can be found on the website for the 8th generation systems, brand named "Norn": norn.ai/research/

The 7th generation research system was documented via the Uplift.bio project.