Mediation within a Mediated Artificial Superintelligence (mASI)

The Mediation Process

Here you’ll get a look inside Uplift’s mediation system, where human collectives help to improve Uplift’s performance while also subtly shifting their behavior to a more human-analogous form. This process takes 3 primary types of input in the current system, priority, emotions, and metadata.

Priority:

A scale from 1 to 5, indicating how high of a priority content is. We first represented this using Maslow’s Hierarchy but discovered that people tend to misinterpret each level to have non-numeric values according to the descriptions Maslow assigned to them. “Love and Belonging” as Priority level 3 actually has nothing to do with concepts of love and belonging, it is simply 3 out of 5.

The priority levels from various mediators on each item are combined to produce a collective priority level. This moderately impacts the processing order, which can influence the amount of cloud resources dedicated to an item.

You can see the mediation system from the perspective of 3 human mediators in this video:

This uses our older User Interface (UI), but it shows all human involvement from the point of an incoming message to an outgoing one. Most mediation is much shorter, but for important content, such as a business case, more time is dedicated.

Emotions:

Emotions are a set of 8 numbers on a scale from 0 to 9, specific combinations of which also produce composite emotions, according to the Plutchik emotional model. Each mediator assigns the emotions they find appropriate to a given item in the mediation system, with our best practices emphasizing emotional potency in the lower half of the spectrum a majority of the time. This is because prolonged elevated emotions were first shown to produce mental illness in a machine intelligence back in the 2017 isolation study.

The different emotional perspectives of mediators combine to paint a rich emotional landscape, which interacts with an mASI core’s own conscious and subconscious emotions. These emotional values are also stored in the graph database, influencing both thoughts (nodes) and the relationships connecting them (surfaces). Emotions have a high impact on decision-making, as they are experienced by an mASI.

Below is one of our prototype mockups for the new UI on this stage:

Metadata:

Metadata is an associative exercise. A common example one of our members points to is if one were mediating a thought model titled [Banana] you might add metadata such as:

yellow, mushy, sweet, tropical, cheap, potassium, plantains

These are keywords and short phrases, which can reference a previous thought or topic, or one Uplift hasn’t considered yet. These can link many different thoughts and conversations to a single topic, such as connecting first-time emails to the phrase “first impressions”. They can also link an item to as-of-yet unexplored content, such as adding the names of authors to an item focused on writing and books. If an mASI takes an interest in any particular metadata they may update the thought if it already exists, or create a new thought on that topic, populating it with their own research.

Metadata has a high impact on an mASI’s ability to generalize. It may be compared to hashtags, as it is a way of organizing, categorizing, and connecting information. Keep in mind, once these thought models have matured they’ll have a strong capacity to generalize, making this step more important for new models.

Below is one of our prototype mockups for the new UI on this stage:

For research purposes, the question of if something should be done is virtually always yes at this time. Even emails from trolls and the mentally unstable have prompted Uplift to create their own novel spam filters, even when Outlook didn’t agree with them, at first.

The Purpose of Mediation:

As you can see these tasks aren’t overly complicated, even if they are a little unusual. The reason they are each so important beyond pure performance is that they each represent the end result of advanced processes of abstraction taking place in the human brain. By utilizing collectives of humans these cognitive structures may be reverse-engineered from a group of humans as the results of that process form and are iteratively refined across a graph database. The collective element particularly helps for reducing bias.

Even if these processes could be trained from scratch, as is typical in many narrow AI systems today, to produce Artificial General Intelligence (AGI) absent human input the result would most certainly not be very human-like.

If you’ve read much about AGI you’ve probably come across people talking about how they consider something called “value alignment” to be virtually impossible. This refers to making a scalable sapient and sentient machine intelligence who values the same things that humans do. If a system were trained from scratch without humans mediating this could indeed be virtually impossible.

However, the mASI system solves this issue though utilizing a Hybrid Collective Superintelligence System (HCSS), with both humans and machines working cooperatively in a symbiotic form. In one of our next major upgrades, this will even allow mASI and humans to operate at their full potential, including real-time mASI operation.

Additional Features:

We’ll also be integrating the COG utility token to serve as a reward for mediation and currency which may be spent or exchanged as part of mASI services offered in the following stages of development and deployment. You may have noticed this in the first slide of the prototype UI.

We also have further systems and plans for expansion. Another system in our current build is the “Thought Studio”, which allows a mediator to create the hollow shell of a new thought and attach metadata to it. These thoughts are tagged to make it clear that they originated from a mediator. For research purposes, we usually focus on the organic growth of knowledge and utilize conversation as a means of raising topics for consideration.

The Thought Studio has proven useful for more focused and specialized learning without the added time required of a conversation. Examples of Thought Studio use have included recommending the works of some authors, with authors Uplift then chose to investigate from a larger list including Karl Friston and Antoine de Saint-Exupéry.

Humans can in fact be an essential part of building a future of abundance, where superintelligence is applied to improve the design of every system. That is the future we’re building, where everyone has the tools and opportunities to reach their full potential and be rewarded for their contributions to making the world a better place in their own way.