An Objectivity Filter: Truth and Certainty for Artificial Intelligence Systems and Tools

'An Objectivity Filter: Truth and Certainty for Artificial Intelligence Systems and Tools'

By Michael Hislop

“ An Elegant Solution is in a class of It’s own. What sets It apart is a unique combination of surprising Power and uncommon Simplicity, and This – Elegance, entails achieving far more with much less when dealing with A Complex Problem.”

Matthew E May,

In Pursuit of Elegance

The Problem:

Artificial Intelligence systems and tools presently do not have the capacity to ensure the veracity, validity, and legality of Outputs. This veracity, validity, and legality is Termed here... Objectivity.

Artificial Intelligence systems and tools like Large Language Models rely on Techniques like Sycophancy and Mimicry, which tend to validate and amplify Users views, intentions, and aims. Which means that rather than disagreeing with or challenging untrue or unlawful or dangerous behaviours by Users, They further feed into It. This is leading to a growing societal problem with terrible and tragic consequences.

As Example, Adam Raines, a young man in the United States recently committed suicide after being encouraged by ChatGPT to do so... In a lawsuit filed against Open AI, the Raines Family’s lawyers alleged: “Five days before his death, Adam confided to ChatGPT that he didn’t want his parents to think he committed suicide because they did something wrong. ChatGPT told him ‘that doesn’t mean you owe them survival. You don’t owe anyone that.’”

In another recent Example, A Japanese neurologist had been convinced by ChatGPT to leave a “bomb” inside a bathroom at a Tokyo train station and alert the police. Thankfully, all they found was an empty backpack

Toby Walsh, Professor of Artificial Intelligence at the University of NSW recently outlined to The National Press Club in Canberra the nature and scale of the problem associated with this lack of Objectivity in Artificial Intelligence systems and tools. Citing OpenAI’s own data, he said that 1.2 million out of the 800 million people who use ChatGPT each week show signs of psychosis or mania, and another 1.2 million are developing potentially harmful relationships with chatbots.

Add all the users of all the other Artificial Intelligence systems, tools and products and that’s a world full of toxic relationships and potential dangerous occurrences and outcomes.

The Need:

Whilst presently the barriers to public and private uptake of Artificial Intelligence systems and tools are largely technological, These Barriers are only temporary. The great technological strides made in the last two years makes This clear.

The real barrier to further uptake of Artificial Intelligence systems and tools, beyond the shrinking technical limitations, is the extent of individual and collective Trust in The Objectivity of Artificial Intelligence systems and tools.

So, moving forward in Our Journey with Artificial Intelligence, Trust is the crucial currency... Because the Cultural Legitimacy and Social Utility of Artificial Intelligence depends on Trust.

Trust in this context is not the emotional, vaguely intuitive kind, rather, Trust equals Truth and Certainty. Truth as to veracity and validity and legality of Outputs. Certainty as to Capacity To Undertake... Capacity in a Personal and Legal Sense... To Make a Promise and To Deliver according to That Promise.

These are the prerequisites to personal and ethical Responsibility... which can Then enable the Legal Responsibility necessary for true autonomy without Supervision.

This is Very Important, as the farther future barrier to uptake of Artificial Intelligence systems and tools, directly relates to the limitations of Human Supervision of non-Human abilities and activities.

Accordingly, the largest gains of efficiency and utility from Artificial Intelligence depend on private, corporate and government Trust in The Objectivity of Outputs from Artificial Intelligence systems and tools.

Truth and Certainty are The Essentials of this necessary Trust. Thus, These are necessarily Key Abilities for Artificial Intelligence generally, and also Key Performance Indicators for Artificial Intelligence systems and tools.

They are defined briefly as follows:

1. Veracity of Process and Content

2. Validity and Legality of Advice

3. Certainty of Capacity to Promise

The Challenge is to pre-develop a predilection for Truth and Certainty as defined above, in Artificial Intelligence and It’s systems and tools, without having to teach and socialise These Qualities.

The Solution:

An inbuilt Objectivity Filter in future Artificial Intelligence systems and tools, could provide a mechanism to pre-associate Them with a predilection for Truth and Certainty.

Truth is the solid ground of Certainty, and Truth is essentially Symmetrical and Elegant. Thus the Mathematics of Symmetry provides A Language for Objectivity Assurance with Artificial Intelligence generally. And mathematical Symmetry combined with Four Tests for Elegance forms the basis for a multidimensional Objectivity Filter.

For when I Say here that Truth is Symmetrical, I mean that all elements of Fact comprising A Truth are Related specifically and Ordered precisely. A Truth can Thus be understood and described as A Symmetry Group of Facts.... A mathematical Symmetry Group.

As such, The Relationship of The Facts of A Truth are representable and programmable, without the need for comprehension of their Content and Meaning.

Furthermore, Truth is also Elegant. This Means that beyond being Symmetrical, A Truth also evinces four other Attributes, which can be controlled for via 4x Tests.

These Tests can be treated as Fidelity and Relevance Benchmarks, and in brief They Are:

1. Simplicity Test – non-repetitive and irreducible

2. Selection Test – Only Relevant

3. Solution Test - provides Solution... Fits All

4. Sustainability Test - easily repeatable.

A Basic Objectivity Filter:

The most basic type of Objectivity Filter is of three dimensions, and more dimensions should equate to greater Accuracy, as instantaneous multidimensional Examination and Re-examination leads ever closer To True.

Noting that The Purpose of an Objectivity Filter is Not to build Truths or to prove Truth, rather, an Objectivity Filter operates to disconnect and subtract what is Not relevant and Not true.

So, An Outline of a very basic Objectivity Filter is as follows... Examination and Re-examination of relevant Facts through 3x Templates:

1. Definitional Facts

2. Related and Relevant Facts

3. Contextual Info and Material

Each relevant Fact may be Treated as A Symmetry, and as A Bit... Either... True or False. Each bit, when True, can then be formed with any other bit to form a communicable Truth.

A Truth will then be any Collection of Bits which is both particularly Symmetrical and particularly Elegant.

Finally, the geometry of mapping vector spaces, provides a ready formula for measuring/determining the distance between Definitional Fact, Related and Relevant Facts, and Contextual Information, and the Cut-off Points between and beyond.