Can We Trust A.I. To Tell the Truth?(reason.com)
reason.com
Can We Trust A.I. To Tell the Truth?
https://reason.com/2023/08/10/can-we-trust-a-i-to-tell-the-truth/
4 comments
We cannot trust humans to tell the Truth.
Better yet, can we ourselves deduce Truth?
So separately is LLM capable of discerning and manipulating Truth? No, not alone.
Humans are not innately truthful. However we may learn and embody principles, of which may be those of truthfulness.
Our technology must undergo this same process of self deception as we ourselves have not (collectively) figured out.
The Truth is the perturbation of Objective Reality (state’s collapse in the moment of now.) the “truth” is a figment of minds. Integrity is the measure in between.
As certainty is truly impossible, determination must come by the least uncertain resolve.
Consider that there are plausibly four representations of truthieness.
- pro-presumptive-truth
- a-presumptive-truth
- actualized-truth
- falsified-truth
The two former might be considered non-terminal yet representative (actuating bias explicitly), and the latter as terminal (mappings to discrete state resolutions.)
Now, ask yourself, can you tell the difference when you hear bullshit or important shit from another human?
Better yet, can we ourselves deduce Truth?
So separately is LLM capable of discerning and manipulating Truth? No, not alone.
Humans are not innately truthful. However we may learn and embody principles, of which may be those of truthfulness.
Our technology must undergo this same process of self deception as we ourselves have not (collectively) figured out.
The Truth is the perturbation of Objective Reality (state’s collapse in the moment of now.) the “truth” is a figment of minds. Integrity is the measure in between.
As certainty is truly impossible, determination must come by the least uncertain resolve.
Consider that there are plausibly four representations of truthieness.
- pro-presumptive-truth
- a-presumptive-truth
- actualized-truth
- falsified-truth
The two former might be considered non-terminal yet representative (actuating bias explicitly), and the latter as terminal (mappings to discrete state resolutions.)
Now, ask yourself, can you tell the difference when you hear bullshit or important shit from another human?
better still, establish a Ministry of Truth
https://en.wikipedia.org/wiki/Ministries_in_Nineteen_Eighty-...
No.
Other AI systems based on other methods (provers and related-logic systems) can derive facts or "reason" based on their inputs and rules. Their "truth-telling" capacity is only as good as their inputs and derivation rules.
Better questions would be:
"What are the capabilities and limitations of this AI system?"
"Who created this AI and for what purpose?"
"How was this AI created and how does it work? (what inputs was it given / trained on and how does it use those inputs to generate outputs?)"