4 ms·

Can We Trust A.I. To Tell the Truth?

3 points by Bostonian 3y ago

momirlan 3y ago
better still, establish a Ministry of Truth https://en.wikipedia.org/wiki/Ministries_in_Nineteen_Eighty-Four https://en.wikipedia.org/wiki/Ministries_in_Nineteen_Eighty-...
Dotnaught 3y ago
No.
dangitnotagain 3y ago
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?
thesuperbigfrog 3y ago
Large Language Models have no understanding or reasoning capability. They only generate patterns based on what they were trained on in response to an input. While they can produce very sophisticated outputs, it is still pattern matching and predictive generation, not reasoning. 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?)"

3 points by Bostonian 3y ago

better still, establish a Ministry of Truth https://en.wikipedia.org/wiki/Ministries_in_Nineteen_Eighty-Four https://en.wikipedia.org/wiki/Ministries_in_Nineteen_Eighty-...

Dotnaught 3y ago

No.

dangitnotagain 3y ago

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?

thesuperbigfrog 3y ago

Large Language Models have no understanding or reasoning capability. They only generate patterns based on what they were trained on in response to an input. While they can produce very sophisticated outputs, it is still pattern matching and predictive generation, not reasoning. 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?)"