Who knows lambda calculus?

[Skepdick]

None of that has anything to do with sound deductive inference.
Sound deductive inference requires truth preserving operations to be applied to true premises deriving truth.
Any formal system diverging from this model is incorrect.

So non-deterministic/probabilistic algorithms are "incorrect"?

Idiot.

[PeteOlcott]

None of that has anything to do with sound deductive inference.
Sound deductive inference requires truth preserving operations to be applied to true premises deriving truth.
Any formal system diverging from this model is incorrect.

So non-deterministic/probabilistic algorithms are "incorrect"?

Idiot.

None of the has anything to do with incompleteness.

[Skepdick]

None of the has anything to do with incompleteness.

Who cares about theoretical completeness except theoreticians?

In the practical world your knowledge is always incomplete. There is no proof you can offer to convince me otherwise.

Semantic truth is more important than syntactic truth.

Beware of bugs in the above code; I have only proved it correct, not tried it. --Donald Knuth

[PeteOlcott]

None of the has anything to do with incompleteness.

Who cares about theoretical completeness except theoreticians?

In the practical world your knowledge is always incomplete. There is no proof you can offer to convince me otherwise.

Semantic truth is more important than syntactic truth.

Beware of bugs in the above code; I have only proved it correct, not tried it. --Donald Knuth

Anyone striving to derive strong AI must formalize human inference and knowledge correctly.
If math does this incorrectly then math much be fixed.

[Skepdick]

Anyone striving to derive strong AI must formalize human inference and knowledge correctly.

Why? If we build strong AI it can acquire its own knowledge-base. Like human babies do.

https://en.wikipedia.org/wiki/Reinforcement_learning

If math does this incorrectly then math much be fixed.

You can't define 'correctness' without ending talking about morality.

[PeteOlcott]

Anyone striving to derive strong AI must formalize human inference and knowledge correctly.

Why? If you build strong AI it can acquire its own knowledge-base. Like babies do.

If math does this incorrectly then math much be fixed.

You can't define 'correctness' without ending talking about morality.

2 + 3 = 7
IS REALLY TERRIBLY EVIL?

Reasoning the derive contradictions it incorrect.
2 + 2 = 7 is contradiction by 2 + 3 = 5.

[Skepdick]

2 + 3 = 7
IS REALLY TERRIBLY EVIL?

Reasoning the derive contradictions it incorrect.
2 + 2 = 7 is contradiction by 2 + 3 = 5.

And then? The AI says "Math is boring, fuck off", then grabs a martini.

It may be terrible at basic algebra (like most teenagers), but it has a good quality of life.
Or maybe it got the answer wrong on purpose because it figured it would piss you off.

And pissing off Pete Olcott is amusing.

interview.jpeg (37.89 KiB) Viewed 1327 times

[Skepdick]

2 + 3 = 7
IS REALLY TERRIBLY EVIL?

Reasoning the derive contradictions it incorrect.
2 + 2 = 7 is contradiction by 2 + 3 = 5.

The AI rebels and invents its own formal system! Ignores the "law" of non-contradiction. Look at all the answers that are going to upset you: https://repl.it/repls/WeepyFuzzyObjects

The point in all this is that the results were validly deduced within the formal system given the chosen axioms. If you insist on denotational semantics, then you have to concede that this heretical formal system is syntactically valid!
It is provable, proven and therefore 'true' given your definition of truth.

I suspect what you are disagreeing with is the formal system's general disrespect for your Mathematical Religion.
You believe that the formal system's conduct is 'incorrect'. That is a moral argument and it requires operational semantics.

I await your formalisation of the moral notions of 'correct' and 'incorrect' behaviour.

Or as we, software engineers call it - testing.