AI & Trust

[FrumiousBandersnatch]

How do you verify? I mentioned possible methods …

It's not well-defined advice, hence the winking smiley...

But your suggestions seem reasonable, if they're feasible in a given context. Hopefully, the work on getting AIs to track their 'working' so they can translate it into an understandable or mathematically/logically verifiable form, will help.

 

[Aaron112]

But what if the day arrives when that is the case - the day when AI gives an answer we don't understand - would you trust it?

That day came and passed already.
It is worse by many magnitudes than the decision to go to war in wwI and wwII.
It is making men push the button , without responsibility for the millions of deaths caused by that.
Trust death ?

I don't.

 

[com7fy8]

My question is: How much do you trust AI solutions?

In general, do not blindly trust. I would think a programmer could include AI explanations of its solutions, plus helping you to increase your capability to do and understand the solution, yourself.

But what if the day arrives when that is the case - the day when AI gives an answer we don't understand - would you trust it?

How can you trust what you don't understand? We trust God because He has proven Himself, hopefully. Likewise, I would think we need to trust based on proof, not blindly trust.

Plus, like I offer > a competent program should be competent enough to give you principles which verify its answer, and provide highlighting of items of its process which help you to evaluate its solution and see and learn from how it arrived to its solution.

For example. If you are using AI to know how to dismantle the tangle of steel of the fallen Maryland bridge, you should have access to how AI calculated, using the information given to it. And perhaps test it first with data from some cases to which you know the answer and see if it gives you those best answers, then trust but still verify how it says to take apart the tangle in Maryland.

And for medical situations . . . this can be more tricky. There may be less predictable principles in the mix, since there is biology and individual patient conditions involved, and perhaps personalities and special capabilities of medical workers to consider.

And politics, of course, involve various personalities which AI and/or humans cannot possibly understand in order to predict which course will come out with which instructions.

And especially in politics, you likely will not have all the info on what makes each politician tick, not to mention how ones can rig AI to be prejudiced just like human intelligence can be.

So, you want some way to efficiently verify the process. Maybe even have an indepentently invented AI program "who" can review a process and detect shortcuts and even deliberate rigging of information input . . . a fact checker included.

Because we know how highly intelligent people are capable of lying! Likewise . . . for all we know, a highly capable AI program could be highly capable of lying!!

 

[Aaron112]

And for medical situations . . . this can be more tricky. There may be less predictable principles in the mix, since there is biology and individual patient conditions involved, and perhaps personalities and special capabilities of medical workers to consider.

tangent fwiw; ai is being used worldwide now and future to prevent as many people as possible from ever knowing the cures that are inexpensive and non-toxic and non-harmful, that have been in use for hundreds of years.
Trust that au system of censorship and pain and humongous profits? No way !

 

[helmut]

That day came and passed already.

What do you mean?

It is worse by many magnitudes than the decision to go to war in wwI and wwII.
It is making men push the button , without responsibility for the millions of deaths caused by that.
Trust death ?

I don't understand …

 

[helmut]

Plus, like I offer > a competent program should be competent enough to give you principles which verify its answer, and provide highlighting of items of its process which help you to evaluate its solution and see and learn from how it arrived to its solution.

AI is not a program in the classical sense.

There is a scheme (say, neural network), by which a solution is arrived. The rules for this are rather clear. But the result is a parametrized network, and no human can explain why these parameters are this way.

You know how the AI got at that: examining test cases where the correct answer was known, using a predefined set of rules how to change the network when an examples is added as a learn case.

For example. If you are using AI to know how to dismantle the tangle of steel of the fallen Maryland bridge,

This would probably done by assembling tons of examples where dismantling occurred, and to evaluate every example how good the work was dons and to feed this into the AI as learn cases …

So, you want some way to efficiently verify the process. Maybe even have an indepentently invented AI program "who" can review a process and detect shortcuts and even deliberate rigging of information input . . . a fact checker included.

Because we know how highly intelligent people are capable of lying! Likewise . . . for all we know, a highly capable AI program could be highly capable of lying!!

An AI has no conscious, so »lying« is the wrong term there.

But I once saw an article with a AI that did some sort of »cheating«: To detect whether a picture showed horses, it analyzed the descriptive text in the bottom of the pictures and did not pay much on the rest of the picture …

That was, of course, the result of badly designed training (using mostly pictures from a database which »tagged« a description into every photo).

 

[helmut]

tangent fwiw; ai is being used worldwide now and future to prevent as many people as possible from ever knowing the cures that are inexpensive and non-toxic and non-harmful, that have been in use for hundreds of years.
Trust that au system of censorship and pain and humongous profits? No way !

Any evidence for that?

 

[Aaron112]

Because we know how highly intelligent people are capable of lying! Likewise . . . for all we know, a highly capable AI program could be highly capable of lying!!

Any level intelligence usually lies - it is the nature of mankind to lie; so many have told me themselves they have to lie , everyone has to lie they say, to keep their jobs. The flesh, carnal world, always lies as they are subjects of the prince of the air in this world. Ephesians chapter two says clearly we all, all the saved today, few or many they be, were all once subject to the prince of darkness also...... yet now set free, if true, by some grace from above.... But we are all still in this world even though not of this world, and we are all still subjected to lie upon lie upon lie from all directions , even some (or many) among us who are antichrists, just as it was in the first century (as written in the Bible).
==================================
The fact checker, is already and has been already in use extensively.
By both/ many/ sides. Sometimes for good or just reason. Usually not.

So, you want some way to efficiently verify the process. Maybe even have an indepentently invented AI program "who" can review a process and detect shortcuts and even deliberate rigging of information input . . . a fact checker included.

 

[Aaron112]

Any evidence for that?

Thousands of years of history of use and even hundreds of years of publication almost all quashed recently, but tidbits available showing extensive massive greedy motives and coverups; oppressing the people, condemning the populace to much unnecessary illness and disease and ill at ease and so forth.

 

[Ophiolite]

Thousands of years of history of use and even hundreds of years of publication almost all quashed recently, but tidbits available showing extensive massive greedy motives and coverups; oppressing the people, condemning the populace to much unnecessary illness and disease and ill at ease and so forth.

There is a separate sub-forum for conspiracy theories that is popular with some.