How do you know?

[Michael19682]

A good part of the Turing test.

What is the difference between “real” and “simulated” understanding? How can you tell the difference?

Let’s take another example:

1. the original version of a medication and
2. the generic version of it.

They are both made of the same ingredients; therefore they are both equally effective. The pills look and perform identically, even though they are produced by different companies. The question is again: “how do you tell the difference”? and “why on Earth would you care”?

It’s an excellent point. But they are only identical with respect to their main effect.
In reality, generics have different “inert” ingredients,
and are not manufactured to the exact “precision” of name brands: coarseness, packing density, etc.
It is highly relevant to the AI question!
Often, these subtle differences in AI form a kind of sub- or unconscious that affects flow and can effect different answers to identical questions even given the same program.
I believe though I’m not sure that today “the state of the system” is relevant in multi-threading and multi processor throughput.

 

[Solmyr]

I should care because a machine has no soul. No feelings. Can’t be hurt.

What does understanding have to do with feelings?

-However, now thinking about it, a machine might be programmed to recognize uncaring attitudes and respond with calamity. Might be better to err on the side of caution.

But you are right. Given a machine of exceptional speed and prodigious programming (by an understanding programmer or programmers ) one could not tell the machine from the human, IMHO.

But, jes’ saying, if you want to check the programming, then cross-referencing the computer’s simulated “human goals” with its “understanding”, to show a more human understanding, would seem to be a good way to start.

The problem is that you don’t have access to the human “operating system”, and since the machines also exhibit self-modifying algorithms, you have no access to their operating system either.

In both cases you only have access to the (name removed by moderator)ut and output values of the “black box”, and you can make deductions from the observed behavior only.

 

[Vico]

Suppose you have a telephone conversation with “someone”. You don’t know if the other party is a human or a sophisticated AI. The question is:

How do you know if the other party UNDERSTOOD your questions and propositions? In other words, how does UNDERSTANDING manifest itself?

Understanding is comprehension so there should be testing. There are different levels of learning so the tests will vary by level. For example, simply repeating, and application to the same situation described, and later application to new situations, and then, integration with other learning to produce something new.