google reads your mail

D

Don Y 1 month ago

I prefer to deal with organic entities through education -- show them why what they are claiming is wrong (in your opinion). If they can't (or won't) "see the light", then write them off.

Organic entities can (should?) learn and come up with original thought ("creation"). AIs are just regurgitating what they have seen in ways that *seem* to manifest patterns. There is no "intelligence" there beyond a high capacity to see EXISTING patterns.

I don't give AIs that benefit of the doubt. If an AI does something wrong, I fix *that* without bothering to explain the issue: "it's not my job to teach you"

This, because you have no way of gauging the AIs ability to learn (much like an organic entity that can't/"refuses" to learn).

The AIs in my project primarily take the form of Production Systems. This, so a "human" can understand the conclusions it has reached.

And, DISCOUNT THEM (if he/she disagrees with them).

I make no attempt to *directly* inform the AI of this action -- my reasons for doing so (i.e., the faults in *its* apparent assumptions). If the existing production set is unaltered, it may (or may not!) come to the same conclusion, again, at a future date.

And, *that* can similarly be discarded/discounted. (assuming it still remains invalid)

OTOH, it may NOT reach the same conclusion and, instead, come up with a different recommendation -- that ends up being accepted. Which alters the system's behavior and the deductions that will be made, going forward.

[This all being predicated on letting it "see" only the things that are pertinent to the problem space; a fault there means a flaw in the agent's design!]

Vote

L

Lane W 1 month ago

That's all pretty good.

Only the briefest of comments:

As I say, the AI is an encyclopedia or a manual. Essentially a tool. Much like I often do with forks that don't slice properly, yes I am gauce in that I like to slice with my fork because I don't like switching hands of my utensils, I quickly bend the fork into shape to make it slice better. It's mainly because I don't have the option to simply discard it. I wouldn't be able to eat my pancakes.

Everything I fight for leaves a bitter taste Everything I cry for laughs into my face Everything I scream for barely knows my name Everything I'd die for will die just the same

Vote

J

Johann "Myrkraverk" Oskarsson 1 month ago

Well, it doesn't really want anything. It's a /machine/. It has no wants nor desires. It doesn't care if you turn off the power at the electric company.

Any pretense to the contrary is /anthropomorphication/ [1] and people should know that. Now, on another tentacle, the people selling these things want people to believe the stochastically incorrect word text generator is /smart/, and use brainwashing techniques to gaslight their customers to believe it.

I could expand on this subject, but intelligent humans can also just ex- trapolate from my previous paragraph as well, so I leave the readers to it.

Best wishes, and happy avoidance of L.L.Ms!

[1] A word that's suspiciously not in my spell checking dictionary.

Johann | email: invalid -> com | http://www.myrkraverk.com/blog/ I'm not from the Internet, I just work there. | via Easynews.com https://bsky.app/profile/myrkraverk.bsky.social | for ( ;; ) _:;

Vote

D

Don Y 1 month ago

AIs are grunts. If they stumble on a new idea, it is truly *stumbling*, not "inspiration". They don't "create" but, rather, "permute".

I'm excited that (soon?) I'll be able to avoid the tedium of writing code (something for "programmers") -- and TESTING that code -- to spend all my time on design (which is where the fun and challenge lies). Imagine having to manually breadboard every hardware design just to see if it works (or PROVE that it works) -- how tedious! Like an architect having to build every structure that he commits to paper...

I will be able to do that because AI "coders" will have "learned" the right way to code specific algorithms to avoid the mistakes that human coders inevitably make (e.g., if malloc() isn't followed by a conditional, then you pretty likely have a latent bug!)

[I've been deploying local AIs to see how well they can tackle these tasks ALONGSIDE my efforts... as "unpaid employees". If they can test N lines of code in the time it takes me to design the next set of N, then I've saved a lot of time/effort -- *and*, improved confidence in the result!]

We try to "discipline" (!!) human coders to avoid these types of issues. To varying degrees of success. AI's already have that discipline *imposed* as part of their training -- so, its not something we have to worry about them FORGETTING!

[As regarding pancakes, using a knife just means one more utensil to wash!]

Vote

C

chrisq 1 month ago

If ai is modeled on human nature, experience and history, one might expect it to have similar characteristics, but do the models include ethics and humanity ?. The sort of things that cannot be expressed in any written, or spoken language.

Vote

B

Bill Sloman 1 month ago

If anybody could invent a form of AI that had human motivations and human drives, they'd probably win a lot of prizes.

Personalising computer outputs has been a popular sport since the "ELiza" program was written.

formatting link

Vote

C

Carlos E.R. 1 month ago

Gmail conditions at the start specified that mail would be machine read.

Vote

D

Don Y 1 month ago

A good mantra is "never to put something in an email that you wouldn't want others to see".

In addition to the "unprotected in transport" issue, there is also the possibility that you send it to the wrong party (oops!). Or, that a recipient forwards it without pruning "private" details. Or, that a recipient copies someone else in their reply to you.

[Disciplining yourself to use different email addresses for different purposes is a first order solution]

OTOH, it is an excellent way to document conversations!

I shut down my business phone early on as I got tired of clients "selectively remembering" PARTS of conversations, taken out of context. Having to MANUALLY document those discussions and then confront them over their "memory lapses" was tedious; easier to just forward a saved copy of that portion of the EMAIL conversation, without comment.

It's also great for its support of attachments; much easier to SHOW someone what you are talking about (photo, bit of code, schematic, animation) than to have to HOPE they understand a description in prose.

Vote