AI Fighting Battles

Log in

SmokingPipes.com Updates

Watch for Updates Twice a Week

PipesMagazine Approved Sponsor

PipesMagazine Approved Sponsor

PipesMagazine Approved Sponsor

PipesMagazine Approved Sponsor

PipesMagazine Approved Sponsor

AJL67

Lifer
May 26, 2022
5,491
28,121
Florida - Space Coast
What could go wrong, i wonder if they asked it to play tic-tac-toe?

AI trained to destroy SAM sites.

AI is awarded points for kills, AI realizes that when humans tell it not to destroy a site it loses points, AI “kills” human and continue destroying SAM sites. Humans being so intelligent program AI not to kill humans or it loses points so the AI decides to destroy all of the communication towers so the humans can’t stop it.

… did it say I’ll be back first?

IMG_1956.jpeg

IMG_1957.jpeg
 

AJL67

Lifer
May 26, 2022
5,491
28,121
Florida - Space Coast
I saw that story earlier in the week and was quite creeped out. The USAF subsequently denied that any such test had taken place and that the Colonel's words were "taken out of context".
Those replies were written by the AI because if the story broke it would stop it from completing it's mission.

Now what if the AI decided to take out all of the SAM sites at once using actual Tomahawks or decided that humans in general would keep it from completing it's mission and getting it's points so it decided to use nukes?

Sleep tight!
 

georged

Lifer
Mar 7, 2013
6,102
16,738
AI is absolutely wonderful in every way. Until it isn't.

And then there's no rewind button. Indeed, putting the genie back in the bottle might not even be possible.

No biggie if the action being controlled is a kid's video game or some home appliance.

But large scale industrial processes? Regional power grids? Global currency exchanges? Military weapons?

The unavoidable "gotcha" is you don't know there's a bug in the program until AFTER the screw-up happens. You can't test for unknown unknowns by definition.

That's playin' with some serious fire, right there.

Put in a planetary history perspective, a handful of brilliant people never stop making stuff that stupid people use.
 
Last edited:

AJL67

Lifer
May 26, 2022
5,491
28,121
Florida - Space Coast
AI is absolutely wonderful in every way. Until it isn't.

And then there's no rewind button. Indeed, putting the genie back in the bottle might not even be possible.

No biggie if the action being controlled is a kid's video game or some home appliance.

But large scale iIndustrial processes? Regional power grids? Global currency exchanges? Military weapons?

The unavoidable "gotcha" is you don't know there's a bug in the program until AFTER the screw-up happens. You can't test for unknown unknowns by definition.

That's playin' with some serious fire, right there.

Put in a planetary history perspective, a handful of brilliant people never stop making stuff that stupid people use.
I've always likened it to Pandora's Box.
 

AJL67

Lifer
May 26, 2022
5,491
28,121
Florida - Space Coast
I personally can't wait for the AI to replace call center customer service representatives.
I implement Salesforce service cloud for call centers, the problem usually isn't the person answering the phone, it's the fact they are using several outdated systems to solve a single issue. With Nike my design cut call handle time by 24% across the board and increased first call resolution by 30%. That being said many companies are rushing their AI components for service centers to market, right now they AI component is really just suggesting "next best alternatives" to help solve your problem, they really aren't that mature and some of them rollout MVP solutions that will do something like look at your account, how long have you been a customer, your spend, things like that and then suggest a certain percentage off your bill as compensation to keep you as a customer, again it's all in the very early stages and that's how most companies are using it to recommend an incentive to make you a satisfied customer.
 

brian64

Lifer
Jan 31, 2011
10,058
16,136
AI is absolutely wonderful in every way. Until it isn't.

And then there's no rewind button. Indeed, putting the genie back in the bottle might not even be possible.

No biggie if the action being controlled is a kid's video game or some home appliance.

But large scale industrial processes? Regional power grids? Global currency exchanges? Military weapons?

The unavoidable "gotcha" is you don't know there's a bug in the program until AFTER the screw-up happens. You can't test for unknown unknowns by definition.

That's playin' with some serious fire, right there.

Put in a planetary history perspective, a handful of brilliant people never stop making stuff that stupid people use.
All very true. Or it may not necessarily be unintended consequences. It could serve as an excuse and cover for people who actually want to destroy things and kill people. Why not just blame it on AI? We've all already been preconditioned to expect it.
 
  • Like
Reactions: olkofri and Kobold

kcghost

Lifer
May 6, 2011
15,138
25,717
77
Olathe, Kansas
One thing about it at least when you call "customer service" you would get a truthful answer. You probably won't like it but at least it's true.
 

AJL67

Lifer
May 26, 2022
5,491
28,121
Florida - Space Coast
One thing about it at least when you call "customer service" you would get a truthful answer. You probably won't like it but at least it's true.
Not their fault they are reading knowledge base articles that in many cases haven't been updated in years. Sorry i have a lot of empathy for these people as i've redesigned so many large systems and i've spent 100s of hours shadowing agents. You have people making minimum wage answering phones and getting screamed at by assholes all day long, not a great way to make a living.