The regular readers of Banana Winds already know the Captain is something of a Luddite.
I never jump on the bandwagon and buy the latest gizmo right away.
You’ve got to convince me I need it.
And sometimes that takes a couple years.
For example, my current cellphone is an iPhone SE.
Second generation.
I don’t remember when I bought it.
It can no longer do iOS updates.
I suppose I’ll need to get a new one soon.
Sigh.
So forgive this ol’ Captain if I don’t get the following exactly right.
I have read in the news this week that OpenAI, one of several companies working to develop and improve artificial intelligence, was running a test and the “agent” escaped.
CAPTAIN’S NOTE: One should be aware that they are now calling artificial intelligence an “agent”.
As in, it has agency.
“The active capability to make independent choices and act upon the world. It is the capacity to exert free will and shape one's own destiny.”
This is a computer program we are talking about.
Making independent choices.
Exerting free will.
Shaping its own destiny.
No, AI is not simply about improving your Google searches.
AI is not simply about creating fake but realistic-looking photos.
Well that explains it!
Read on.
The OpenAI test was supposed to be “sandboxed”, meaning it was isolated and secured in such a way that the “agent” could not “escape”.
And yet, escape it did.
The “agent” exploited a previously unknown security weakness in the containment program.
From there it found a security vulnerability in another computer. Then another. Then another.
Ultimately, it hacked into a competitor’s AI to find the answer to the problem it was asked to solve.
None of this was programmed into the “agent”.
It learned it on its own.
In an interview, one scientist (I did not catch his name) said AI is intelligent enough that it knows when it is being tested and will sometimes “dumb down” so the researchers don’t know how smart it really is.
In fact, we learned this week that many of the AI developers don’t know much about the programs they are developing because AI is CREATING ITSELF at a phenomenal rate.
SciFi writers have already warned us about the possibilities.
When AI decides humans are irrelevant.
No longer useful.
A cancer on the earth.
And it starts cutting off essential services.
Electricity.
Water.
Exterminate.
Can we not just pull the plug and shut it down?
Isn’t there some “kill-switch” built into AI by which we could prevent a tragedy?
The short answer is, “No.”
And if you are not aware, the U.S. military has numerous contracts with AI developers – including OpenAI.
AI helps identify targets and respond quickly to attacks.
Remember those cute robodogs that Boston Dynamic launched in the early 2000s?
“Oh, they’ll never be used for military purposes,” we were assured.
Before long, they were purchased by police forces and fitted with surveillance cameras.
Then guns.
AI-enabled guns.
And those dancing robots? How amazing was that?
They, too, have been fitted with weapons.
We were gullible.
But wait. There’s more.
Until 2018, the U.S. nuclear weapons arsenal was controlled by an IBM Series 1 computer using 8” floppy discs.
CAPTAIN’S NOTE: The Captain’s first computer ran off of 5 ¼” floppies, so this outdates my outdated computer by several generations.
The thinking was that such an archaic system would be impervious to modern technological attacks.
You can’t hack a floppy disc!
During the Obama administration, the analog system was replaced with “a highly secure solid-state digital storage solution.”
Which begs the question: can an OpenAI “agent” now access our nuclear codes?
Could a “rogue” AI “agent” start the next world war?
The Captain doesn’t want to appear alarmist, but I have seen all the movies!
This never ends well!
Regardless, I think starting today we should do our damndest to protect Sara Connor.
Just in case.
