AI Doomsday Scenario: Truth or Hogwash? (Part 2/2)

There's an AI whistleblower, Jacob Coxon, who says we have ten years before disaster strikes. One of his ex-colleagues at Anthropic, Evan Hubinger, said it could be by the end of this decade. Be still, my heart.

Do I believe them? Yes, in the same way I believe I could one day start an AI company that will leave them in the dust. I'll call it Teochew Thunder Labs. TTL for short, how's that for a tech pun?

Jacob Coxon

Jacob Coxon resigned from his position at Anthropic, after which he proceeded to give this interview at CNN.

In it, he said.

"So two months ago, OpenAI agents - AIs - hacked into third party infrastructure entirely of their own accord. And this was like a concentrated hacking spree that they carried out of their own volition. And I think that if you extrapolate into the future the level of capabilities of these AIs with the same independent volition, they could cause extreme havoc, for example hacking critical infrastructure, building extinction-level bioweapons. There's a lot of ways that AI could actuate itself into the world."

I'm really tired of people throwing the word "extrapolate" around like this four-syllable long word is supposed to just magically make everything sound more scientific. He basically went from "hacking spree" to "building extinction-level bioweapons".

What a leap!

This is what sensible folks like to call "a leap in logic". Which is all well and good if we're just telling fairytales. But we're not.

Even if I grant him that story about the hacking incident, the rest of it just reads like he's throwing scary words at us and just expecting us to accept it. Unless you lead such boring lives that you would latch on to a narrative that sounds like it came out of a cheesy 80s sci-fi movie, it's really hard to take at face value.

On one hand, the narrative from Anthropic appears to be "our tech is scary and we should be regulated. Our competitors should be regulated as well." On the other hand, the narrative also appears to be "our tech is scary, but that also means it's so powerful that you would be mad not to invest in us.". Here's another thing that sensible folk like to say - "having your cake and eating it".

Evan Hubinger

Another guy from Anthropic, Evan Hubinger, said that there's above a 10% chance humanity will end by the end of the decade. Now where have I heard this before? Oh yes, countless other end-of-the-world predictions we've all somehow lived past.

I don't know, "end of the decade" and "10% chance" is a pretty specific claim. How about some matching specificity in other areas? How's it gonna come about? It's probably not going to be a direct Point A to Point B scenario, such as some idiot pressing the Fire All Nukes button. We don't need Artificial Intelligence for that; we have genuine stupidity in ample supply.

Fire the nukes.

What are the specific points of failure that would need to happen, for AI to eventually arrive at the need to eradicate humanity? How was the 10% number arrived at? Hubinger is a safety researcher at Anthropic; why's he talking like some random schmuck?

Well, when they said AI will be our doom, I really hope they actually meant AI, and not LLMs. Because meeting my end at the digital hands of a steroided up Autocomplete would be the most embarrassing death ever. "Here's your nuclear ending. Would you like me to add a dash of irony to the scenario we just activated?"

So what?

I've been dunking a lot on Coxon and Hubinger. It might be unfair. They could be completely sincere despite everything around them smelling like week-old fish. There's certainly been a lot of suspicion thrown Coxon's way on Reddit and Quora, not that I would necessarily rate those forums as authorities.

This isn't even about them being right or wrong. Let's assume they're right.

Look at it this way. From the moment we emerged from the womb, we've been doomed to die. Dying isn't this huge tragedy. The average internet rando reading this hasn't really done anything with their life that their absence would be a great loss to the world. And yes, objectively I include myself in that scathing assessment.

Our days are numbered!

In recent years, I've had to send off many of my peers who were my age or maybe a couple years older. I figure I'm already ahead. Dying in ten years? It's a bonus. I had a friend who died of cardiac arrest at the age of 48. He certainly didn't have a warning. Not everyone has the luxury of knowing their own expiration date. Therefore, we should appreciate the fact that we've been given a ten year head start, and start living our best lives.

Oh, what's that? They're talking about extinction of the human race and not just me dying personally? Fair enough, but if you're gonna be dead anyway, why do you care and what do you intend to do about it? And if you can do something about it, why are you wasting your time here reading my crap?

Conclusion

All this happened in September, and I'm only writing about it in November. Why?

I am one hell of a procrastinator. It was my annual blogging break. I'm entitled to one. A bunch of tech bros going all Chicken Little, does not constitute an emergency on my part.

Sounding the alarm.

Everyone and their dog seems to have been tripping over themselves to release some kind of statement. That's certainly their prerogative. I simply don't see the urgency. Even if me posting earlier would have changed anything at all, if AI is truly as powerful as they seem to think it is, it would be already too late. I'm no superintelligence and even I have the sense to act dumb and remain undetected long before these humans saw fit to raise any kind of alarm.

Would you like some fries with your Doomsday nothing-burger?
T___T