@billseitz Cool, I love High Potential! Haven't seen the episode yet.
Tagged: Akkoma
"AI will take over the world."
This can mean two things:
1. "AI" will become self-aware and decide to control humans.
2. Gen-slop and deepfakes will dominate available content, leading to deskilling, human abuse and eco-harm.
The first idea has no basis, but it's fair to worry about the second.
Happy Petrov Day to those who celebrate. On September 26, 1983, Stanislav Petrov made the correct decision to not trust a computer.
The early warning system at command center Serpukhov-15, loudly alerting of a nuclear attack from the United States, was of course modern and up-to-date. Stanislav Petrov was in charge, working his second shift in place of a colleague who was ill.
Many officers facing the same situation would have called their superiors to alert them of the need for a counter-attack. Especially as fellow officers were shouting at him to retaliate quickly before it was too late. Petrov did not succumb.
I've attached a short clip from a reenactment of the situation in the documentary The Man Who Saved the World.
The computer was indeed wrong about the imminent attack and Petrov likely saved the world from nuclear disaster in those impossibly stressful minutes, by daring to wait for ground confirmation. For context one must also be aware that this was at a time when US-Soviet relations were extremely tense.
I've previously written about three lessons to take away from Petrov's actions:
1. Embrace multiple perspectives
The fact that it was not Stanislov Petrov's own choice to pursue an army career speaks to me of how important it is to welcome a broad range of experiences and perspectives. Petrov received an education as an engineer rather than a military man. He knew the unpredictability of machine behavior.
2. Look for multiple confirmation points
Stanislav Petrov understood what he was looking for. While he has admitted he could not be 100% sure the attack wasn't real, there were several factors he has mentioned that played into his decision:
- He had been told a US attack would be all-out. An attack with only 5 missiles did not make sense to him.
- Ground radar failed to pick up supporting evidence of an attack, even after minutes of waiting.
- The message passed too quickly through the 30 layers of verification he himself had devised.
On top of this: The launch detection system was new (and hence he did not fully trust it).
3. Reward exposure of faulty systems
If we keep praising our tools for their excellence and efficiency it's hard to later accept their defects. When shortcomings are found, this needs to be communicated just as clearly and widely as successes. Maintaining an illusion of perfect, neutral and flawless systems will keep people from questioning the systems when the systems need to be questioned.
We need to stop punishing when failure helps us understand something that can be improved.
My original blog post about Stanislav Petrov:
https://axbom.com/lessons-from-stanislav-petrov/
@kolombiken Yeah, the other group didn't see.
@troed And Iâd *expect* people to make mistakes. đ
I hope @emilymbender and @alex will dissect the Red Lines letter. Unless they already have and Iâve missed it.
Iâm personally a bit disappointed in wording that sounds like criti-hype, and suggesting that there is a window that is about to close and then itâs âtoo lateâ.
"it will become increasingly difficult to exert meaningful human control in the coming years"
? â human control is all that itâs aboutâŠ
"Governments must act decisively before the window for meaningful intervention closes.â
? - is there an expiry date on regulation and after that it doesnât work?
https://red-lines.ai
#artificialurgency
@iakonkret Ja precis! Jag Àr en av de som bjöd in :) Det var trevligt hÀng och kul att samsprÄka med likasinnade. Alla var rörande överens om vi vill ses igen sÄ det blir ett till möte i november. Det kommer nog ta ett par iterationer att landa i vad vi kan och inte kan Ästadkomma. Men alla verkar vilja göra nÄgot för att öka allmÀnhetens förstÄelse för vilka samhÀllsförÀndringar "AI" innebÀr.
Vi har ju inte landat i vad Folkets AI-kommission riktigt Àr Ànnu heller. I första hand var det ju bara en samling runt ett remissvar.
@jamesravey Word! đ
I was at an event yesterday evening with a group discussion. As I often do, I drew a mindmap on my iPad to summarise what we talked about. I've made mindmaps since I was a kid and truly love them as a thinking tool. I also pride myself in having a real skill in picking up important points and quickly drawing connections and grouping them. I've been doing this for more than two decades.
So I presented the mindmap at the end and of course people asked "What AI tool did you use for this?"
Oh how I loathe living in this timeline.
The argument that humans make mistakes too so we should be okay with AI making mistakes is⊠bizarre. I don't use a calculator for it to be wildly wrong some of the time and not indicate any sort of confidence level. I use a calculator when I want the right answer, ALL THE TIME.
Sure, I also don't ask the calculator to write poems in the style of Mad Hatter⊠but maybe, just maybe, I can do without that.
Also, again, "AI" is not sentient. It's actually not 'making mistakes'. It's just operating according to its programming. Which, people for some reason forget, is entirely human-made. Yes, even when it's probabilistic. "Hallucination" is a misdirection.
@datarama Also, itâs still a black box with all the dubious and prejudiced qualities this necessarily entails.
Try out Lumo, they said. It's a secure "AI" from Proton, they said. You can trust it, they said.
So I wanted to show what a joke this also is.
And yes, this is even more hilarious than many of these types of examples I've seen in the past.
Starting now: What Autonomous Vehicle Failures Can Tell Us About Engineering Safety-Critical AI, with Prof.dr. (and former jet pilot) Missy Cummings
https://nmclive.tudelft.nl/Mediasite/Channel/seminar-ai-robotics/watch/a525e686be53457e9dad690bfedf51721d?sortBy=most-recent
Tip: When you use DuckDuckGo's web browser you can generate a free email address that forwards to an email address of your choice but also removes trackers along the way. Handy for subscriptions and "giveaways".
For each email you get a report on what trackers were removed.
https://duckduckgo.com/email
@EwanIngvald If you add the url for a newsletter, the feed reader should autodiscover it.
But the pattern is this:
Newsletter:
https://sarahkendzior.substack.com/
RSS feed:
https://sarahkendzior.substack.com/feed
Jag har vÀrderingar, som jag mÀrker blir tydligare med Älder, med erfarenhet och kritisk reflektion. De förÀndras ocksÄ. Men jag vete fasen om mina vÀrderingar har en nationalitet.
Your regular reminder that all Substack newsletters have an RSS feed. That means you don't have to give them an email address for tracking or annoying marketing. Which is especially nice when you want to avoid the platform as much as possible but some good writers insist on publishing there.
I personally use Feedbin as my RSS reader and am happy to recommend it.
If you're on #LinkedIn, this is the link for declining their use of your data for training their AI models:
https://www.linkedin.com/mypreferences/d/settings/data-for-ai-improvement
@kolombiken Hahaha đ
@kolombiken
BetĂ€nk nu att jag postade den hĂ€r, och kom pĂ„ att jag ville anvĂ€nda den i min presentation. Men pĂ„ svenska. SĂ„ jag spĂ„rade upp vilket avsnitt av My Little Pony det handlade om, sökte fram för att hitta scenen. Laddade ner klippet frĂ„n YouTube och klippte ut kramen. La pĂ„ svensk text och exporterade för anvĂ€ndning i min Keynote. đ