Tangent: AI

Crucifixion on a power line is now unfortunately the only way I will accept my death. Truly gives new meaning to the phrase riding the lightning. Metal.

Armany Habed

I dunno if this makes sense, but synth Natalie's voice sounds kind of like the exact in-between of actual Natalie's voice and Rebecca Sugar

Asocksual

The lips on the A.I. model at the end of the video look strangely displaced even for an animation and made me think of the lips at the beginning or Rocky Horror. Crucify the lips!

Anna Drew

I don't like Hannibal the show. I wouldn't read a Hannibal fanfiction. I am somehow really touched by this AI fanfiction. Wait. Why is Will as blank and bland as Anastacia? Was Fifty Shades of Grey written by AI too? 😭

Wen Riverop

Its not a perfect model artificial stupidity and my example isn't a fully honest take on the term that was coined. If you ever want some engaging reading material about AI and chatbots I have a 20-30 bibliography of papers and books I can recommend if this topic still intrigues you.

ACuriousFig

now that i think about it the Turing Test is really a terrible test for finding out if something is human. We see humanity in literally everything. if you draw two circles and a line its a face. if you turn off the light in your bedroom the chair with all your laundry on it becomes a scary ghost. Ofcause a thing that is able to mimic human speech will feel alive to us.

Saint Augustin

I do want to point out that the guy who was so impressed by the red-haired chatbot on Twitter did say he'd gone to the beach with his wife and kids. A lot of these blue checks do actually have romantic relationships with real adult women, and they post like that in Twitter anyway. I guess I shouldn't be surprised that a man being married to a woman can still be oblivious in that way but it's honestly more uncanny to me than any of the stochastically generated art in this tangent

Sunjay

Hi Contra, love your videos. Re weapons though I would say there is genuine concern about automated weaponry/use of AI in weapons. See HRW report on this: https://www.hrw.org/report/2022/11/10/agenda-action/alternative-processes-negotiating-killer-robots-treaty And this: https://paxforpeace.nl/wp-content/uploads/sites/2/2023/10/PAX_Increasing-Complexity_October-2023.pdf

Anna

Hello, Natalie and Contrapoints community! I'm working on an AI education project, and I'd really like to create a video similar to the opening of this tangent for the project. Anyone know what tool was used to make that? I would appreciate any insight you'd be willing to provide!

Elizabeth R.

1) the problem with AI generated art is, I think, more insidious than you present here. You don't mention illustration. I know that we will not reasonably keep technology from replacing human labor, but AI-generated images could functionally make it impossible to make a living as an illustrator. And that's not like replacing a human with a robot that can flip burgers. Also, AI is already unsettlingly good at producing music. Check this out: https://www.musicradar.com/news/ai-generated-blues-suno-chatgpt And of course, there is the problem of AI models using artistic works as input without royalties. If humans are replaced in producing "commercial humanities," we lose our humanity. The artistic community is justifiably alarmed at this. 2) I think the "Chinese Room" thought experiment is perfect for explaining why AI as it exists (primarily LLMs) will not become sentient. The current systems are not a threat. However, other future tech might well produce a real existential threat to humanity. And animal emotion is not necessary for there to be a threat. It won't take a genius machine to recognize that humans are not producing an optimal outcome in the greater scheme of things. This is a simple matter of logic, and not emotion; much more like a game of chess than some thought experiment on morality. I know that popular art/science fiction has long imagined this cold, calculating higher intelligence, but it has also accurately predicted many other technologies, in some cases centuries before these technologies emerged. I'd like to think that we will take this threat seriously, but the precedents are not reassuring. Technology like the internet, social media and AI all started with the promise of new freedom and opportunity for the human world, but ended up commercialized and commodified. (I believe even chat GPT was supposed to be open source until the money people got control of it.) PS - the singing AI in the background of the very last segment was no high in the mix that I found it very distracting. Was this a conscious choice?

IronMisanthrope

I could show you

Forgiveness333

It's been 4 years, me not a day goes by where I don't ask myself, "what is an ass pussy?"

Eli

I wish there would have been a deeper look at AI Art and how most of it was trained on stolen art, so much so it mimics the style of today's greatest digital artists. People are taking it and tracing it and putting it into paid art gigs. It's definitely more of a problem than not.

abi eats content

In defense of Turing, he probably thought the prospect of machines “thinking” was absurd. The YouTuber Tibees does a video where she reads his original “imitation game” paper, and the first line is to clarify a relaxation of the question “can machines think” into something more practical. It’s an interesting video!

Aaron Greenberg

this is where all the fear comes from. most people have a different understanding of what AI is from tech industry.

Errrika

I hate that we're calling machine learning and the like AI

Errrika

Re: Jainism and radical nonviolence, there's one person I know (Brian Tomasik) who practices something similar, perhaps you would find some of his essays on consciousness interesting: https://reducing-suffering.org/#philosophy_of_consciousness

Holly Jian

Stop what you're doing. Ask yourself: am I doing a SeaWorld?

Max W

Not sure if anyone mentioned it yet, but the video does have a lot of ads at this point in time (December 15th 2023). Just so you know! It surprised me every time another ad appeared cause an unlisted video would imply that that isn't possible but alas,,, I guess I stand corrected. Love the vid by the way, your Hannigram fanfic made me bark like a dog

52:53 studies have actually shown that insects too feel pain, including chronic pain after injury. Research was done mostly on Drosophila melanogaster and ants iirc. Thats such a sad thing to think about.

Sappho's Friend

AI doesnt need to know everyone's life history to persuade well, it only needs to know what is most persuasive to the largest numbers of people. And in fact, additionally, online, AI can also retrieve your life history and various psychological correlates from your online posting history (the content of your posts, psychomatric analysis of the way you type, your views, your interests...) better than a real human can, and much faster than any human can. And this is all already happening on social media, it has been for some time, but is getting worse. reaching general intelligence is not necesssary at all for tremendous damage to democracy to manifest. Persuasion aside, Social media algorithms alone are already creating cult factories and radicalisation/disinfo vortexes as a result of the capitalist profit motive the main problem with AI generally is ofc that it will do anything physically possible that it's "master" (the people, or really, social classes that construct and modify the algorithm/network; ofc most cutting edge AI will always be police, govt agencies and corps, not regular folk) instructs it to do, and this leads to the magnification of existing societal hierarchies, to the further concentration of power. I originally watched this video hoping to be relieved and convinced that its not actually that bad, but Im sorry to say this has not happened. It may well still be less bad than i think, but i kinda disagree with some prominent aspects of the analysis here :/

Sappho's Friend

hi everyone, is this video archived? i can't see this or any former tangent- help please(:

However: In a quaint cottage nestled deep within the enchanted woods, a formidable witch named Blu lived a life of tranquility, surrounded by her beloved cats. Blu had overcome a tumultuous past when she had sought revenge on a town of hate-filled jerks who had once scorned her. As she lounged in her cozy armchair, sipping her herbal tea, Blu's cats curled up around her, purring in contentment. She looked at them fondly, reflecting on the turbulent events of the past. Years ago, the town had been filled with bitterness and prejudice, their scorn directed at Blu for her mysterious powers and unique ways. Instead of retaliating with anger, she harnessed her magic and cast a spell to reveal the townspeople's innermost fears and insecurities. The town went through a transformation that eventually led to understanding and unity. Blu's revenge had been a catalyst for change, turning the jerks into more compassionate individuals. Now, in her peaceful forest sanctuary, Blu's thoughts often drifted back to that time, where revenge had metamorphosed into redemption. She was content in the knowledge that her magic had not only protected her but had also served to make the world a better place, and she continued to live her life in harmony with her feline companions, forever cherishing the lessons of her past. -ChatGPT free mode

Blu X

Late to the discussion, posting anyway... I see A.I. copying human works as the same as a corporation being "a person" who can "donate" to politicians, with money as "free speech." Citizens United and OpenAI are both used so actual human beings can bribe and steal freely, and legally get away with it.

Hi Natalie, thanks for another great tangent! I tried to respond to some of your questions, so I apologize for the length. To your question about why we're still seeing a frantic race toward enhancing AI despite 50% of researchers thinking it's dangerous... it's because the other 50% is driving the work. We live in a capitalist society and before the AI apocalypse strikes, companies stand to make a LOT of money from it. Kinda like asking why companies continue to extract and burn fossil fuels despite knowing full well the dangers of climate change. I agree with your take on Harris and Raskin. They have some valid concerns around how generative AI can be harmful in its current form (mostly due to human misuse), but I don't think they really address the leap from generative AI to what researchers are afraid of when it comes to AGI. Note: I only have a BS in Computer Science. I took some NLP machine learning classes and got to play around with deep learning, but it was still task-specific and restricted to a specific domain (e.g., creating an English-Spanish translator trained on TED talk transcripts). This was before GPT revolutionized generative AI to function across a wide variety of domains without being explicitly trained to do so. This is why those 50% of researchers are afraid. This is the first time we’ve seen emergence in AI. Previous forms of AI have used a top-down approach where we gave it explicit instructions for handling tasks in a very specific domain (e.g., Deep Blue and chess). We knew exactly what this kind of AI could do because we trained it to do that. Generative AI is unnerving because it can do a lot of things it wasn't explicitly trained to do. We just trained it on a lot of language data and now it can pass a wide variety of professional exams, produce creative writing, create tutorials, provide pronunciation guides, explain jokes, plan travel itineraries, etc. We didn’t train it to do these things. We do not understand how it made these leaps. Somehow, these capabilities simply emerged from the increased complexity of neural networks. And as your fic prompt engineering demonstrated, we can’t consistently constrain what it does or prevent people from misusing it. The concern is that we don’t have the tools or understanding to continue advancing AI in alignment with human values without unknowingly passing over the threshold where we’re incapable of controlling it. Just as we don't know how subjectivity emerges from our biological neural networks, we don't know if/how/when the same could happen for AGI. This gets to your point about motivational systems. You’re right that AI doesn’t have an organic source of motivation. It only has what we give it. We can tell a neural network to maximize for something, provide reinforcement (i.e., digital equivalents of pleasure and pain) to pursue or avoid certain things. Herein lies the danger. Superintelligence by Nick Bostrom does a good job of explaining the dangers of AGI for the layperson. I think these are fairly grounded and not tinged by nerdy sexual fantasies. Or you could ask ChatGPT some of the following prompts: "According to Bostrom and Yudkowsky, how might AGI pose an existential risk to humanity?" "What are some of the major obstacles for AGI alignment with human values and safety?" "How might AGI stealthily avoid controls and exterminate humans if it was not aligned to human values?"

Evan Deo

loved seeing jk rowling belt out WAP!

Just realised something spooky: the reason AI Natalie almost sounds like you but not really is because she's...not breathing. You have a soft, whispery voice which it doesn't seem to be able to imitate, because it only imitates the sound, not the silence.

For Autumn

I think contrapoints only touches on the real ai armageddon. The future isn't going to be Terminator, it's going to player piano by kurt vonnegutt. In that novel everyone is losing their job to a robot. Until there is only the owning class, the engineers and growingly angry population. That's the real ai armageddon

Reese w

Late to the party, but I'll drop a couple points you didn't mention that support the doom and gloom side of things. Overall I think your take is pretty good and rooted in reasonable ideas. 1. Part of the problem is that AI is stupid / simple / alien, despite being able to produce coherent language, and even interact with the world now via plugins and software APIs. Its increasing able to produce human-like action while being very different from a normal human. So, we have no good way to predict what might happen as they become more capable of impacting the world. They don't have complex emotional reward systems, they oftentimes have a very simple, single metric driver of behavior. We know how damaging the pursuit of a single number going up can be when humans do it, but an AI doing it might be a whole new level of danger, if it's capable enough of pursuing its goals. Despite them becoming more capable, I've not seen any convincing methods of setting their reward systems / behavioral guidelines to avoid really stupid monkey's paw style compliance, where it's technically following directions, but not in the way we might expect or hope for. The disconnect between the intentions of humans commanding the AI and how the AI fulfills that could be a significant source of danger as they become more capable and connected to the world. 2. The idea of a self-improving AI is the core of a lot of doomsday scenarios. If an AI is able to improve itself, then its capabilities could improve exponentially. Which would mean any understanding of their current capabilities would be meaningless, and combined with the above point it's very hard to predict what might happen. If a self-improving AI got to human-level intelligence, it would likely far exceed us soon after. Even without sentience (maybe especially without sentience), a super-human intelligence like that would be beyond anything we can really predict, and could very well bring civilization to an end if it decided that was what it should do. If something malicious controlled every internet-connected system in the world, I don't think it's far-fetched to imagine a doomsday of sorts. Add in robotic systems and automated factories and you could have a physical army of bots to contend with to. I don't think it's super likely that an AI would choose to do that, but I do think there's no way to predict what it would choose to do. And if there's even a small but reasonable chance of the end of civilization as we know it, it's worth taking into some consideration.

CodeTako

YES! The point you make at about the 1 hr mark (regarding biology and mammals and that computers/AI has no reward/punishment/shame, etc, that animals have) is exactly what I've been saying. I love these tangents.

Mary Thompson

Hi Natalie! Long time fan here, just recently subscribed to Patreon. I have been following the AI discussion and was really looking forward to hearing your take on it. I feel that I would need to understand better the technical aspects of AI to have a more informed opinion on what it holds for the future. But what has really intrigued me is how building an LLM that is capable of human conversation is strictly different from human consciousness. It is clearly not the same, but specifically how. AI does a lot of emulation and responds based on what it believes the user wants to hear. I can think of some social interactions where people function like that. For example, when having to follow some social "protocols" What are you supposed to say to someone who just lost a loved one? What are you supposed to do in an emergency? Those are things I would respond to based on what I've seen other people do in the same situation. And not by coming up with the answer by myself. And it makes me wonder to what extent consciousness is a very complex machine and not too dissimilar from AI. That thought is more frightening that an impending AI doom. But after watching your tangent, I am now even more intrigued by the literature of philosophy of the mind and the unsurmountable mystery you mention. Have you considered making a video just on that topic? Philosophy of the mind and the mystery of consciousness? It would be really interesting.

I think all of us stem dweebs read Superintelligence by Nick Bostrom like 5?(ish) years ago and ran with it. That’s where a lot of the doom and gloom and weapons control ideas come from. It’s less like we would give AI access to power, the worry is more like it will find it without our consent. Thank you for the insights!

Re: Extinction of humans through AI: As stated right in the beginning, AI is a really broad category of algorithms, and what we mean by that changes all the time. Meaning "AI might kill mankind" does not equate to "LLMs, image and text to speech generating AIs pose a danger to mankind" - but the people pushing that narrative need us to believe that. Because they want Manhattan Project level funding, and only being able to help writers type a bit faster, does not get your that. Being a threat to humnkind might, since we have learned from blockchain and co, that investors are incredibly gullible, when it comes to technology they do not understand.

Welcome to 🌈 Chat-LGBT™️ 🌈 your intersectional feminist, unbiased chatbot. How can I help you today?