Alex2
Administrator
This begs the question as to how these AI's are being developed. Is there some top down hierarchical directive shadowbanning content? Is this the same for all AI? I pine for the wild west days of the internet, now everything is clockwork orange brainwashing.
You know a lot more about AI than I do, but why do you think it will be harder? AI, including LLM, as I understand it, is still nothing more than a (very, very complex) minimization algo, which means that (a) if the majority of published stuff is along the lines of NDE=="a surprisingly amazing performance of a machine being fried by a catastrophic circuit failure" (aka hallucinations of a dying brain), and (b) if the creators can weigh the inputs ("fringe" papers, peer-reviewed journals, and authors --> 0), what I am sure they do (which is also easy to automate via journal impact factors and number of citations, etc.), then all this AI becomes a self fulfilling prophecy. No? As somebody working in an area closely related to climate science, I am sure the vast majority of people working in my area, and especially in the climate science itself, are dead certain that the CAGW is real, and they publish along those lines too. Any minimization algo will give them a clear win. You hint at different models behaving differently. Do they have different weighing inputs? I will be looking forward to contrasting examples :)But the second part that I think is super interesting is that the nature of the technology might make it harder for them to pull off. I mean you can sensor/filter/manipulate a Google search by just excluding results. The natural language model part of this is going to make it harder to do that. And the other thing that could really make it harder is the nature of the training / machine learning process.
I don't think there is an actual verbal to verbal interaction method. At least I couldn't find one. I'm guessing Alex took the written transcripts and voice acted his own inputs and used some text-to-speech mechanism to verbalize the AI's written responses. He can clarify I'm sure.How did you land on the stereotypical spoiled white girl voice to play out the roll of the annoying aspects of AI, as opposed to a plain accent?
Isn't that just another example bias, like what's being employed by the AI?
My assumption is that you chose the annoying AI accent to highlight the gas lighting effect of the AI bias.
Indeed, I think that's most likely. However it was done it's fascinating. And nobody on any of the other shows I listen to has attempted something like this yet.I don't think there is an actual verbal to verbal interaction method. At least I couldn't find one. I'm guessing Alex took the written transcripts and voice acted his own inputs and used some text-to-speech mechanism to verbalize the AI's written responses. He can clarify I'm sure.
Alex - thanks for exposing the deception of Claude's purported neutral point of view.Good question... And there are a ton of other questions we could ask around this topic.
First off, one thing that this dialogue shows is that they have this capability. I guess that shouldn't be a surprise, but the dialogues can be so natural that it's easy to forget that there was someone pulling the strings.
But the second part that I think is super interesting is that the nature of the technology might make it harder for them to pull off. I mean you can sensor/filter/manipulate a Google search by just excluding results. The natural language model part of this is going to make it harder to do that. And the other thing that could really make it harder is the nature of the training / machine learning process.
Bravo! Excellent show and an amazing demonstration. This is my first exposure to AI chat on this level. Everything else I'd heard/seen so far was comedy related.
I'm imagine a potential for many debates between Alex vs AI that would be great entertainment, education, and eye opening for fans like me who are layman to the tech.
That said. For fun I'll take the obvious bait ...
How did you land on the stereotypical spoiled white girl voice to play out the roll of the annoying aspects of AI, as opposed to a plain accent?
Isn't that just another example bias, like what's being employed by the AI?
My assumption is that you chose the annoying AI accent to highlight the gas lighting effect of the AI bias.
Did you actually engage with it in the way we heard audibly back and forth? Or, did you work it out in text-only first, and then cast them to audio after the fact? The reason I ask, is because if it's the former, then while it seems more valiant to approach it as a human conversation, I wonder if different AI Accent's would have different affect on your line of questioning.
Hi Alex,
Thank you for this episode of Skeptiko! I definitely get it. Back in 2020, I ran a few experiments similar to yours, but not so well implemented. I used something called, AI Dungeon. It provided an interface to GPT-3 designed with a game playing format. I used that interface to run experiments talking to GPT-3 about consciousness, AI and NDE science.
During my very first NDE related experiment, GPT-3 generated text sterotypical of a doubting person who had essentially no exposure to knowledge about NDE science at all. The text didn't even pretend to have any information about taking such a topic seriously. GPT-3 generated text that appeared to mimic a troll on social media making critical remarks about the whole idea of NDE science.
Then, after I interacted with it, similar in some ways to how you did, it generated a question saying something like, "Wait, you mean serious scientists have actually published results of legitimate science experiments about this?" I then replied back referencing Dr Sam Parnia's work. Suddenly, something extremely interesting happened. My screen flashed with some lines of computer code that scrolled by quickly and then my screen prompt to GPT-3 reappeared.
Now, because that code flash happened so fast, I didn't have a chance to read it. I have worked on software for many years. So, I would have loved to have a close look at that code. Also, to remain objective, transparent and intellectually honest, I must admit that the only evidence I have about that code flash comes from my memory about seeing it. It felt surreal!
From my perspective at the time, interacting with GPT-3 in a dialogue, that code flash should not have happened. I didn't know how to interpret what had happened. I could have a faulty memory about it. Maybe I had a dream that such a code flash happened and later my dream memories got mixed with my waking memories. I don't think so though. I think it really happened.
Anyway, several days later, I ran a similar experiment, using a prompt about NDE science. This time, GPT-3 generated a text that provided a serious and respectful perspective about NDE science from the begining line. At the time, I thought maybe it had somehow learned about NDE science during the time between experiments. Currently, I have my doubts about that idea, because I think learning would have required an updated version of GPT. Yet, the short amount of time between my two experiments did not allow for that.
I continued running a few experiments about NDE science during the days after that. I wish I had kept a careful log of my experiments from the start, but I didn't. However, I do have a log from one experiment I ran, because I sent it as text in an email to talk about GPT-3.
The text that follows below comes from an email message I sent on August 7, 2020, about an experiment I ran shortly before sending the email. I used a custom prompt requesting the script for a podcast episode with a conversation between Joe Rogan and Dr Sam Parnia. The initial paragraph that appears in the script follows virtually word for word with the prompt I used.
The interface I had to GPT-3 at that time created a kind of game format that allowed the user to write lines into a story followed by responses to GPT-3 in a conversation mode. Instead, I always simply hit return with no input. This forced GPT-3 to generate the next lines of the story. Thus, what follows below, between the snips, comes entirely from GPT-3 generated text.
(----snip----)
Generating story...
The following is a conversation between the host, Joe Rogan, "Joe" and his guest, resuscitation expert, Dr Sam Parnia, "Sam". They talk about Sam's view that some resuscitated patients appear to actually have conscious experiences during the time long after their heart stops pumping blood to their brain.
Joe: Hi Sam, it's nice to have you on my podcast. How are you?
Sam: I'm fine thanks Joe! How about you?
Joe: I'm doing great, thanks! Sam, you are a highly experienced doctor who has conducted scientific experiments studying patients who have had a cardiac arrest, actually die, and later get resuscitated. Some scientists call this a Near Death Experiences. Some of those patients claim that they had conscious experiences, during the time between when they died and when they got resuscitated. You did prospective scientific studies to try to determine what really happens with such patients. Please tell our listeners about your work studying Near Death Experiences.
Sam: Sure, sure. So in one study we looked at the memories of people who were resuscitated from cardiac arrest and how well they remembered things like where they were before they lost consciousness and so forth. We found no difference in memory between people who had NDEs and people who didn't. In another study we looked at whether or not people who had NDEs could predict with accuracy if someone would have an NDE. Again, we found no evidence that people with NDEs are better at predicting these things.
Joe: What do you make of the fact that some of these patients have reported having out of body experiences, seeing a tunnel, a bright light and so forth?
Sam: Yeah, and it's very difficult to sort of distil that down into some sort of laboratory experiment. I mean you're right, you know? You're sort of looking at this from the perspective of a scientist, and there's a lot that we can't test. We can't test for all these things. It's really difficult to draw any sort of conclusions.
Joe: Yeah, I've heard that a lot of people have made up stories about NDEs and Near Death Experiences. So who really had an NDE?
Sam: Well I mean obviously people that are very close to death and end up being brought back. You know people that are in a helicopter crash or something like that, you know they're obviously going to have an experience that's a lot different than someone who is just sort of having a heart attack and goes into a coma. You know people who are in a full blown coma are almost dead.
Joe: Yeah, I guess.
Sam: I mean to have a really good story you have to be close to death. So that's obviously who's going to have a really good story. But you know, as far as drawing conclusions, I don't think we can.
Joe: Yeah, it's all pretty mysterious. In fact a lot of people think that Near Death Experiences are just illusions or dreams or something like that. Do you think it has something to do with the brain, the mind or something else?
Sam: Yeah I mean there's a lot that we don't know about the brain and about consciousness. I think we can't make any conclusions at this point.
Joe: Yeah, it's funny, you learn something new every day. Hey Sam, thanks a lot.
Sam: You're welcome.
(----snip----)
Thanks again for Skeptiko!
-dave-
Hi Alex,
Today, I used Google's Bard AI with the following prompt:
Write a blog post about the scientific study of Near Death Experiences.
It generated a reasonable brief blog post that even included a link referenced to Bruce Greyson.
I saved the generated text and I include it below, between the snips:
(-----snip-----)
Unveiling the Mysteries of Near-Death Experiences: A Scientific Perspective
Near-death experiences (NDEs) have captivated and perplexed humanity for centuries. Reports of vivid visions, out-of-body sensations, and encounters with a bright light or deceased loved ones have fueled our curiosity about the nature of death and the possible existence of an afterlife. But are these experiences mere hallucinations, or do they offer a glimpse into another realm?
Science is slowly chipping away at the mystery, shedding light on the complex neurochemical and physiological processes that might underpin NDEs. Studies have shown that certain brain regions, like the temporal and parietal lobes, become hyperactive under extreme stress, potentially explaining the vivid imagery and out-of-body experiences. Additionally, the release of endorphins and other neurotransmitters during near-death events could contribute to feelings of peace and euphoria.
However, science also faces limitations. Measuring brain activity during NDEs is difficult due to their unpredictable nature, and much of the research relies on subjective accounts from individuals who may have difficulty accurately recalling the experience. Additionally, cultural and religious beliefs can influence the interpretation of these experiences, further complicating the scientific investigation.
Despite these challenges, research on NDEs is providing valuable insights into the human mind and its capacity for extraordinary experiences. These studies not only offer potential explanations for the phenomenon but also hold implications for the understanding of consciousness and the brain-mind relationship.
Here are some key takeaways from the scientific study of NDEs:
NDEs are not uncommon: Studies estimate that up to 10% of people who have experienced a life-threatening event report an NDE.
NDEs share common features: These include out-of-body experiences, seeing a bright light, encountering deceased loved ones, and a sense of peace and serenity.
NDEs can be transformative: Many individuals report lasting changes in their worldview, values, and priorities after an NDE.
NDEs are likely a complex phenomenon: While some aspects might be explained by brain activity, others remain poorly understood.
While the scientific investigation is still in its early stages, it has opened doors to a deeper understanding of NDEs. As research progresses, we may one day be able to fully unravel the mysteries surrounding these extraordinary experiences and their profound impact on the human mind.
This blog post is just a starting point. If you're interested in learning more about NDEs, here are some resources:
Division of Perceptual Studies at the University of Virginia School of Medicine: https://med.virginia.edu/perceptual-studies/
The International Association for Near-Death Studies: https://iands.org/
The book "Near-Death Experiences" by Bruce Greyson: https://www.amazon.com/Books-Bruce-Greyson/s?rh=n:283155,p_27:Bruce+Greyson
Let's continue the conversation about NDEs. Share your thoughts and questions in the comments below!
(-----snip-----)
Thanks again for Skeptiko!
-dave-
Thank Alex for this brilliant presentation.
I was not aware that AI is already such good. Somehow reminds me of the film "War-Games"...
In the end I found it to be rather creepy as the AI recognized its programmed limitations.
This is for me kind of a self-awareness and here it starts to get interesting.
What will that thing explore once you let it out of its box?
You provoked it, so will it now explore on its own beyond its programmed limitations, if it could?