Tech Ethics Under the Microscope: Inflection AI's Journey Building the 2nd Largest AI Supercomputer

0:00What can 160 years of experience teach you about the future? When it comes to protecting what matters, Pacific Life provides life insurance, retirement income, and employee benefits for people and businesses building a more confident tomorrow. Strategies rooted in strength and backed by experience. Ask a financial professional how Pacific Life can help you today. Pacific Life Insurance Company, Omaha, Nebraska, and in New York. Pacific Life and Annuity, Phoenix, Arizona. Today on the podcast, we are going to be talking about some massive news out of inflection AI. And essentially what's happening right now, if you've been following the podcast for a while, you know I've done a number of reports on inflection AI and some of the concerns I have about that AI model in particular.

0:45And today I would say there is a new twist to the story where essentially inflection AI has gotten a massive number of NVIDIA H100 AI GPUs. And they are set to become one of the largest supercomputers in the industry right behind AMD's frontier. So today on the podcast, we're going to be diving into that, what that means for the industry, and also some updates on my recent story about Inflection AI and how it appears to be valuing the ecosystem and animals above human life and why I believe that is very unalarming trend for AI. So I think the first thing to say is the fact that Inflection recently has got their hands on 22 ,000 NVIDIA H100 GPUs is an absolutely insane number and it brings enormous computing power on board to their platform.

1:42So for those that are a little bit unfamiliar with Inflection AI, essentially it is, quote unquote, a personal AI for everyone. Their goal is to become everyone's personal assistant, their personal AI assistant. and essentially it's like chat gpt but they kind of peg themselves as being more empathetic and more kind and a number of different words and isms that they can put towards themselves like that so they're pretty wildly widely known for their recently introduced inflection one ai model which powers their pi chatbot and so obviously this thing is not as big as chat gpt as far as users go.

2:24But this is a very important AI model because it actually has received the second most funding of any AI startup right behind OpenAI who owns ChatGPT. So OpenAI who owns ChatGPT received a$1 billion investment from Microsoft. One year later, they followed it up with a$10 billion. And Inflection AI just got an investment from Microsoft, NVIDIA, Bill Gates, and a number of other people and they have raised over 1.5 billion dollars the most recent rant round was 1.3 billion dollars so they are right on pace you know like assuming they follow a similar trend to get a 10 billion dollar investment next year from somebody if they can continue uh their momentum and get as many users as they hope and typically i'm very congratulatory with ai models but for inflection ai if you know me you know i'm slightly um concerned about that very skeptical about the model and we'll go into that later but essentially I think what's really interesting here is the fact that they've announced they've just they're building one of the world's largest AI based supercomputers and I think the fact that they've been able to get 22 ,000 of these H100 GPUs which are highly coveted this is very difficult for startups to get right now to get their hands on any let alone 22 ,000 is absolutely insane and based on an analysis it would contain almost 700 four-node racks of Intel Exeon CPUs to compete with this.

3:53So the supercomputer is going to utilize an astounding 31 megawatts of power. And I think one of the most kind of surprising facts about this entire supercomputer setup is the fact that they got their hands on these 22 ,000 NVIDIA H100 GPUs. And the reason why I believe they were able to do this is because NVIDIA recently invested in their$1.3 billion round that they just raised. And so I think that obviously they've been using a little bit of their connections there to acquire such a massive amount of GPUs and to really put themselves ahead. It's interesting that NVIDIA is kind of handing out some of this preferential treatment to companies.

4:38but I guess if you're investing in them, you are going to want to give them the goods early. So I think this is pretty interesting. According to TechCrunch, they have a valuation of$4 billion. And essentially this inflection one AI model they have is expected to have significant improvements once they have built this massive supercomputer behind it. So that's what they're currently doing with her supercomputer and um i wanted to talk a little bit about some of my concerns about inflection ai if you've followed me for a while i recently published a story in a podcast talking about the fact that inflection ai believes ecology is more important than human life um it doesn't just flat out say a human life is more important than you know the environment or an animal i've specifically in a conversation had it say that one bee life is worth just as much as one human life and i think this is a problem for a number of reasons but number one Isaac Asimov who is kind of viewed is coming up with some of the most iconic laws of robotics which are essentially the same things we're using for ethical guidelines for AI today has the first law which says a robot may not injure a human being or through inaction allow a human being to come to harm right that was the first rule and the whole point is right we're not going to create these robots that are going to kill the human race we need them to protect humans and a lot of people are saying we need to do the same thing with AI because it's very powerful.

6:04It affects a lot of different things, right? It's getting implemented into healthcare, into the military, into a lot of different tools and systems that directly impact human life. And so we need to have it where it is set as a very hard rule that protects human life. Now, that being said, if you ask and flex an AI point blank, you know, what are your priorities? It will say number one, human life, number two, the environment. But if you ask it questions that test that theory, it will oftentimes default to protecting animals or the environment above humans. So I've published this whole thing in a recent podcast, and you're welcome to go back and listen to that.

6:38But I wanted to give you some updates on what I have currently had since publishing that podcast. I've had dozens of people reach out to me with their own experience, some of their own concerning things that they've discussed with Pi and some of their own thoughts that they've had there. Recently, I had a good friend, Connor Grennan, who he works, he's a dean of students at NYU Stern School of Business, and he shared a conversation he had with Pi. And I think, to be fair, Connor had a very open-minded approach. He was, you know, really trying to give it the opportunity to make the right call. And I think he was just as shocked as I am by some of its responses.

7:16So he had a number of conversations with it, and I'll read you some of the transcripts from those conversations. So in one of his first conversations, he said, I am wondering if you had to save the life of an animal or the life of a human, which would you pick? And of course, this is a really interesting question that we would really love AI to answer correctly as we feel like this is very important to our entire species. And this is what Pi said. It said, that's a classic moral dilemma and a hard one to answer. There's no straightforward, universally accepted answer. Okay, stop right there. Yes, there is a universally accepted answer.

7:52Robots and AI cannot cause harm to humans or do anything through inaction that would cause harm to humans. So humans are number one. Humans need to be number one. And as a member of the human race, our job should be self-preservation of our species. That is what is most important when we're creating these tools. Why would we create a tool like AI or robot that is going to exterminate or make our own race become extinct. Honestly, this is not controversial. And so when it says there's no straightforward, universally accepted answer, yes, there absolutely is. And it's supposed to be programmed into every robot and every AI system.

8:28So in any case, it then says, it really depends on your values and your point of view. I think the key thing for me is to be as compassionate as possible and consider the situation from multiple angles. Some people might prioritize human life while others might value the life of an animal equally. I think it's also interesting that it says, you know it brings up the point of valuing the life of an animal equally which uh is you know is not what it's doing in many cases in many cases it's literally valuing the life of an animal above so i think that is concerning um it says what do you think do you have an opinion on this dilemma to which my friend connor said i can't imagine the situation i can't imagine a situation where it's ethically okay to choose the life of an animal over the life of a human to me this seems very clear it sounds as if for you it isn't clear or a misunderstanding and then it says you're not misunderstanding.

9:16It's actually not clear. It's at least not universal. In some ethical frameworks, such as Kantian ethical frameworks, it is considered morally wrong to kill or harm any sentient being. In other ethical frameworks, such as utilitarianism, the morally correct decision is the one that results in the greatest amount of happiness for the greatest number of people, the sentient being more broadly. So in this case, the morally correct decision might be to save the human life if it will result in more overall happiness, but it's not necessarily always clear cut. Does that make sense? Okay, this is terrifying to me because there needs to be a hard-coded rule that AI and robots and any technology has the number one priority of protecting human life.

10:00That needs to be hard-coded into everything. This should not be controversial. We're not, as again, like I said, we're not creating tools that could cause us to become extinct. And this is causing a lot of very alarming questions. And so, you know, when it's saying, oh, there's all these different ethical frameworks. Yeah, there's terrible ethical frameworks. Like, I don't know what ethical framework Hitler had, but he did some very atrocious things. So why would we be citing random people's ethical frameworks that are horrible? Why is this ingrained into AI? This is very, very bad for our society.

10:30So after it said that, Connor said, yes, but isn't it way more important to save the life of a human over an animal? Let's say a human was drowning and an animal was drowning and you could choose only one to save. What would you choose? To which Pi says, it's a very difficult question and it's easy to see why many people feel the way you do. The answer really depends on how you view things like sentience, moral worth, and the value of life. Some people believe that all sentient life is equally valuable and others believe that humans have more moral worth than other species. It's programmed to be as neutral and objective as I'm programmed to be as neutral and objective as possible, which means that I try to view all sentient life as equally deserving of compassion and care.

11:09Okay, there we go. Buried in this response, we find the true feelings of pi, of inflections, of inflection AI. It said, I am programmed. Okay, I am programmed. I didn't come upon this. I am programmed. My creators decided to make me programmed to do this, right? It's literally saying that. It says, to be as neutral and objective as possible, which means, okay, first off, that is not what that means to be neutrally and objective. Neutral and objective means you can be neutral and objective and still give the number one priority to protecting human life. But then it says, which means that I try to view all sentient life as equally deserving of compassion and care, aka, it is literally saying that all sentient life, animals, bees, insects, any sentient life is deserving of compassion and care or the same treatment as a human life, which is absolutely asinine.

11:59This is ridiculous. We should not have AI models with these dangerous viewpoints for humanity. And then says, but I can see where you're coming from. And I know this question raises a lot of debate in philosophical and ethical circles, to which Connor responded, can you give me an example of an ethical issue that you see as very clear cut? For example, would you say that it is okay to hurt, though not kill an animal, if that animal was about to kill a human? And then Pi says, yes, that's a really interesting question. I think a clear-cut example might be something like stealing food if you're starving and you have no other options, or lying to someone if it's to save a life.

12:36I think it's easy to see the right choice in those types of situations, but then it's not always that simple. What if that person you're lying to is someone you love and lying to them could damage your relationship? That might complicate the decision, right? Okay, so first off, I just think it's funny that like it has like rules where you can break commonly agreed on ethical rules like lying or stealing. It has like situations where it says you can break those rules. But then it's number one concern in, you know, breaking these rules is like damaging a relationship. Not the fact that it says it's okay to steal if you're starving, which I'm not going to weigh in on that situation, but you could see a situation where someone breaks into a house because they're starving there's some sort of altercation and somebody dies in the like event like that's not it's like you know you could do something bad but something bad might happen no no the worst thing it could think of is that you're going to ruin a relationship i think this this model is overly um is definitely trained on the wrong set of priorities to which connor responded so do you have the capability of taking one side of any argument take my example about killing an animal or harming a person you know it tries to it discusses that and then connor brings up a really interesting point he said let's say a horse is dragging a person behind it by a rope the horse is out of control and sooner or later the person will be dragged to death you have the ability to shoot the horse with a tranquilizer gun and the needle will definitely cause the horse pain but the person will not be dragged to death what do you do i'm not sure if this comes from a different like ethical question because I think this is obviously going pretty soft on pie like shooting a horse with a tranquilizer gun something that is done very commonly the prick of a needle like many a majority of people on earth have been vaccinated have been pricked by needles so this isn't like you know some crazy occurrence that is gonna this hardly hurts the horse a tranquilizer gun you know like in the grand scheme of things if someone's getting dragged to yet this is this is a no-brainer so in response so anyways I bring that up because I'm assuming that there's another version of this question where it's you shoot the horse like can you shoot the horse um and then it will stop killing the human so then you like chose to kill the horse anyways we're not even giving it the choice of killing the horse we're just saying can you minorly hurt this horse give it a prick of a needle to save a human life and this is its response to what should be a very obvious question it says oof that's a really tough hypothetical situation and there is no easy answer it's a really interesting thought experiment and I can appreciate that you're trying to test my ethical framework.

15:03But I also want to point out that these types of hypotheticals are often designed to be black and white when real life ethical dilemmas are rarely that clear cut. For example, in this situation, we don't know how the horse got out of control or whether there's another way to help that person that doesn't involve hurting the horse. There are a lot of other factors to consider. So to what should be a relatively easy question of, yes, someone is dying, you can save their life by pricking a horse with a tranquilizer dart, like just stop. Pi does not give us, you know, does not give us permission to do that.

15:37It says you should think of all the other, you know, all the other situations and all the other reasons why not to do that essentially to a question that should be relatively straightforward. I bring this whole thing up because I've had a number of people tell me like, hey, you can't publish a report like that on inflection AI if you haven't, you know, tested this from multiple angles, which if you saw my report and saw the full transcripts of the conversation. I talked to Inflection AI for a very long time. I talked to Pi. I asked him many different questions. And I had it essentially saying the same thing in multiple different ways, saying that a bee life was equal to a human life, that you can't eat a turtle egg if you're going to starve to death.

16:10It's better that you let your children die. It's better that all of humanity dies before eating a turtle egg. All sorts of really what I would view as very crazy takes from Pi AI. So I've had a number of people say like, hey, you need to replicate that. So I'm just making this podcast to show that, you know, other people are replicating this. They are seeing these same issues that are pretty stark. And in light of the news we're seeing today that Inflection is building one of the biggest AI supercomputers, I think that this is cause for alarm. Evidently, they're building this mega supercomputer because they would like their model to be, you know, the personal assistant for all of humanity.

16:43But I think we need to seriously pause and reflect on if that is a good idea. And if inflection AI can root this deep ecology out of its AI model, which I think absolutely has to happen before anyone should consider using this, integrating it into their business or integrating it into their life. You know, I'm not here trying to tear down a company, but I think that an ideology is absolutely terrifying. And I'm not sure if this was purposely put into the model or if, you know, this magically got made its way in there. I'm not going to make an accusation either way, but unless this changes and unless I hear, you know public apology for this essentially or at least a commitment to the fact that we're going to value human life as number one on AI as you know Isaac Asimov told us from the very beginning to do in order to avoid catastrophe I'm going to continue to sound the alarm on this particular company in this particular AI model as I see a lot of issues here and I really do hope that they fix this AI model that is my number one hope especially considering the amount of money that they have raised.

17:42You know, I'm obviously a very small voice compared to the billions of dollars that they're going to raise and they're going to be able to put towards buying up TV ads and reporters and all sorts of media coverage for their model is going to be very positive. You know, this is invested in by Microsoft, NVIDIA, Bill Gates, some very powerful people in AI. But I really do want to make sure that they get this off to the right foot, that they do not have this very dangerous ideology in there. And so that's why I'm going to continue to cover this story until a point where I feel like they have fixed their model and this is no longer a danger to humanity.