Blurred Lines: The Challenge of Distinguishing AI from Humans

0:00Some shocking news has recently come out about humans and our ability to detect if something was written by AI or not. And along this lines, there's also been two pieces of legislation that have just one is being proposed in the US and one is just being broken or is coming out in the EU that directly, I think, impacts and is kind of addressing this one specific issue. Now, I for one, if you listen to my podcast, I will say I am not typically super pro pro kind of regulating AI right now. I definitely think it's an area that needs to be looked at. I definitely think we need to be working on it.

0:35I would just worry that if we jumped the gun too quick, it could slow us down. So today on the podcast, I am going to talk about some regulation that I think actually can be helpful. I think it has some precedent and I think it has some value and is a good first step. And we're going to talk about why this is happening. So I think the first thing to bring up is the fact that a recent report just came out. If you remember, I think back in April, there was a viral game on Twitter called Human or Not. And essentially, it was one of the largest scale Turing tests to date. And it was just essentially assessing people's ability to determine between humans and AI bots.

1:10Essentially, in the game, you would have a two minute conversation with a bot or a human. And then you'd have to guess if it was human or not. And this resulted in over a million conversations and, you know, people's guesses were all analyzed. And the results showed that only 60 % of participants correctly identified AI bots. So participants very often relied on, I would say, some false assumptions there. You know, they'd expect the bots to avoid typos or grammar mistakes or slang, despite, you know, bots being specifically trained to incorporate these features. So I think that's one way that, you know, they got them.

1:45but like the other thing I would say is this was also you know 60 % of participants got this right so 40 % got it wrong but I think what's important is this was literally people being given like a test and knowing that this was a test on AI or not and I think that that's one thing and you'd actually get a lot higher but if you just come across AI content out in the wild on the internet an image or text I think the likelihood of you getting that right would be much much lower now Now, I bring all of that up because some recent regulation that was proposed in the U.S. essentially proposes having, and this is coming from a congressman, Richie Torres from New York, Democrat, and he is planning on introducing an AI Disclosure Act to address the risks of generative artificial intelligence.

2:33And essentially, his proposed legislation mandates that all outputs generated by AI carry a disclosure statement saying, disclaimer, this is an output that has been generated by artificial intelligence. Now, I think that this isn't very far off as far as precedent goes. His idea here is that the FTC Federal Trade Commission is going to be entrusted with the responsibility of enforcing this requirement. And, you know, I know I'm not a huge proponent of regulation and AI right now. I think this one actually is not bad. And the reason I say that is I believe there's precedent, you know, coming from a marketing background.

3:06There was a lot of issues back in the past with influencers and they would be promoting a product or talking about something they loved. And in reality, it was an ad, but they wouldn't tell you that. And so it kind of felt like a conflict of interest where they were just, you know, maybe tricking people into using a product. Who knows if they used it or not. And so eventually that came out with the rule where you had to do hashtag ad or you had to pretty much tell people that something was a sponsored post. and you know you see this in youtube it's not like that clunky but an influencer will be like hey i love this product i would have you know talked about it anyways but today it's also a sponsor on the thing so i don't know if they would have actually talked about it anyways but you know i do appreciate the honesty and messing around with the ftc federal trade commission is not something a lot of people want to do so i think that this you know putting it in the same kind of ballpark uh here for you know hashtag sponsored ad or hashtag ai generated is going to be um is going to be a good move.

3:57I think, you know, if you just want people to know, if you don't want to worry about viral content going viral where people don't know what's written by AI, I've already seen like on Twitter, in the Twitter comments, when people are arguing one way or another politically, it's very prone for people to like start showing proof of like a person being bad. And they're showing like all these AI generated images. And people are arguing about them. Some people are being tricked by them. I just think it's a obviously ridiculous, like why does that need to exist but b um you know it is some people are using it deceitfully intentionally um and so i do think you know making it legal that you have to uh say this was generated by ai because essentially this isn't like a meme it's not like a uh you know a piece of artwork like people are using these to mislead people there's images that are meant to look literal or real and likewise i think it's just good to have a disclaimer on content written that's generated by AI.

4:55Because, you know, if you're reading an article, and it's generated by AI, you don't know that things been fact checked by humans, maybe they have a note at the top, this was generated by AI and has been fact checked by our experts at health line, blah, blah, blah, or whatever company decides to do this. I think even that would be great. I personally wouldn't hold against the company because I use AI generation for different businesses and different things I use. So obviously, the tool is going to proliferate into everything. But I think people do want full disclosure, and people do want that transparency.

5:21I think that's going to help them to make informed decisions and know how trustworthy content is. And so, yeah, personally, I'm actually all for this proposal. I think that this was a good move from Torres and his office. Now, do I have I read the bill? No. Did they slip sneaky things in there that might be bad or like, you know, on the face, this might look like a good idea, but there's other stuff in there? Yes. I mean, that's politics. I see that all the time from Republicans, Democrats, both sides where they'll propose a bill and there's, you know, this is the idea of the bill, but there's sneaky stuff in there.

5:50So I can't say I specifically, I guess, endorse this bill without reading it. But conceptually, I think that is a good concept. And so it's left to be said how that gets litigated and if that gets passed. But conceptually, I think that's a great idea. And it is interesting to say that this is also a battle that is currently happening. So this is a proposal in the US, but also in the EU. Google and Facebook recently have been urged by the EU to label all of their AI generated content. You know, Google already said that they were going to be doing this on Google Images. And so I think that this is something that is going to be interesting to see.

6:27Apparently, the head of the EU, she met with Google Sundar Pichy, the CEO, a while ago, and she asked him if they had technology to detect fake news. And he said, yes, but we're developing our technologies further. I think in the EU, it would appear from the articles I've read that a lot of this debate revolves around Russia and essentially them wanting to label Russian mission misinformation campaigns. They're worried about with the Russian invasion of the Ukraine, that Russia is going to be pumping out a bunch of different, you know, fake news. I mean, you could understand, right, Russia pumps out, let's say, some images of Ukrainians doing some horrific war crimes.

7:08And, you know, all of a sudden people are saying, oh man, Ukraine's bad. Or, I mean, this could go either way. Ukraine could be doing it against Russia. Anyone can be doing it against anyone. And it's bad in all ways. But in any case, I think that this is the big issue that they're currently kind of centering it around there. I guess they're kind of in the middle of a, they're in the middle of a big war. And so this is kind of top of their mind for them there. And I do think it's interesting though, because they're they're talking to, you know, the CEO of Google, and he says they have tools. Now, we all know there are a handful of tools, I wouldn't put it past Google to have more powerful or more advanced tools, because it really kind of undermines their search if everyone can use AI to rank things that are not true or written by AI.

7:51So I would hope Google has some good advanced tools for detecting AI and making, you know, algorithmic adjustments based off of that. But at this same time, I am a little skeptical just because a lot of the AI detection tools I've used are known for putting forth false positives. And a lot of them are not super accurate. Like personally, when I want something written by AI, and I don't want it to be detected. And the reason I don't want it to be detected is because I just don't want Google to algorithmically derank an article. I'm, you know, trying to write it for SEO purposes, is I will grab that I will throw it through Quillbot, and I will have Quillbot rewrite it or whatever.

8:27And there's gonna be a number of tools. So like if all of a sudden I saw that all of the articles on my website started getting delisted from Google, search was also written by AI, I would just go find a new tool, I would just re-spin them. What's the difference between using that and having an actual human go and rewrite them or re-spin them? I mean, okay, obviously I understand the differences, but I mean, from Google's perspective, so I think it's going to be a cat and mouse game forever. Google says they have tools that detect AI, but I don't know how in depth those go. For example, I recently discovered and this is a great chat GPT use case, but I recently discovered that you can get chat GPT you can say hey act as a copywriter and copy my style of writing exactly you can feed it a few articles that you've written and then you say write an article on xyz topic and it can write an article in your style.

9:14So I think that you know chat GPT says they have some things embedded in there to help you detect that that was written by AI. I think the way that they like put words together they have algorithms so that you can detect that but uh what's to say i go train my own model or there's a anti-ai detection model that people have created um that you could train off of your own writing style so i think it gets kind of gray lines and blurry there um where like an article was written about a certain topic by ai but it was copying my style and maybe i gave it like uh five content pieces or sources or bits of information to write about and it's like it's it's tricky i'll definitely say it's tricky and detecting this is going to be really hard it's going to be cat and mouse forever so i don't know if this is really going to be something that can be very easily enforced and with the false positives right like if you go and say hey this was all of a sudden the ftc starts um you know fining the wall street journal because they say your article is written by ai and they're because it got a false positive on something it's going to be interesting it's going to be a little bit hairy and i think at the end of the day it's kind of like influencers on YouTube can go and shout out a product and talk about it and pretend it's not sponsored and probably nothing will happen.

10:24But on the off chance that someone finds out or there's some controversy around it, they find out it was a sponsored post and they didn't disclose it, they would obviously get fined. I are influencers really getting fined in reality. I highly doubt I haven't seen any big cases. I highly doubt this is a very common thing. I think it's gonna be the same with AI where it's kind of a self regulated thing. Like you should be a good person and denote if something was created by ai uh enforcing it will be pretty tricky so it's going to be interesting to see if there is actually any um any kind of credence to this if there's going to be actually application and if this is actually uh taken or implemented but conceptually i think it's a good idea um because transparency in my opinion is always the best option so this is going to be an interesting area to continue to follow