This week on the Publisher Lab, Whitney Wright and Tyler Bishop continue the discussion on generative AI and also touch on Meta’s new ad-targeting transparency announcement.
You can listen to the podcast here and watch the video accompanying it here.
Google publishes new article about AI-generated content
Google Search Central Blog recently published an article about how “AI-generated content fits into our long-standing approach to show helpful content to people on Search.” The blog also states that Google will always try to reward “original, high-quality content,” with an emphasis on the quality of the content over how it is produced. However, it made a point to note that AI-generated content with the intent to manipulate search ranking is in violation of Google’s spam policies. However, how can one determine intent?
This all seems to go against what Google originally announced before releasing Bard, its own generative AI chatbot. Back in April, Search Engine Journal released an article that discussed Google’s John Mueller’s quote on auto-generated content and how it has always been against Google policies. Now, it seems Google is backtracking a bit.
In the blog, Google relayed that AI has been used to generate content for some time now—stock market prices, sports scores, weather forecasts, etc.—so using a chatbot to create content is all in the same vein.
Really, what Google may mean is that ‘original content’ is not necessarily content that comes from an individual source or blog, but that the content itself is not readily available in the way that people want and need it.
As an example that shows unoriginal content, Tyler discusses a site that was scanning local newspapers’ obituaries and creating an online database that updated every day. Google pinged the site because it it was technically ‘scraped content, even if the newspapers weren’t online, and so wasn’t original. Using generative AI to do something similar would be against Google policies.
The potential for generative AI to do harm
Last week, Microsoft, Google, Alibaba, Baidu, and Tencent all announced new AI capabilities last week. AI is exciting but there’s still room for caution about misinformation, cybersecurity, e-commerce fraud, and data privacy breaches.
For example, consider how Facebook’s Cambridge Analytica scandal sparked new data privacy debates or how self-replicating malware worms of the late 1990s and early 2000s prompted cybersecurity experts to reconsider standard operating procedures for protecting computer networks. Some similar event is bound to happen for AI as well.
Specifically, there are concerns with AI generating wrong answers, called “hallucinations,” and then someone having the power to spread that misinformation with an unlimited amount of “writers” on hand. Additionally, there is a concern with e-commerce and people using AI to generate fake reviews.
However, there isn’t necessarily much that can be done; it’s inevitable that because AI is spreading so fast, we will be unable to get ahead of it to stop something from happening. This is especially true when you think about how open-sourced versions of generative AI are being created right now; even if restrictions are built into generative AI, if it’s open source, you can easily go in and remove the restrictions. You could even use AI to generate workarounds.
For example, when we were doing machine learning at Ezoic, the machines would often test the bounds of the Sandbox even though there was a rule against breaking policy, just so it could fulfill its goal or purpose.
AI has a double reputation as of right now—that it will solve the world’s problems and be the world’s collapse, and it is likely that it will do neither. For example, the internet has brought many great things to the world but also a lot of unique challenges. It will be the same with AI.
However, one of the long-term issues that may arise is identity verification and encryption; hacking and breaking into things involves difficult math and problem-solving, but machines are really good at both. It wouldn’t be hard to trick someone into giving out information about their identity and then use that information to break in.
It’s similar to CAPTCHA tests; for example, when you’re clicking the boxes of everything that includes a stoplight, the machines are learning what a stoplight includes. This makes the machine smarter so that it can better understand when people are wrong or right, but also could make it smart enough to solve the puzzle itself.
Machines better understand humans more than what a human actually understands, meaning that it knows what a human expects and what humans see when we see a stoplight, but it doesn’t actually know what a stoplight is.
Overall, generative AI and the rise of AI is probably getting a little blown out of proportion, and people shouldn’t worry as much about it as they possibly are.
Meta touts more transparency with ad-targeting
Meta collaborated with privacy experts and stakeholders to get feedback on increasing transparency in the ads system and recently launched an updated version of “why am I seeing this ad?” This includes information on how they use machine learning to deliver ads and information on how your activity on and off the platforms informs the machine learning models. For example, liking a friend’s post or interacting with a sports website may inform the machine learning models to deliver the ads you see.
Additionally, the update includes new examples and illustrations explaining how the machine learning models connect various topics to show relevant ads.
In our opinion, the average consumer isn’t going to like this because they don’t like the idea of their data being collected (even if they know it’s happening in the background anyway) because of security issues.
Advertisers get all the heat for this data collection and then using it to serve ads, but really it’s the conglomerates collecting the data and hiding in its T&C’s how it’s going to be used and distributed to whomever they choose that people should place their attention. It’s likely giving people more transparency into that isn’t going to sit well with people.
This is probably Meta giving into the pressure of what it perceives to be a public problem but it’s really all in reaction to companies like Apple and Amazon, who live at a level below Facebook and collect data and information without having to get direct permission from users like Meta.
Apple is a big proponent of security and privacy and is a driving force behind Big Tech having to answer to growing privacy concerns. Apple builds its data collection verifications into every app and feature on their devices, allowing people to opt-in or out of it; however, outside apps and additions are always toggled ‘off’ by default and Apple ones are, by default, always toggled ‘on.’ Even if you toggled everything off, in their T&C’s, there are still conditions that you have to share some of your data. And what are you going to do–not use your iPhone?
The big question is—is Meta’s reputation salvageable?
It’s possible if it latches on to its popular and trusted parts—like Instagram—but as a brand with Facebook and its current leadership, it’s likely not going to make an about-face any time soon, no matter what they do to try to gain trust. Partially, it’s because they are behind on a public image standpoint, especially when you consider a company like Apple; they are masters of setting up a problem and then also providing the solution to that problem.
The saying is, ‘if you’re defending, you’re losing,’ and Meta is constantly defending itself and its actions.
It’s likely that these transparency changes Meta is making will have no effect on its contemporaries; other brands in thise realm are more concerned with how to change the system that we operate within so that it looks different when they’re still doing the same things they do today. For example, Google’s Privacy Sandbox and not tracking third-party cookies anymore—their solution is to just keep the data in Chrome. However, that means that Google will still have access to all of our data and give it to advertisers, just in a different way.
We’ll see how this all pans out for Meta.
