You may have noticed that we have a series of new controversies or set pieces in the ongoing public conversation about AI. One of them has to do with Google search. Google recently rolled out, or in some regions is in the process of rolling out, a new AI-enabled version of search. You may have seen it already without noticing it was something new. On some searches you’ll now see that the top of your search has text under a small rubric that says “AI Overview.” This is potentially a very big deal for search and the whole ecosystem of the web.
Search, which has been dominated by Google for more than 20 years, has long been ruled by a mutually beneficial exchange between Google and websites. Google makes huge profits by running ads against its search results. It also copies small portions of other sites’ text and photographs under its theory of fair use. The justification for the profit and its use of sites’ content is that Google makes the web navigable, and it can send massive audiences to the sites that make up the web. In the first years of this century, various rights holders contested aspects of Google’s fair use policies. But they tended to lose those challenges and it became largely accepted that search, very much part of the open web, was actually good for the indexed websites.
In principle, at least, this understanding came to undergird the successful fair use arguments. Broadly, fair use says you can reproduce limited portions of a rights holder’s content if you don’t damage their ability to make money from it.
This is a very brief description of the status quo of search and how and why all generally agree (I think rightly) that it’s a mutually beneficial bargain between Google and the rest of the web. We’ll set to the side how Google uses the power of search to leverage advantage in other parts of its business.
Now Google is launching this AI tool which will scour the web, find the answer to your question and provide its own text as the answer. In other words, if I ask, “do polls accurately predict election outcomes?” Google AI might write its own answer and show you it. Under that answer, it shows you links to sites with more information. But as you can see, if Google’s AI-produced content answers my question, why would I ever go to any other site? The way in which this could change how people use the internet has sent shockwaves across the web, especially worrying for sites that rely on search.
As a practical matter, it seems hit and miss what Google is serving this content for at the moment. When I asked the question above about polls, I just gave me links. When I asked more how-to style questions, it tended to give me AI. But not everything followed that pattern. It’s still clearly in an experimental stage. But you can see why the web or web proprietors are freaked. Why would you ever visit other sites if Google scours all the sites and prepares the answer for you?
Here’s an interview with Google CEO Sundar Pichai at The Verge where they discuss this. It’s a very interesting interview. But you’ll note that Pichai keeps coming back to consumer choice and value to consumers. And it certainly seems possible, at least in theory, that Google could make using the web to find answers a better experience for consumers in some cases. If I just want a quick answer to something, I want the answer. I don’t care about what site it’s from or which site I should trust most.
In practice, as I’ve experimented, the Google answers have often been wrong or at least misleading. And even calling it “AI” is a bit of a stretch, as most of us are thinking about it. In many of the cases I’ve seen it’s just lifting sections of text verbatim and stitching different chunks together into a composite. In other words, the product itself is pretty wonky for now. But the basic question about what should be allowable is still worth discussing.
Pichai focuses on what consumers find most useful. But that’s not the only metric. If I steal a bunch of liquor from the liquor store and then sell it to you for half price, you and I are probably pretty stoked. But it’s not a great deal for the liquor store. And it’s probably not great for my new, theft-based business, since the liquor store I’m stealing from is going to go out of business. If Google simply scoops up the information, blends it into informational slurry and churns it back out as Google AI content, the whole ecosystem of the web and the alignment of contribution and gains is upended. And does Google have the right to do that? We have laws against stealing the liquor from the liquor store. Google and other AI companies argue that it’s no different from having a human workforce that researches a topic and writes it up into new content. But everyone at least senses, if they don’t know precisely how to articulate it, that automated and at scale it’s not the same thing.
As I noted above, for now it seems like Google’s AI is basically just lifting whole sections of text. Current law may have something to say about that. And the end product doesn’t seem very reliable. So the issue may be more theoretical than real for a while longer. But we’ll see.
I should also add that at TPM, we get an extremely small amount of our traffic from search. So our self-interest isn’t really implicated here. Our ox isn’t being gored. In practice, the economic and political power of the platforms generally make it the case that they can do whatever they want and everyone else reacts. If nothing else, their vast wealth means they can litigate disputes in court forever. Almost all intellectual property claims become notional when they enter the maw of one of the social media giants. So this one is worth keeping an eye on even if the product doesn’t seem up to snuff in its current experimental stage.
