Updated: July 24, 2026
Zooming out, AI may not yield novel consequences in principle; it may change the distribution of benefits between individuals who use it more efficiently.
AI reduces information search costs, but the Internet and typography had a similar effect of search cost reduction.
AI increases scale, scope, and automation of execution, but software and the printing press had similar improvement effects.
Recently, I’ve been watching a Stanford class on AI where an invited guest from Databricks, made a point that caught my attention: AI drops the cost of searching for information dramatically.
The next day, I came across a piece in one of the Ukrainian technology news channels citing a study that, in the AI era, people are becoming much less likely to say that they simply do not know something—only about 3% now give an “I don’t know” response.
Lastly, during a conversation with an older family member, they casually mentioned that they had learned about the topic we were discussing only after the Internet appeared. Before the Internet, they simply had no easy way to access this kind of information.
The sequence of these three observations made me think: Do we really observe something fundamentally unique with AI's impact on information access? Or are some of the changes that we currently attribute to AI simply another step in a much longer technological trajectory—one in which access to information becomes progressively cheaper, knowledge becomes easier to retrieve, and the boundary between what we need to know and what we can simply look up keeps shifting?
Below, I outline the three points that may (or may not) convince you that AI is (not) that different.
Before the Internet, access to information was really costly. After the Internet appeared, people were able to go online and learn things that previously would have required much more time, effort, or access to specialized sources. With the development of fast broadband Internet and search engines, the need to memorize information also decreased. Instead, individuals became increasingly confident that they could simply Google something when they needed to know it.
This likely prompted a similar decline in “I don’t know” responses long before AI. If people know that an answer is only a search away, they are less likely to treat not knowing something at the moment as a hard constraint.
If this logic holds—which I believe it does—then AI should not be fundamentally different from previous waves of digitization of knowledge and the continuous lowering of the cost of its access. AI may push search costs much lower, but the underlying mechanism itself is not new.
Before LLMs as a the most popular type of AI, executing on information—writing code, drafting essays, generating designs, or developing new ideas—was costly. It required not only access to knowledge, but also the skills and time needed to turn that knowledge into an output.
However, this hurdle was even larger before the Internet. Accessing examples, comparing approaches, learning new techniques, or finding inspiration required much more effort. And before that, technologies such as the printing press dramatically lowered the cost of reproducing and distributing knowledge, making it easier for more people to build on what others had already created.
The mechanism, therefore, is not entirely new. Each technological wave has reduced the cost of turning existing knowledge into new output. What seems different about AI is the scale and scope of execution that suddenly become available to an individual. One person can now write, code, analyze, design, and iterate across domains that previously required much more time—or several different specialists.
AI may therefore represent less of a new mechanism and more of a large expansion in how much execution capacity an individual can access.
Before the conveyor belt, assembling industrial goods such as cars and heavy machinery was cumbersome, slow, and heavily dependent on skilled human labor. The conveyor belt changed this production process by simplifying some human tasks, specializing others, and replacing parts of human execution with machines.
AI may be doing something similar for intelligent work. The difference from the previous argument is that AI does not only make it cheaper for a person to execute on information—it increasingly makes it possible to automate parts of that execution altogether. Writing a standard email, producing basic code, summarizing a document, or generating a first design may no longer require the same amount of human involvement.
But the economic mechanism is familiar. As the cost of automation falls, firms automate more tasks, including what Acemoglu calls “so-so automation”: tasks for which technology may not be dramatically better than a human but is sufficiently cheaper or more convenient to replace some human work. AI may therefore move the boundary of automation far into cognitive work, without fundamentally changing the economics of automation itself.
After drawing these parallels between AI and previous “disruptive” technologies, one common pattern becomes noticeable: many of their consequences repeat themselves. There is always a distribution of people and firms in how effectively they access, adopt, and use the available technology.
What changes is the competitive positioning of individual economic agents within that distribution. Getting access to a technology that was previously unavailable can improve an individual’s competitive position. Using the same technology more intensively or more effectively than others can improve it further. The arrival of AI follows the same logic. Even if the underlying economic effects are not entirely new, AI can substantially reshuffle who benefits from them and who gains a relative advantage.
In this sense, technological progress may change the tools available to everyone, while competition increasingly depends on who learns to use those tools better.
Given the large intervals between the technological paradigms it is typical of humanity to forget the prior ways of doing things. Zooming out to capture several epochs back is not an easy everyday task either. This is also the reason why we get overly excited by the arrival of the new technology, such as AI, as we see the promise to disrupt the status quo. But many changes are less abrupt than they first appear. Rather than creating an entirely new trajectory, a technological shock often changes the slope of an existing one. AI may therefore be less of a step-function break and more of an acceleration—or, in some cases, a reversal—of trends that were already underway.
Why is such analysis of the differences and similarities between the two waves of technology important? For recent graduates and early-career professionals, it helps to reflect on their competitive positioning relative to the existing, already familiar set of technologies—and to extrapolate how they need to use the new technology to maintain or change that position.
From a research perspective, such comparison is equally necessary. Distinguishing what is genuinely new from what is simply an extension of an existing mechanism helps pinpoint the novelty more precisely and, at large, moves scientific thinking forward.
Stanford Online. (2026, July 13). Stanford MS&E435 economics of the AI supercycle | Spring 2026 | Infrastructure, enterprise AI, SaaS [Video]. YouTube. https://youtu.be/sRvrXL83N-c
Marcoccia, C., Quattrociocchi, W., & Capraro, V. (2026). AI advice suppresses people's willingness to say “I don't know”, even when the advice is wrong and accuracy is incentivized [Preprint]. arXiv. https://doi.org/10.48550/arXiv.2607.13562
Acemoglu, D., & Restrepo, P. (2019). Automation and new tasks: How technology displaces and reinstates labor. Journal of Economic Perspectives, 33(2), 3–30. https://doi.org/10.1257/jep.33.2.3
Please cite this article as:
Petryk, M. (2026, July 24). AI is Different. Not. MariiaPetryk.com. https://www.mariiapetryk.com/blog/post-33