-
CERN, “The Birth of the Web,” accessed April 10, 2026, https://home.web.cern.ch/science/computing/birth-web.
-
John McCarthy, “From Here to Human-Level AI,” April 19, 1998, https://www-formal.stanford.edu/jmc/human/human.html.
-
John McCarthy, “Concepts of Logical AI,” May 30, 1999, https://www-formal.stanford.edu/jmc/concepts-ai/concepts-ai.html.
-
“Behind Deep Blue,” Slashdot, November 18, 2002, https://science.slashdot.org/story/02/11/18/1810222/behind-deep-blue.
-
Distill, “About,” accessed April 10, 2026, https://distill.pub/about/.
-
“A Neural Network Playground,” TensorFlow, accessed April 10, 2026, https://playground.tensorflow.org/.
-
David Ha and Jürgen Schmidhuber, “World Models,” accessed April 10, 2026, https://worldmodels.github.io/.
-
LessWrong, accessed April 6, 2026, https://www.lesswrong.com/.
-
Eliezer Yudkowsky, “Recursive Self-Improvement,” LessWrong, December 1, 2008, https://www.lesswrong.com/posts/JBadX7rwdcRFzGuju/recursive-self-improvement.
-
Sam Altman (@sama), “it is possible at some point he will deserve the nobel peace prize for this–I continue to think short timelines and slow takeoff is likely, but certainly he got many of us interested in AGI, helped deepmind get funded at a time when AGI was extremely outside the overton window, was right about many technical issues, etc,” X, February 3, 2023, https://x.com/sama/status/1621621725791404032.
-
Mozilla Foundation, “YouTube Regrets,” accessed April 8, 2026, https://www.mozillafoundation.org/en/youtube/findings/.
-
John McCarthy, “Professor John McCarthy,” accessed April 10, 2026, https://jmc.stanford.edu/.
-
John McCarthy, “John McCarthy’s Home Page,” accessed April 10, 2026, https://www-formal.stanford.edu/jmc/.
Here are concrete edits you can plug in to strengthen the use of evidence without changing the overall argument. I’m giving them as either replacement sentences or sentences to insert after a specific line of reasoning in your draft. They’re based on your uploaded essay.
In the McCarthy / early expert-web section, the easiest improvement is to make the evidence do a little more interpretive work.
After this sentence:
“Read together, these hypertext pages framed AI capability less as an imminent transformation than as a constrained and unfinished research project.”
Add:
“McCarthy’s phrasing is important here: terms such as ‘difficult’ and ‘have not been solved’ do not merely describe technical obstacles, but actively position AI as a field defined by unresolved problems rather than public triumph. In the context of the early web, where a researcher’s homepage functioned as a relatively stable extension of scholarly authority, such language made caution itself appear as a credible mode of expertise.”
That helps because it turns the quotation into analysis rather than leaving it as support that sits on the surface.
A second sentence you could add near the end of that paragraph is:
“What these pages offered readers was not a spectacle of machine performance, but a durable explanatory frame through which performance could be interpreted in advance as partial, technical, and incomplete.”
That sharpens the connection between evidence and your larger claim about web form.
In the Slashdot / Deep Blue paragraph, you can make the forum evidence feel less like a single anecdote and more like proof of a discursive pattern.
After this sentence:
“What mattered here was not only disagreement over a single machine, but the way the forum’s structure turned AI capability into a matter of public argument rather than expert declaration.”
Add:
“The evidentiary value of the thread lies precisely in its back-and-forth structure: capability was not presented as a settled fact, but produced through visible disagreement over definitions, criteria, and boundaries. In that sense, the forum preserved not just opinions about Deep Blue, but a public record of how web users collectively negotiated what counted as intelligence at all.”
You could also slightly revise your concluding sentence of that paragraph. Replace:
“The significance of forums, then, lay less in changing that conclusion than in shifting authority outward, creating participatory spaces in which capability could be interpreted, disputed, and bounded through public argument.”
With:
“The significance of forums, then, lay less in changing the conclusion that AI remained narrow than in shifting interpretive authority outward, creating participatory spaces in which capability claims were tested, disputed, and bounded through public argument visible to all readers.”
That version makes the evidence feel a little more specific and grounded in the forum form.
In the Distill / TensorFlow Playground section, this is where you can most obviously raise the evidence score, because right now the examples are good but move a bit quickly.
After this sentence:
“Rather than simply telling readers that neural networks learned by adjusting internal parameters, it let them alter features of a model directly in the browser, adding or removing layers, changing the dataset, and watching the decision boundary shift in response.”
Add:
“That detail matters because it changed the status of the claim being made. The user was no longer asked only to trust an expert description of machine learning, but was invited to witness the model’s behavior as something dynamic, responsive, and experimentally legible in real time. Evidence of capability therefore became less purely textual and more procedural: understanding came through manipulation, observation, and repeatable interaction.”
That addition directly strengthens your “use of evidence” because it explains exactly what the evidence proves.
You can also revise this sentence:
“In that format, machine learning did not appear only as a technical claim to be accepted on expert authority; it appeared as something users could test, manipulate, and observe for themselves.”
Into:
“In that format, machine learning did not appear only as a technical claim to be accepted on expert authority; it appeared instead as a process users could test, manipulate, and observe for themselves, which made capability more immediate and therefore more rhetorically persuasive.”
That makes the evidence serve the thesis more explicitly.
A good additional sentence for the Distill portion would be:
“Distill’s self-description as a venue for explanations ‘native to the Web’ is especially revealing, because it shows that the form of presentation was itself part of the intellectual claim: clarity no longer meant prose alone, but interactivity, animation, and visual inspection inside the browser.”
In the LessWrong paragraph, the main problem is that the evidence is slightly too concentrated in Yudkowsky and Altman. Even if you do not add a brand-new source, you can make the existing evidence work harder by emphasizing the platform and its recurring discourse.
After this sentence:
“Terms and ideas such as ‘recursive self-improvement,’ ‘intelligence explosion,’ ‘alignment,’ and existential risk did not simply appear there as isolated phrases; they circulated through serial posts, extended comment threads, and ongoing cross-reference, giving participants a shared language for interpreting new developments in AI.”
Add:
“That accumulation is itself evidence of a distinct web form at work. LessWrong did not simply host opinions about AI; it archived, linked, and recursively amplified them, allowing concepts to gain force through repetition, commentary, and reuse across many posts rather than through a single definitive publication.”
That helps show the evidence is not just “Yudkowsky said X,” but “the platform sustained a discourse.”
You could also revise this sentence:
“On LessWrong, capability was therefore described less as bounded task performance than as a potentially open-ended and strategically significant force.”
To:
“On LessWrong, capability was therefore described less as bounded task performance than as a potentially open-ended and strategically significant force, a shift visible not only in individual posts but in the site’s cumulative vocabulary of acceleration, alignment, and existential risk.”
That gives the paragraph a broader evidentiary base even before you add another source.
In the 2020 platform infrastructure paragraph, you can make Mozilla’s example feel more like evidence of a broader shift in discourse.
After this sentence:
“The significance of such evidence lay not only in showing that recommendation systems were powerful, but in showing that public discussion of AI capability had become inseparable from questions of platform responsibility, user vulnerability, and the governance of everyday online experience.”
Add:
“Here, capability was no longer persuasive because it appeared spectacular, as in a milestone victory or research demo, but because users could describe its effects in ordinary life: what they were shown, where they were led, and how difficult those systems were to see or resist. The evidence of AI’s power thus became experiential and infrastructural at once.”
That sentence makes the evidence richer by clarifying what kind of evidence Mozilla is actually providing.
You could also strengthen your transition into that paragraph. Replace:
“As a result, online debate increasingly shifted from what AI could do in principle to what algorithmic systems were already doing in practice: steering attention, shaping judgment, filtering visibility, and sometimes pushing users toward material they had not actively sought out.”
With:
“As a result, online debate increasingly shifted from what AI could do in principle to what algorithmic systems were already doing in practice, as users encountered recommendation, ranking, and personalization not as abstract innovations but as forces shaping attention, judgment, and visibility in everyday browsing.”
That reads a little more evidence-driven and less abstract.
The fastest high-impact revision, though, would be these five insertions:
For McCarthy:
“McCarthy’s phrasing is important here: terms such as ‘difficult’ and ‘have not been solved’ do not merely describe technical obstacles, but actively position AI as a field defined by unresolved problems rather than public triumph.”
For Slashdot:
“The evidentiary value of the thread lies precisely in its back-and-forth structure: capability was not presented as a settled fact, but produced through visible disagreement over definitions, criteria, and boundaries.”
For TensorFlow Playground:
“Evidence of capability therefore became less purely textual and more procedural: understanding came through manipulation, observation, and repeatable interaction.”
For LessWrong:
“LessWrong did not simply host opinions about AI; it archived, linked, and recursively amplified them, allowing concepts to gain force through repetition, commentary, and reuse across many posts rather than through a single definitive publication.”
For Mozilla / 2020:
“The evidence of AI’s power thus became experiential and infrastructural at once.”
If you want, I can next turn these into a clean revised version of the actual paragraphs so you can paste them directly into the essay.