What I've Been Reading
AI, 2026, and Why It Matters That We Look Carefully
We have evolved, for good reason, to make quick judgments on fragmentary evidence. In an evolutionary sense, when confronted with a new beast or shaken by a rolling rumble, the decision we made in the next couple seconds might determine whether we lived or died. Mediated by our amygdala, the fight-or-flight control center in the most ancient part of our brains, “think quick” is a nearly irresistible impulse. Its continued presence in our lives is reflected in the Nobel Prize awarded to Dan Kahneman whose work on decision-making under uncertainty was later popularized in Thinking Fast and Slow (2011), and by the extraordinary robustness of the “thin slice” research tradition. The term “thin slices” was coined in 1992 to encompass judgments made after “short observations of expressive behavior (under 5 min)”.1 Think of it as tests of the adage “first impressions matter.” As one meta-analysis notes, “thin slices reliably and validly measure behavior across various domains.” How thin might those slices be? Two Princeton researchers discovered that we make stable judgments about a stranger’s attractiveness, likeability, trustworthiness, competence and aggressiveness.
In 100 milliseconds2.
Giving people more time to study a stranger seemed to increase their confidence in their tenth of a second call but didn’t change it.
That pattern seems to hold in our encounters with large language models, which generate personae that are simultaneously deeply human and completely inhuman. Rushing to answer Samuel Morse’s question, “what hath God wrought?” we see, we label, we move on, secured by the comfort of having assigned a label.
And yet there is a case for stepping back, double-checking our footing before proceeding much further. In pursuit of that goal, I’d like to raise four possibilities – “Big things” – that might help frame our thinking, followed by a celebration of the coolest bits of AI-related research that I’ve read in 2026, aka “Little things.”
In preface to all that, I should explain my place in the discussion. I routinely work with three collaborators: my wife Chip who reads and reacts to everything, Claude Sonnet, who reviews drafts for structure, coherence and lapses, and Perplexity, who has an unrivaled ability to quickly locate and summarize academic research. Claude and Perplexity routinely cross-check one another’s work (Claude thinks Perplexity has the writing style of a tax form, Perplexity thinks that Claude mostly locates research by stumbling over it.) I have been teaching about communication and emerging technologies since 1999, when email was The Big Thing threatening to disrupt Corporate America3 and Mark Zuckerberg was a high school sophomore. Since then, I’ve taught through, and about, the rise of social media (who now remembers the great Friendster / MySpace wars?), blogs, vlogs, wikis, tablets, the mobile internet (initiated by iPhones in 2007), the Invasion of the Chromebooks (a $36 billion misadventure) and now AI. While the words have changed, the melody remains largely the same.
Or so I’ll argue over the next several weeks, where I’ll attempt to share four big things and ten little ones with you. Let’s see how it goes.
1 Ambady, N., & Rosenthal, R., “Thin slices of expressive behavior as predictors of interpersonal consequences: A meta-analysis.” Psychological Bulletin (1992): 111(2), 256–274. https://doi.org/10.1037/0033-2909.111.2.256. Ambady and Rosenthal have been cited over 3000 times as of June 2026.
2 Willis and Todorov, “First Impressions: Making Up Your Mind After a 100-Ms Exposure to a Face,” Psychological Science, July 2006. The mothership is Ambady and Rosenthal, “Thin Slices of Expressive behavior as predictors of interpersonal consequences,” Psychological Bulletin (1992) which has been cited over 3000 times. Their meta-analysis found that “thin slice” (under five minute) exposures to a person lead to judgments as accurate as those reached after hours. A striking example of such research: the 15 character ratings made by experimental subjects seeing 2- to 5-second silent clips of a teacher are strongly correlated to the ratings offered by students after 14 weeks with the same professor. “Snap Judgments Work!” Harvard Magazine, 7/2001.
3 For example, Sarbaugh-Thompson, “Electronic mail and organization communication: Does saying ‘hi’ really matter?” Organizational Science, Nov-Dec. 1998.