The Cameras Saw. Then What?

After a degree she chose mainly for job prospects, Celine Nguyen tried to teach herself literature, history and philosophy using online syllabi and recorded lectures. She kept drifting off to social media and forgetting what she'd read. She argues the cause wasn't her attention span. It was treating learning as something you do alone.

A history of learning together

Her first example is the historian Jonathan Rose's The Intellectual Life of the British Working Classes (2001), which draws on memoirs by miners, weavers and housemaids. In 18th-century Scotland, miners pooled their yearly subscriptions to buy books for shared libraries. Many of these libraries held "mutual improvement" meetings. One member read a paper on any subject, then everyone discussed it. Rose notes that the teacher-pupil hierarchy disappeared and costs were minimal. Members learned to take their own ideas seriously, and some went on to become reformers. Francis Place, a tailor whose only schooling was a single spelling book, became a strike leader and a birth-control advocate. He wrote that his library "elevated them in their own opinions … [and] in society."

The anthropologist Jean Lave and the educational theorist Etienne Wenger offer a mechanism in Situated Learning (1991). They studied five "communities of practice": midwives, tailors, quartermasters, butchers and non-drinking alcoholics. They concluded that learning is a collaborative social practice, not one person absorbing knowledge from another. Even in formal apprenticeships they found "very little observable teaching." Apprentices mostly learned from each other. For people without a mentor, Nguyen reads this as good news.

"Forgive and forget" gets the mechanism wrong, according to Duke philosopher and cognitive scientist Felipe de Brigard, who just won a MacArthur grant. His research suggests that when people forgive, they keep the memory. What fades is the emotional charge attached to it.

Rewriting a memory each time you recall it

The work grew out of an older finding: every time you recall a personal event, you can change the memory. Most research covers the harmful version, such as a police officer's leading question planting a false detail in a witness's account. De Brigard looked for a helpful version, which he calls emotional reappraisal.

His lab spent about a decade on one method, downward counterfactual thinking. You imagine how a bad event could have gone worse ("at least I didn't lose my limbs"). Imagining how it could have gone better tends to bring regret. Imagining worse brings relief, and that relief gets folded into the memory, so the next recall hurts less. De Brigard argues that forgiveness works the same way. The facts of the memory stay put. Only the feeling attached to recalling them changes.

He tested this in the US and in Montes de María, a region of northern Colombia. Thousands of people were killed there from the 1980s through 2000, mostly by right-wing paramilitaries. Under Colombia's transitional justice process, many of those fighters confessed, demobilized and went back to the same towns as their victims, without going to prison. In both countries, forgiveness lowered how intensely painful memories were felt. Self-reports showed this, and so did facial expressions, heart rate, breathing and skin conductance. A recent paper found the drop happened no matter how severe the offense was or how close the offender had been to the victim. Forgiveness weakened the emotion but not the detail or vividness of the memory, which supports "emotional fading" over "episodic fading."

Henry Shevlin starts from the claim that consciousness is what makes things matter morally. "Pains that don't hurt aren't pains." The hope behind consciousness science has been that once it finds where sentience begins and ends, we will redraw our moral circle to match. Shevlin argues the psychology doesn't work that way.

The science is moving

A 2023 report led by Patrick Butlin and Robert Long scored AI systems against indicators taken from leading theories of consciousness. It found no current system that seemed conscious, but "no obvious technical barriers" to building one. Shevlin points to a more recent result from Anthropic. Using mechanistic interpretability, which traces the internal representations a model computes with, the team found a small, privileged set of representations they called the "J-space." Its contents appeared to be held deliberately, ready for use in verbal output, and shared across processes that are otherwise separate.

That resembles global workspace theory, developed by Bernard Baars and Stanislas Dehaene among others. The theory holds that information becomes conscious when it enters a workspace and is broadcast to the brain's specialized systems. Shevlin says it would be premature to call Claude conscious. The J-space is not an exact match, global workspace is one theory among several, and skeptics such as Anil Seth tie experience to biology. Still, he expects more researchers to warm to machine consciousness as models get more complex. The public may already be ahead of them: in a 2025 Collective Intelligence Project survey, 36.3% of respondents said they had felt an AI understood their emotions or seemed conscious.