According to the zeitgeist, the AI age began when ChatGPT was released in November 2022. This narrative is false. Starting in 2015, one billion people began interacting with AI daily, when Google used powerful neural networks to redesign YouTube's recommender systems. These systems may still be AI's most influential form: they have fundamentally altered how we attend to the world, causing us to fixate on hyper-salient online content in ways that have upended our politics, emotions, wellbeing, and freedom. My NEH- and Templeton-funded work asks how good attention is possible in this age of distraction.

My philosophical work on this project––generated during my NEH project on the Dangers and Opportunities of Technology––argues against the dominant view of the harms of digital distractions. Within the philosophy, cognitive science, and the tech industry, there is an implicit consensus that digital distractions harm us by interfering with controlled, goal-directed attention. There is some truth to this: smartphones are hyper-salient in ways that interfere with our capacity to reflectively control attention. But this reflective control is what makes attention reasons-responsive, allowing us to direct attention to what we truly value. Developers who create hyper-salient apps therefore undermine our capacity to respond to reasons in a way that is morally harmful. Philosophers disagree about the precise moral harm: do apps make us less autonomous? Less free? Do they manipulate us? All of the above? Yet they agree that these harms stem from how technologies use salience to bypass our deliberative reasons and choices about where to attend.

I argue that the standard view mischaracterizes the threat of digital distractions in two ways. First, salience is reasons-responsive. Specifically, what we find salient depends on implicit learning mechanisms that track what we have reason to find relevant, given our values, emotions, and history of learning. This raises deep problems for views across the philosophical spectrum. Current arguments that digital distractions manipulate us depend on the strong assumption that all non-deliberative influences are manipulative, for example, whereas current arguments that digital distractions undermine our autonomy depend on the strong assumption that autonomous actions are necessarily based on our deliberative reasons. But there are strong general reasons to reject such strong assumptions. We argue instead that hyper-salient distractions are harmful not because salience is sub-rational, but rather because tech designers and recommender systems systematically undermine the reasons-responsiveness of salience. As such, digital distractions pose more dire problems than the literature recognizes. Hyper-salient digital distractions undermine (1) our capacity to regulate attention in response to deliberative reasons (as previous philosophers argue), (2) the synchronic capacity for salience to track our values, and (3) the diachronic capacity for implicit learning systems to update salience in a way that tracks our reasons to attend.

Second, hyper-salient technologies not only make us less attentive, but also change how we are distracted. Our minds used to wander during idle times such as riding a bus. Now mind-wandering has been replaced by the hyper-salient contents of our phones. Digital distractions replace mind-wandering, in other words, with salience-driven attention. This change is a problem because spontaneous forms of attention like mind-wandering help us explore alternative points of view. Under normal conditions, salience plays an important role in our mental lives: it directs attention to topics that we (in some sense) consider relevant, such as moral transgressions, dangers, and exciting opportunities. Yet salience can also be self-reinforcing, insofar as it can lead us to focus on what we think is relevant, which can lead us to ignore information that challenges our point of view. This is an especially poignant problem on social media platforms, which generate self-reinforcing epistemic bubbles. Mind-wandering and other forms of spontaneity (e.g. creative idea generation) function to escape these self-reinforcing attentional loops by exploring ideas that often turn out to be irrelevant but sometimes challenge our assumptions. Without spontaneity, relevance is a trap. Digital distractions have made the trap worse, not by reducing attention, but rather by making distraction less spontaneous.

My applied and scientific work on digital technology asks how we can find ways to harness––rather than eliminate––spontaneity in the digital age. Some solutions require no new technologies, but only a new philosophical approach to distraction. For example, companies like Serene currently use app blockers––software that will not allow you to use distracting apps––to promote “laser focus” on high-efficiency work. Those same app blockers could instead promote spontaneity, encouraging users to log off and let their minds wander. Or we could encourage people to utilize the “shower effect”, an effect I found in my empirical research, where moderately engaging activities like showering or walking, which induce mind-wandering to facilitate creativity. The problem is not that we lack the technology: over a century ago, Virginia Woolf generated ideas while walking around London. Rather, the problem is with our normative theories: people view spontaneity as frivolous idleness, rather than something valuable. My academic and public work attempts to foster this shift, encouraging people to recognize the value of spontaneity.

Another kind of strategy utilizes the recommendation algorithms that helped to create the spontaneity deficit. Recommendation algorithms ultimately use AI to maximize whatever variables we tell them to maximize. Often, that is watch time. Yet we could instead re-train these algorithms to instead recommend content that promotes spontaneity forms of attention. To do so, my long-term goal is to combine my philosophically-grounded methods to measure spontaneity with AI-based detectors to train AI to recognize and harness spontaneity. Two studies take the first step towards this long-term goal, showing that it is possible to use AI to identify spontaneity. One study (“Detection of freely moving thoughts using SVM and EEG signals”, IEEE) reliably detects whether someone’s mind is wandering, on the basis of electrophysiological data (i.e. their brain activity). Another paper-in-preparation finds that we can train AI to identify where one’s experiences fall on the Thinking Grid, on the basis of eye gaze data. This is significant, because it could allow our methods to be scalable: rather than have to train recommendation algorithms on the basis of thousands of reports, we could instead use eye gaze data. Next, we plan to train a Thinking Grid detector from facial expression, using computer vision algorithms pre-trained to detect fine-grained emotions such as relaxation, interest, and boredom.