Copyright Newsweek
A central theme of this series has undoubtedly been how humans and AI systems can − or should − interact in a harmonious, symbiotic way. And award-winning author and neuro-ophthalmologist, Mithu Storoni thinks she has the answer: AI systems should reduce problems that are beyond human cognitive ability to the sweet spot mix of “80 percent easy, 20 percent hard” that results in a cognitive “flow state.” She boldly conjectures that the net effect will be that “AI is our tool to achieve an intrinsically motivated state of productive bliss.” William Shakespeare famously wrote that “the eyes are the window to your soul,” and it turns out that this is more than romantic conjecture, there is an underlying neurological basis for this observation. Storoni always sensed there was more to this statement than met the eye (pardon the pun), and this inkling drove her entry into what many would regard as a rather esoteric area of research. As she describes it, “I did my Ph.D. in neuro-ophthalmology, and I worked in a very niche field at the time…looking at pupil dynamics. And I realized that the pupil is an extraordinary organ because it gives you a window into not only into the autonomic nervous system but also into this little region in the brain that's blue in color and called the locus coeruleus.” The locus coeruleus (LC) or “blue dot” (a literal translation of the Latin) network is the subject of Storoni’s latest book, Hyperefficient, that details how this body, and the associated network, controls a set of different “gears” that our brains use to manage our cognitive and physical lives. Storoni identifies the essential gear types and their respective roles, and how each gear is engaged in response to different stimuli, as well as the natural duration of the gear-modes required to be maximally productive, or “hyper-efficient.” Importantly, she also sees a critical role and opportunity for AI in assisting humans to achieve this efficiency, by optimizing our problem-solving abilities. In order to understand the nature of these gears and their interaction, it is important to have a basic appreciation of the operation of the relevant neurobiological systems. The autonomic nervous system (ANS) is comprised of the sympathetic system that controls our fight or flight behavior and the parasympathetic system that controls the complementary rest and relaxation behavior. One of the primary controls of the ANS is the LC network. The LC is the primary producer of the neurotransmitter norepinephrine (NEP) and it controls the distribution of NEP throughout the brain to different regions that support a wide variety of functions such as: arousal and the sleep-wake cycle; attention and memory; behavioral and cognitive flexibility; creativity; personality, behavioral inhibition; psychological stress; cognitive control and decision making. Storoni explains that the NEP release is uniform across the network, but different regions of the brain contain different receptor types that respond differently to concentrations of NEP, with some becoming active at low concentrations, some at intermediate and some at high concentrations. In addition, they have different gain functions, so the rate at which the associated neurons fire varies significantly, allowing control to be coherently shifted from one region of the brain to another as NEP is released by the blue dot. Simply put, the blue dot network is the critical “modulator” of the ANS that enables the switching of the human mode of operation by using the ANS to tune the sympathetic (acceleration) and parasympathetic (deceleration) nervous systems to meet the demands and stimuli of the current physical environment. As she explains in Hyperefficient, the different gears are “associated with three levels of tonic and phasic LC activity” (tonic being background, constant rate pre-synaptic activity, and phasic being the short-bursts of activity seen after activation), that can be summarized as follows: Gear 1 (Slow Mode): The mode used for cognitive rest and recovery, as well as what we would call “daydreaming.” In a photographic analogy, focus is panoramic and soft, allowing for wider exploration of thought-space, facilitating spontaneous insights as well as the complementary function of mental "slate cleaning" for avenues of cognitive exploration that are not pursued. Gear 2 (Medium Mode): The core mode for cognitive work as it supports focused concentration, problem-solving and learning. The prefrontal cortex is fully engaged, providing sharp, adaptable focus with two sub-modes: Low-energy Gear 2: Used for spontaneous creative exploration; it allows attention to broaden and shift around, but retains the ability to sharpen and narrow focus on specific ideas or thoughts. High-energy Gear 2: Used for complex conceptual learning and divergent thinking as it supports facile creation of neural pathways and fluid intelligent reasoning. Gear 3 (Fast Mode): The mode used for rapid response to critical situations. Prioritizes speed over accuracy by enhancing automated abilities and suppressing refined thought and judgment. Storoni explains that the movement between these mental gears occurs when “the locus coeruleus changes the rate at which it fires. And by changing the rate at which it fires, there is a gradual increase in norepinephrine, and that results in these different states.” And she correlates the origin of the firing rate change with a change in the perception of uncertainty, saying, “We are constantly trying to reduce uncertainty in the world around us.” In fact, she observes that “a very broad definition of intelligence is using information to dynamically navigate a dynamic, complex space and draw conclusions,” and thereby minimize uncertainty. In other words, our intelligence is designed as an uncertainty-minimization function, and this is instantiated as a set of mental states (gears) controlled by the blue dot, LC network. Interestingly, Storoni argues that uncertainty is perceived both by the instantaneous information load and the perception of the passage of time – that is, when a wealth of new sensory information and data about the current environment arrives over a short period of time, the resultant uncertainty triggers the LC, which increases the release of NEP to allow the shifting of focus and activity (a gear change) to reduce this uncertainty. There is a clear connection to Kahneman’s System 1 and 2 modes of cognition, with Gear 3 operating based on learned and inherited heuristics that are automatically applied to allow the most rapid response and minimization of energy consumption, and Gear 2 corresponding to the deeper thinking and analysis that is characteristic of System 2. In Kahneman’s conception, effort is required to shift from (effortless) System 1 to (effortful) System 2, but Storoni argues that the high energy Gear 2 mode feels can actually feel effortless when it is instigated by intrinsic motivation; what is popularly described as the “flow state.” The concept of the flow state was conceived of by management psychologist Mihaly Csikszentmihalyi who described it as the state of human contentment and productivity that arises from "being completely involved in an activity for its own sake. The ego falls away. Time flies. Every action, movement, and thought follows inevitably from the previous one…your whole being is involved, and you're using your skills to the utmost." He further argued that to achieve a flow state, a critical balance must be struck between the challenge of the task and the skill of the person involved; if the task is too easy, flow cannot occur as it requires that both the skill level and the challenge level must be significant. In contrast, if the skill level is low and the task is not challenging, apathy results, or if the task is too challenging, it is just frustrating. Storoni argues that the key to achieving a cognitive flow state is consistent improvement in terms of learning progress that is matched with work that is sufficiently challenging to require focus. This has the net effect of creating pleasure from effort (with the pleasure mediated by the second order connections from the LC to the dopamine network) and explains the near-euphoria that is often experienced during flow, with more work accomplished with less effort. She is also is quick to point out that effort and reward have, until recent times, always been connected. For example, the effort applied in hunting or gathering or similar physical work (and the mental planning associated with it) was matched with the reward in the form of larger quantities of sustaining food or materials. However, with the mechanization of the industrial revolution, goods were produced by machines requiring little human physical effort, and negligible mental engagement. In short, we have disconnected effort from reward for the sake of production efficiency. Yet we, as humans, derive pleasure from rewarded sensory experience and physical effort. As Mihaly Csikszentmihalyi puts it, “Humans feel happy through working.” With the advent of so-called “knowledge work” as a primary form of human endeavor, the effort and the reward have become almost completely disconnected, with a few keystrokes giving rise to manifold change or, conversely, the arduous process of writing documents (or code) having little or no impact. Storoni summarizes this state of affairs succinctly, “We divorced the body from the mind with assembly lines, then reversed this by divorcing the mind from the body with knowledge work. This divorce has robbed the mind of the plethora of sensorimotor signals from the body. The net result is that the physical and diurnal nudges to the brain state are no longer natural….it's a confusing environment and your brain naturally thinks there is too much uncertainty in this environment” which it will attempt to minimize, but without the clues and feedback mechanisms required. So how will human-AI interaction impact this reality? Storoni sees that “in the near future and moving forward, tasks will require really analytical problem solving, and original creativity. And all of those functions are optimized in a brain state that's intrinsically motivated, because intrinsic motivation holds your brain in the right configuration.” On one hand, AI clearly has the potential to exacerbate the effort-reward disconnect by further reducing the effort required, creating more apathy towards mundane work tasks. But Storoni sees the potential for the exact opposite – AI can make possible tasks that were previously too hard, or impossible, increasing the effective skill of the individual and the resultant sense of accomplishment. In her view, AI will enhance the potential for achieving the nirvana of intrinsic motivation. Storoni observes that with higher loads of information you need to operate in higher gear(s), and sees intrinsic motivation existing on the boundary of Gears 2 and 3, describing the process as follows: “You're [in Gear 2 and] about to hit gear three, but then you pull down, and so you have the maximal possible level of alertness coupled with perfect cognitive control.” She points to recent research that actually quantifies the requirement for achieving intrinsic motivation or flow state, “one of the ways in which we seem to derive it…is by applying effort in a way that what we are working on is around a 20 percent difficulty level.” She continues, “Whenever the brain meets a problem that it perceives as being impossible to surmount, it immediately leaves the creative zone. So, we need to achieve the difficulty level where your problems are no longer impossible, and we can actually shift people into the right state. There are many tasks which are very mundane and are akin to the old-fashioned assembly line in terms of monotony…we get very bored, and they put us into Gear 1, because we just drift off; we don't have the drive to get into Gear 2.” She also argues that across the overall set of tasks we need to perform in any role, there are many that are beyond our capacity, but “if we bring AI in there, we can lower the load or difficulty of these tasks, we can also fragment the task into smaller tasks, and let AI do smaller tasks, allowing us to re-imagine tasks to be at the 20 percent difficulty level.” So, her defining argument is that AI can move tasks that were previously impossible or too taxing into the solvable and intrinsically motivated human “flow” realm. Storoni has another key takeaway derived from the preceding observations: we need to change our way of working to allow the human part of this emerging equation to be further optimized. “We have created [the] template for work modeled on the assembly line, based on muscle action. But muscle and mind are very different, because the muscle works while it works, rests when it stops. The brain…can often work while it rests, and rest while it works. So, this fundamental difference between muscle and the brain has not been respected in the way we've designed our knowledge work template.” In essence, the production line concept now underpins our rhythms, when the brain operates on a different timescale of bursts of 90-minute cycles, followed by rest and recovery periods. We are all aware of the circadian rhythm – a 24-hour biorhythm that is also mediated by norepinephrine levels, rising in the morning for alertness and falling at night for rest. Embedded in this cycle is what is known as the basic rest-activity cycle (BRAC), which is most apparent during the classic REM sleep cycle. However, there is an emerging body of evidence that this ultradian cycle is actually present throughout the day, and is the reason that many human activities last between 60-120 minutes, e.g. sports games, meeting or class times and entertainment (films, concerts, theatre performances). Moreover, brain activity is higher during the first half of the cycle which corresponds to feeling maximally alert and focused; during the last 20 minutes brain activity slows, and we begin to daydream and feel mental fatigue. We are capable of overriding this 90-minute natural cycle, but only by engaging the fight-or-flight response or using artificial stimulants such as caffeine. In Storoni’s gear system, we naturally descend into Gear 1 mode towards the end of the 90-minute cycle; if we then force our brain to continue to work it will immediately move into a Gear 3 mode, where only rote tasks are possible, with little or no original thought or creativity. She explains, “The longer your brain spends on a task, we see that the pathways through which information crosses your brain actually become less and less efficient, which may explain why the task becomes more and more difficult.” The brain has two complementary networks that manage this cycle: the cognitive control network (CCN) of the prefrontal cortex hands off to the Default Mode Network (DMN) to manage descent into the lower gear. The prevailing wisdom is this latter phase of the 90-minute cycle is critical as it allows neurological recovery, e.g. to regenerate neurotransmitters and remove byproducts, as well as eliminate unproductive lines of mental exploration. So, Storoni proposes that, in order to maximize our Gear 2 (System 2) cognitive output, we need to recognize the criticality of this 90-minute cycle. She consequently argues that work should be comprised of many shorter, lower-effort explorations, combined with a few more intensive attempts during the day. Indeed, there is evidence that this so-called Levy walk behavior describes the optimal cortical strategy when exploring a rich mental space. These observations suggest another potential role for AI: it will not only accelerate the rate at which humans can explore a topic − as well as the scale of exploration − by reducing the mental load of the task to the 80:20 sweet spot but, in turn, this will also allow the analysis to complete within the minimum number of 90-minute cycles. What about inter-human interactions? The Media richness theory developed by Daft and Lengel observes that all communication channels vary in their richness, with richness defined as "the ability to change understanding within a time interval.” A key component of this theory is the idea of “social presence,” which refers to the degree to which a medium is perceived to convey the actual presence of the communicating participants. The more complex or uncertain the message, the richer the communications channel needs to be, because multiple sub-channels such as gestures and other non-verbal cues must be used. This multimodal and multisensory complexity has been shown to result in five to 15 times more effort required to communicate complex ideas when a simple communications channel such as email is used rather than in-person communication. Storoni also underscores the critical role of interpersonal or collective synchrony: “We use a whole wealth of nonverbal cues…our pupils open up or close, depending on the synchrony. The synchronization happens once there's trust between two individuals, but really it shows a deeper level of interpersonal synchrony, which is a synchronization of the autonomic nervous system in both individuals.” She continues, "Virtual interactions do not allow for this…you can barely discern the pupils, so that level of synchrony is not taking place. The pupil of the eye is a window to your autonomic nervous system…so it is the source of information that's acting as part of the whole process of interpersonal synchrony. When you're surrounded by people who are calmer, or when you're with someone who is very calm, you automatically feel slightly calmer, and the opposite happens if you're next to someone who's very tense, you will feel slightly more tense, and that is really your autonomic nervous system mirroring the other person.” Storoni sees that this interpersonal interaction goes beyond just the facilitation of information exchange, and that shared synchronous effort matters – a topic we have previously covered in this series, “We are learning ever more about things like synchronization, neural synchrony and the power of influence and achieving shared goals…these factors have contributed to creating this whole new world that we have to navigate, not just individually, but also collectively. We know that at times of intense emotional reaction, if you experience that with someone else, there is a level of synchrony that is not present if you were to experience that alone.” In the most extreme form, this can be thought of as the “revelry” that is associated with the Greek God, Dionysus. Returning to the idea of the brain’s blue dot network working to minimize uncertainty, it is clear that in-person interactions are an effective way to engage in collective analyses and creative explorations, with less individual uncertainty. So, Storoni proposes a new way of working: “When we're doing a task that involves focusing, then it's very helpful to have one target you can anchor your focus onto and not be distracted from. Whereas, when you are looking for new ideas or when you're brainstorming ideas and you hit a wall and you want to hear a different opinion, shared effort is often better.” She highlights that “it's a delicate interplay…creativity goes through different phases; you first have a phase of collecting and gathering information, then you have a phase of filtering that information, and then you have a phase of, okay, this is it: I'm going to sit down and converge my focus and zoom in onto this. With focus, you want a solid target, a closed room, silence, whereas with more creative tasks, like brainstorming, you want something a little more stimulating, where you're forced to run down avenues you haven't encountered before.” Famously, this was the mode of operation of one of the most successful innovation organizations in history, Bell Labs, about which I can claim some insight. The scientists were afforded a lab and an office, both of which were private spaces (although doors were left open when interaction was welcomed), but the lunchroom was the common daily meeting place where ideas were exchanged between a diversity of people and disciplines. Moreover, the building was built around long corridors that acted as common conduits that everyone traversed, resulting in numerous additional serendipitous engagements. The modern equivalent is perhaps the combination of remote versus in-person working, which can provide a similar complement of private, focused cognitive work and collaborative, creative exploration. Working remotely lends itself to a self-enforced 90-minute mental cycle, which can be more difficult to arrange in complex multi-person environments with multiple asynchronous and competing agendas and distractions. Putting this all together, Mithu Storoni has an interesting way of summarizing the current state of affairs and the potential for AI: “Your mind is the new assembly line worker: it processes information to manufacture knowledge products. But the AI era may force a change for the good by removing the assembly line template for work.” Specifically, she argues that AI will allow the natural 90-minute cycle of the blue dot network to be better adhered to, with maximum use of the Gear 2 mode of operation, by concomitant reduction of the difficulty of complex problems to the 80:20 sweet spot, and optimal use of in-person and remote ways of working. She also highlights that humans will be needed to make AI-operated systems “anti-fragile,” given the lack of real world understanding of the models in these systems: a persistent theme throughout this series. Storoni concludes that “so far, humans were forced to mimic and out-compete machines, but now that machines have their own realm, it's time to return to being human again,” effectively creating a personal and professional utopia or “bliss.” In other words, hyper-efficiency is now a realizable goal, and a much more grounded one than the mythical quest for super-intelligence or AGI.
