.
On Human Nature(s) 2: System 1 vs. System 2
Marc Green
Overview of Human Nature
I start by considering the human condition from the broadest perspective. The highest level view of a system is its architecture, that determines which functions the system may perform easily and which functions it can only perform with difficulty or not at all. In short, the architecture places constraints on what the system can do, so it is a critical component of any conceptual model. The human physical architectural constraints are fairly obvious, but we also have mental architectural constraints. Three are most relevant. First, humans have access to only two sources of information for decision and action. One is the information that enters the senses, most importantly the light that enters the eyes. This sensory input is termed "bottom-up" information because it flows from world up through the lower parts of the nervous system (the sensors) to the more central part (the brain).
The other source of information is the contents of the brain. Some of this content consists of previous experiences stored in memory. In addition, we also have some built-in, innate predispositions. The two sources together constitute "top-down" information that flows from the higher level of the nervous system downward. It interprets the sensory input, makes decisions and schedules behavior. This may or may not involve conscious information processing.
Second, both the quantity and quality of the bottom-up sensory and top-down information are limited. Elsewhere, I have already discussed some visual system's constraints, such as small visual field size and lack intrinsic 3-D depth information. The brain is also limited in both storage capacity and speed of processing. Brain size is limited by energy consumption—brains are avaricious users of energy. The human brain is likely a tradeoff between processing power and the ability to find and consume calories.
Third, our architecture has a limited processing capacity. Consider the problem that we face every day: we exist in a world that has far more information and complexity than our limited sensory and cognitive apparatus can handle at once. The mismatch is most obvious in time-limited situations because it may not be possible to switch among information sources and process them separately or deeply. The mismatch is perhaps the single most important determinant of human cognitive architecture because:
- Much of our mental evolution has been an attempt to find workarounds for the mismatch to expand our abilities. We have evolved many to arrive at satisfactory decision-making despite the limited resources that we have. It is important to understand why these workarounds exist and why they are necessary in assessing human decision-making and behavior;
- These kludges usually work but like all such short-cuts, they are not perfect. They can meet circumstances where they fail. This has an important consequence: normal adaptive decision-making and behavior can still lead to bad outcomes. But the bad outcome alone is not proof that the decision was bad. Last, I use the term "kludge" to emphasize that there is grand no design and that they are simply a collection of haphazardly evolved mechanisms that band-aid over a human limitation;
- This is just the way we are built. We can no more process all the information in a scene or pay attention 100% of the time than we can lift a ten-ton weight. And there is nothing to be gained by penalizing people for their innate limitations or wishing that it were otherwise. Remember Skinner's Law; and
- Humans have a strong bias toward "cognitive ease," information that is more easily digested as it lowers the drain on mental resources making us better able to handle new sensory input. This is discussed below.
Dual Process Theory
The starting point for understanding the architecture of human nature is its duality. To handle our limitations, evolution has given us two very different ways of processing information and interacting with the world. Each is appropriate for a specific type of circumstance. The duality of human nature is a thread that runs through most psychological disciplines and much of Western (and probably Eastern) philosophy and thought going back at least to the ancient Greeks. Modern researchers have contrasted the two natures using the various dichotomies shown in Table 1.
Table 1. The distinctions between the two human natures suggested by various authors. The entries marked with "*" are my additions and not found in many, if any, other sources. Some differences are not explained here, but see Green (2024) for more detail.
| System 1 | System 2 |
|---|---|
| Fast | Slow |
| Effortless | Effortful |
| Intuitive | Logical |
| Holistic (experiential) | Analytical (reductionist) |
| Automatic/unintentional | Controlled/intentional |
| Generally unaware | Often aware |
| High capacity | Low capacity |
| Intuition & Emotion | Facts & rules |
| Reptilian | Mammalian |
| Errors (slips, lapses)* | Mistakes, violations* |
| Resistant to stress | Degraded by stress |
| Dorsal/magnocellular | Ventral/parvocellular |
| Ambient | Focal |
| Ecological* | Constructivist* |
| Direct perception* | Indirect perception* |
Not all authors would agree with every listed difference between systems or use similar language to describe them, but all agree that humans have two natures. Many are familiar with the notion of the System 1 vs. System 2 dichotomy, Nobel Prize winner Daniel Kahneman's packaging of dual process theory, so I will use this terminology1. In short, the major distinction between the two systems is that System 2 is slow, conscious, and effortful while system 1 is fast, automatic, and effortless. The two systems have several other differences as I'll explain. Last, System 1 is a mashup of different mechanisms. It might be useful to simplify by slicing off part of System 1 and labeling it "System 0" as it is the oldest and most reptilian part of human nature, the part that is perhaps most deeply ingrained.
Disclaimers
It is often easier to discuss a topic by presenting its polar opposite, yin vs. yang, Hertz vs. Avis, System 1 vs. System 2. On close inspection, however, dichotomies begin to fray at the edges. In the case of System 1 and System 2, it is probably wrong to think of them as completely separate processes. They are perhaps best viewed as being analogous to rods and cones. Each represents a different mechanism that operates in somewhat different conditions by trading off one optimization (contrast detection) against another (light sensitivity.)2 The result is a visual system that can operate over a much wider range than would be possible with either alone. However, their functioning also overlaps somewhat (mesopic vision). The System 1/System 2 tradeoffs are suggested in the Table 1 dichotomies, the most obvious being speed, accuracy, and effort.
Further, System 1 and System 2 are also not places in the head. (Aside from the reptilian/mammalian distinction which may be taken either literally, or perhaps better, metaphorically.) They are merely descriptions of behavioral mechanisms. I would urge the reader to completely ignore neuroscience when thinking about System 1, System 2, attention, emotion, or most other psychological processes. It adds nothing to the explanation but merely glosses the explanation the patina of "real science" compared to just talking about behavior. For example, there has been much recent research saying that the amygdala is the neural basis of threat perception (e.g., Phan, Wager, Taylor, & Liberzon, 2002). OK, but how does us help explain understand someone's behavior better than simply saying that the person was under threat? Moreover, despite what any researcher claims, the connection between known physiology and any halfway complex behavior is tenuous at best3. The putative connection could change tomorrow. I've seen it happen many times. It is better to stick to behavior.
However, I sometimes break this rule when talking about the visual system, perhaps because it is my core competency and/or because it is the best understood part of the brain. There is no doubt that the visual system has multiple pathways that go from the eye to the brain, each with its own specialized characteristics. In Table 1, the dorsal/ventral, magnocellular/parvocellular, and ambient/focal distinctions refer to contrasting visual pathways, their functional role in guiding behavior, and the information that they convey. In addition, section 4 discusses "predictive coding," the reciprocity between expectation and coding in the visual and other sensory systems. I cannot explain this without at least some reference to physiology.
System 2—Controlled
Now let's dig down into some details. I'll start with controlled System 2 information processing because it is easier to understand. The label "2" reflects its later evolution compared to System 1. It is sometimes termed "mammalian" because it is a recent evolutionary advance that many believe is restricted to humans. (But I wouldn't try telling that to my cats.) System 2 gives rise to decisions and behavior that are controlled in the sense that they seem volitional and largely conscious. The mode of information processing is logical (in the broadest sense of the term), analytical, sequential, rational, and rule-based. You look at one thing, then the next and then the next. If you are solving a puzzle, looking for a parking space or deciding what to order off the menu, you are likely in an analytical mode—or so it would appear.
A good example of logical System 2 conscious, analytical processing is the "feature vector" approach to decision-making: break the problem down into a set of variables and assign a value to each. Choose the option with the highest score. Imagine you go to the car dealership and see two potential vehicles that you might purchase. Which as the lower mileage? Does either have any damage? What are the prices? What is their gas consumption? You could just pick the car that, as they say, checks all the most favorable boxes. You could further even weight each variable and then add up the scores to pick the better car based on a weighted sum (a linear regression analysis). Either way, you are aware of how the decision was made.
People are generally consciously aware of being in this analytical mode. It is intentional and has "cognitive penetrability." That is, decision-makers are consciously aware of their mental processes. They can introspect about what they are doing and why. This is why System 2 is relatively easy to understand and why people tend to overestimate its importance in determining decision and behavior.
System 2 supposedly dominates when humans encounter complex and/or novel situations, putting them into an information-seeking mode. However, humans prefer not to use it because of its downsides. It consumes extensive mental resource leaving little for other things, like noticing that saber-tooth tiger about to pounce on you. It is slow and effortful. It is a recent evolutionary advance, and we often struggle with it. It is contrary to the "cognitive ease" that humans prefer. As a result, we avoid using System 2 whenever possible4.
In fact, it is doubtful that many people make important decisions using the feature vector or other seemingly rational analytical methods even though they may believe themselves doing so. The feature vector approach often results in what I like to call the Pittsburgh Paradox. Survey after survey scores Pittsburgh as the best, or at least one of the best, places to live. They arrive at this conclusion by enumerating a list of features and then scoring each. Despite the high score that Pittsburgh always receives, I doubt that many people would see it as the best possible place to reside. I also doubt than anyone would make a life-decision such as choosing a spouse this way using a similar approach.5
My old real estate agent nicely articulated the System 1 grounding of many supposedly logical System 2 decisions. She often liked to say that "Buyers are liars." She said this because in her experience house buyers gave her a list of required features for their future home, e.g., number of bedrooms and bathrooms, square footage, etc. They were seemingly going to make a System 2 decision. But the buyers then routinely end up with a house that has few if any of these features, instead choosing a house because it "felt like home," "was cozy," etc. Even apparently analytical System 2 decisions are readily influenced by the "affect heuristic," which is described below with other System 1 heuristics.
System 2 reasoning also so often fails because we live in a world better described as "complex" rather than merely "complicated." Although the two words are often treated as synonyms in everyday language, they have different technical meanings. Complicated systems can be decomposed into separable components/variables that can be reassembled into the whole by following a set of "syntax" or "rewrite" rules. The easiest systems to analyze are linear, i.e., have a context-free grammar. They are readily amenable to the kind of logical, reductionist analysis that System 2 performs6. Unfortunately, many, if not most, real-world situations are better described as "complex" because they consist of components/variables whose interactions are tightly bound in nonlinear relationships. The components/variables cannot be readily isolated and examined in individually so logical analysis cannot be performed, at least in any reasonable time. System 1 is better in this environment because it operates holistically and does not engage in any logical analysis, which would be slow and possibly futile. The best approach to complex systems is management, not analysis. This is what System 2 attempts to do.7
System 1—Automatic
Someone has estimated that only 2 percent of our behavior is generated by this conscious System 28. So where does the other 98 percent of information processing originate? The answer is System 1, which is fast, automatic, and seemingly intuitive and effortless. By intuitive, I don't mean anything mystical. I mean that it operates by direct perceptual recognition. We see a situation and pick up cues that are automatically interpreted and that tell us what to do with little or no conscious thought. A more detailed examination of System 1 is presented later and comprises the bulk of the discussion.
Comparison of System 1 and System 2
Some simple perceptual problem-solving exercises demonstrate the differences between System 1 and System 2. The first task is to find a vertical red bar in a display. Look at the empty frame below. The task is to click on it and then to find the vertical red bar in the array that appears. Ready: click in frame.
This was very easy. The red bar just appeared automatically and immediately without the need to exert effort. Now look at the next empty frame below. The task is again to click on the empty frame and then to find the vertical red bar in the array that appears. Ready: click in frame.
This task exemplifies System 2 information processing. It was much more difficult, required effort, and was slow. Some introspection will reveal that it required fixating each item in turn and moving attention sequentially through the display until finding the vertical red bar. The sequential nature of the task is the reason that this called "serial search." In contrast, the first frame required only System 1, so it was intuitive and effortless. It involved holistically processing the entire array at a glance so that it was a "parallel search." There was no need to decompose the array into its components.
This first example was purely sensory. The next demo shows the differences between System 1 and System 2 in a somewhat different task that involves learning and memory as well as perception. Look at the frame below. The task is to click on the frame and to find the dog in the display that appears. Ready: click in frame.
Keep searching for a few moments if the dog doesn't appear. Some people will see the dog quickly, some slowly and many not at all. For those who don't see the dog click on the next frame below, which shows its outline. Ready: Click in frame.
Now look back up at the previous frame. The dog becomes apparent. In this case, learning has transformed the task from System 2 to System 1: from sequential, effortful search to intuitive, holistic recognition. Lastly, look at the previous frame and try not to see the dog. It's impossible—once you recognize the dog, you can't make it go away. This highlights some key points:
First, recognition did not depend on noting any single blob or small group of blobs. Instead, it was a holistic processing of the entire image;
Second, most people would prefer to perform the System 1 task because it is automatic and effortless. Humans will almost always attempt to transfer a task from System 2 to System 1.
Third, imagine that someone just walked into the room and only saw the first frame with no outline and didn't recognize the dog. Viewers who saw the dog might blame the newbie for not seeing the "obvious," which of course is only obvious in hindsight. This is what often happens during mishap investigations. Those who have studied the scene and know what has happened often fail to understand how much of their perception, interpretation, and causal attribution has been influence by already knowing the facts.
Fourth, it is easy to view this the urge toward cognitive ease in a negative light. How often have you heard said that you should think about what you are doing and focus on the task? However, the desire for cognitive ease is not a matter of laziness. Rather, it reflects the reality that mental resources are limited and should be conserved to handle novel or unexpected circumstances. There is a tendency, often in 20/20 hindsight, to scornfully suggest that people should always be logical (i.e., analytical), think before they act, and operate in System 2 mode. This is short-sighted, as many have said (e.g., Malcolm Gladwell in Blink)9. Whitehead (1811) summarized this issue best:
It is a profoundly erroneous truism that we should cultivate the habit of thinking of what we are doing. The precise opposite is the case. Civilization advances by extending the number of important operations that we can perform without thinking about them.
This is accepted doctrine in most design fields, especially in human factors. The goal is to make the object or environment easy and intuitive to use. Good design requires no manual because merely looking that the object/environment tells the user what to do. The goal of training in many cases is also to make the behavior automatic. Despite the obvious benefits of such "user-centered design" and training, humans are routinely scolded for not thinking about what they are doing (using System 2), especially when something unexpectedly goes wrong. This view is popular in mishap investigations because System 1 works its magic out of sight, and the great benefits that it presents are taken for granted as normality—until one of its relatively rare failures occurs10.
Next: System 1 Kludges —Affordances, Heuristics and Biases, and Schemata
Endnotes
1Kahneman originated neither the notion of dual processes nor the System 1/System 2 terminology.
2This dual system design to extend human abilities is a constant thread in human evolution. See for example, the "sustained-transient dichotomy."
3The studies on threat also have debatable ecological validity because the methods used to induce threat are seldom highly threatening. This is also major limitation in most behavioral studies of threat behavior.
4However, there are exceptions. When arousal level is too low, humans crave the stimulation of using System 2 to solve puzzles. This is why crossword and jigsaw puzzles exist.
5Although Elizabeth Barrett Browning did write "How do I love thee? Let me count the ways."
6If this sounds familiar, it is how science works, at least in theory. This also reveals the limitation of scientific reductionism.
7Worse than complex, the system could be chaotic. See the "Cynefin" framework of system analysis.
8Some say that the number is 5 percent. I can't remember who said this. It doesn't matter because it is irrelevant. Never take numbers like this too seriously.
9In fact, paying focal attention while performing a System 1 task can impair performance. A good example is found in the Stephen Potter's book, The Theory and Practice of Gamesmanship or The Art of Winning Games without Actually Cheating (1947). Potter describes how to mess up an opponent's golf game. Just as he tees up, say "That's a wonderful pause you have at the top of your backswing." The opponent then can't stop focussing on the pause while hitting and has his rhythm completely disrupted. The book is a classic of applied psychology.
10Many internet sites claim that the average adult makes 35,000 decisions a day. I tried to track down the source of this number but was unable. The web pages just cited another web page, which cited another web page... in an endless daisy chain, although some simply refer to unnamed "experts." This exemplifies a new species of knowledge that might be called "Internet science." Whatever the exact number or how "decision" was defined," humans do make many decisions per day, but are not aware of it because most are automatic and outside of awareness (System 1) and most produce the desired result.
