Contra Kahneman, your mind’s fast and intuitive “System One” is capable of logic

Numbers of the MindBy Christian Jarrett

Nobel-winning psychologist Daniel Kahneman’s international best-selling book is titled Thinking Fast and Slow in reference to the idea that we have two mental systems, one that makes fast, automatic decisions based on intuition, and a second that is slower, deliberate and logical. A further detail of this “dual processing theory” is that given time, and if we make the effort, the second system can step in and correct the intuitive illogical reasoning of the trigger-happy first system.

It’s an elegantly simple model supported by a huge number of studies, but it’s far from perfect. As demonstrated by a new paper in Cognition, it seems that contrary to Kahneman’s caricature of the mind, our intuitive System One is perfectly capable of logic, without the need for any help from System Two. Moreover, it’s actually rather rare for System Two to step in and overrule System One; more common is for System One to find the logical answer all by itself. 

Bence Bago and Wim De Neys tested hundreds of participants across four experiments that involved mental puzzles that call for logical thinking and reasoning. For example, participants were presented with questions that often provoke illogical answers if we allow our prejudices to overshadow basic probability theory, such as:

This study contains IT technicians and boxers.
Person ’L’ is strong.
There are 995 IT technicians and 5 boxers.
Is Person ‘L’ more likely to be?:
∘ An IT technician
∘ A boxer

The temptation is to run with stereotypes – to think “heuristically” – and say that L is more likely to be a boxer because boxers are typically stronger than technicians, even though the sheer volume (the “base rate”) of IT technicians means in fact it is logically more likely that L is a strong IT technician. For comparison, some of the questions were also phrased in such a way that thinking in stereotypes or in terms of base rates would both provide the correct answer – in the above example, for instance, the number of technicians and boxers would be reversed.

The participants also had to assess the logical soundness of syllogistic statements, such as:

All dogs have four legs
Puppies are dogs
Puppies have four legs
Does the conclusion follow logically? Yes or No.

Crucially, the researchers set conditions for answering these questions such that participants were initially forced to rely on intuitive System One thinking – for example, in one experiment there was an extremely tight deadline (just three seconds) for providing initial answers. In another experiment, as well as imposing the deadline, the researchers also engaged participants’ System Two with a secondary task that involved memorising a grid pattern, so that participants had no choice but to rely on their fast and intuitive System One. In all cases, after participants had given their fast System One answer, they were given a second chance to dwell on the problem for as long as they liked, to reflect deliberately and effortfully and, if they wanted, to change their answer – that is to engage in System Two thinking.

Across the experiments, about half of the time, participants provided the wrong initial answer and they stuck to that wrong answer, even when given more time to deliberate. This was as expected – most of us are not great at logic and we’re often too mentally lazy to engage System Two successfully. There were also some instances of participants’ fast System One thinking getting the answer wrong, and then being corrected by their System Two thinking. This scenario is consistent with Kahneman’s model but it was rare – occurring only ten per cent of the time. Far more common – around 30 per cent of the time – was the scenario where the participants actually answered logically straight way, after relying purely on their System One.

Bago and De Neys aren’t claiming that System One conducts logical thinking in the same way as System Two (it’s likely it does not), but their results do provide compelling evidence that System One is capable of logical intuitive thought, which seems to run counter to standard dual processing theory (in his book Thinking Fast and Slow, Kahneman states System One has “little understanding of logic and statistics”). Additionally, the model of thought described by Kahneman, in which it is common for System Two to take over from System One “when things get difficult”, seems to happen only rarely.

There are other interpretations of these findings – for example, some may question whether the researchers were successful in disabling System Two from contributing to participants’ initial answers. Another potential criticism might be that correct answers from System One were produced by chance, not through intuitive logic. But in fact, the distinct difference in participants’ intuitive answering styles to questions that were designed to provoke conflict between bias and logic and those questions that were not, showed that they were not simply guessing.

Bago and De Neys propose a new “hybrid model” to explain their findings, in which the automatic and intuitive System One is capable of both logical and heuristic thinking (with the two forms of intuition competing for dominance), and with later, logical System Two thinking being slower and optional. If one of System One’s intuitions is much more dominant than the other, it is more likely that it will be selected, and less likely that it will be changed later. In contrast, when the two intuitions – logical/heuristic – are more similar in dominance, then confidence will be lower and it is more likely that System Two will make a change. To some extent this account was borne out participants’ ratings of their confidence in the current experiments – if they expressed less confidence in their initial answer, they were more likely to spend more time reflecting on it later, and more likely to change it.

“To conclude,” the researchers write, “the present studies indicate that fast and automatic Type 1 processing can cue a correct logical response from the start of the reasoning process. This pattern of results lends credence to a model in which the relative strength of different types of intuitions determines reasoning performance.”

Fast logic?: Examining the time course assumption of dual process theory

Christian Jarrett (@Psych_Writer) is Editor of BPS Research Digest

9 thoughts on “Contra Kahneman, your mind’s fast and intuitive “System One” is capable of logic”

  1. It seems to me that both Kahneman and these researchers are overlooking the plasticity of the brain and the effects of learning and practice. Someone who spends a lot of time learning statistics will almost certainly come to logical answers using system one because they have more experience identifying the underlying facts of a problem. I think is less that system one is more logical than Kahneman thought but rather that learning can provide system one more resources for quick logic.

  2. Is the blogpost written using speech recognition technology? because i found some spelling errors that sound the same as the correct words.

  3. But isn’t part of Kahneman’s hypotheses that it is in fact simple decision making that is System 1 and more complex System 2 – by the very type of questions asked in the article I would expect System 2 to kick in immediately. The respondent recognises the complexity of the question and the auto-pilot bails even within 3 seconds.

    1. That’s a possibility but remember they also used a secondary task to occupy System Two, to prevent it from getting involved until later. Also, if System Two can get involved immediately, you wouldn’t expect such poor performance on the questions, or any evidence of corrections by System Two (which although rare, did occur).

  4. Does Kahneman’s theory say that System One makes decisions based on intuition?

    Thought he actually opposes the idea of “intuition” and sees it as a very well trained and fast thinking process: i.e. often many tasks that were previously performed by System Two, when done often enough to be automatic are then performed by System One.

    Were participants intelligent enough to deal with an easy task correctly by using System One alone (and not due to System One’s ability to deal with complex tasks, but because the task was easy enough for them to give a fast and obvious answer)?

Comments are closed.