Understanding the Decision Making Apparatus

The structure of the mind forms the basis of decision making. This paper cobbles together the highlights of previous posts that focused on the underlying machinery and how it fits together.


The human brain is distinguished by its connectedness and our ability to co-opt parts of it to our advantage. The connectedness is provided by overwhelming quantity of neurons and by a broadcasting system known as consciousness.  Our ability to co-opt comes from cultural systems that we have created such as writing and numbers that have allowed the specialization of  certain neurons.

Compared to electrical wiring hooked to a multi-test meter, neurons are slow and noisy. Electrical wiring is a million times faster–really.  Neural responses are noisy.  As an example, a radiologist may have tumor detecting neurons. These hypothetical tumor detectors will give noisy and variable responses. After one glance at a scan of a healthy lung, our hypothetical tumor detectors might fire 10 spikes per second. After a different glance at the same scan and under the same conditions, these neurons might fire 40 spikes per second.

And then to make things worse, neurons encode changes in stimulation–relative judgments and not absolute judgements. Absolute judgments on any dimension are much more difficult than relative judgments. This lies at the root of Ernst Weber’s 1834 observation that detectable increases in visual or auditory signal intensity are proportional to the starting value, i.e., need to be larger for larger starting values.

Connectedness allows better performance in spite of these shortcomings. The corpus callosum linking the left and right hemispheres of the brain has tremendous capacity and myelin sheaths that speed signals up to a hundred times faster. The huge fiber bundles under the frontal lobes allow massive increases in connectedness. The human brain has about 100 billion neurons. This quantity coupled with the differentiated hierarchy of neurons makes it quite likely that I have a neuron dedicated to Salma Hayek, etc.  Then human culture has boosted our abilities. In Reading in the Brain, Stanislas Dehaene introduces the idea of “neuronal recycling” whereby portions of our ventral visual system are turned over to reading and writing.  He says that after centuries of trial and error, writing systems evolved to a form adapted to our brain circuits.

Neuronal Architecture for Decision Making

The processing network. This vast distributed parallel machinery for decision making accumulates multiple sources of evidence, is a major part of what we call intuition, and moves us to action. It includes functionally specialized processors or modular subsystems that “encapsulate” information relevant to its function. Sometimes this entire processing network is just called System 1 or the default mode network or model free or non-conscious. In this network, processing is fast, automatic, and largely unconscious. The processing network has a dual architecture.

An example is the model of predictive coding in the visual cortex. At the lowest level, there is some pattern of energetic stimulation, derived by sensory receptors from ambient light patterns produced by the current visual scene. These signals are then processed via a multilevel cascade in which each level attempts to predict the activity at the level below it via backward connections. The backward connections allow the activity at one stage of the processing to return as another input at the previous stage. So long as this successfully predicts the lower level activity, all is well, and no further action needs to happen. But where there is a mismatch, “prediction error” occurs and the ensuing (error-indicating) activity is sent to the higher level. This automatically adjusts probabilistic representations at the higher level so that top-down predictions cancel prediction errors at the lower level yielding rapid perceptual inference. At the same time, prediction error is used to adjust the structure of the model so as to reduce any discrepancy next time around yielding slower timescale learning.

What is most distinctive about this duplex architectural proposal is that it depicts the forward (to the brain) flow of information as solely conveying error, and the backward flow thus achieves a balance between cancelling out and selective enhancement. This is made possible by the existence of “two functionally distinct sub populations,  encoding the conditional expectations of perceptual causes and the prediction error respectively”.  Superficial pyramidal cells are depicted as playing the role of error units, passing prediction error forward, while deep pyramidal cells play the role of representation units, passing predictions (made on the basis of a complex generative model) downward.

When a neuron or population is predicted by top-down inputs it will be much easier to drive than when it is not”. This is because the best overall fit between driving signal and expectations will often be found by inferring noise in the driving signal and thus recognizing a stimulus as, for example, the letter m say, in the context of the word “mother”, even though the same bare stimulus, presented out of context or in most other contexts, would have been a better fit with the letter n.  A unit normally responsive to the letter m might, under such circumstances, be successfully driven by an n-like stimulus.  (from post the Prediction Machine)

Research indicates that subliminal processing can be quite deep. The brain non-consciously recognizes the abstract identity of pictures, words and faces, the quantity attached to a number symbol, the fact that two words are related or synonymous, the emotional meaning of a word, or the reward value of a coin or an arbitrary symbol. Beyond even this, in chess experts, a brief non-conscious flash of a chessboard suffices to determine whether the king is in check. Transitive inferences can also be deployed non-consciously: after non-conscious exposure to arbitrary word pairs such as ‘winter-red’ and ‘red-computer’, word association effects generalize to non-adjacent pairs (‘winter-computer’), a transitive link.  All in all, these findings refute the idea that non-conscious processing stops at an early perceptual level: meaning and value can clearly be assigned non-consciously. There is also considerable evidence that attention can be deployed and enhance processing even if its target remains non-conscious.  Overall, findings support the view that virtually any cerebral processor may operate in a non-conscious mode.

In the paper “How do we convert a number into a finger trajectory?” Dotan and Dehaene find that sometimes processes that seem conscious start unconsciously. Study participants performed a number-to-position task on an iPad tablet computer, which allowed continuous measurement of the finger trajectory. On each trial, a two-digit number between 0 and 40 was shown on the iPad screen, and the participants dragged their finger from a fixed starting point at the bottom of the screen to a position along a number line that was at the top of the screen. The experiment software digitized the entire finger trajectory. Finger trajectories are a powerful measure because the finger position at a certain time during the trial tightly tracks the underlying cognitive operations.  If the finger movement was based on a conscious process one would predict a linear relationship with the finger moving straight to the number, but in actuality finger location was predicted by the logarithm of the two digit number so that the finger movements are clustered more in the middle toward 20 at the beginning. Dotan and Dehaene believe that since the log factor started early and then disappeared that the representation is automatic and not conscious. It is not based on a sequential conscious model but a non-linear non-conscious model.



















The global neuronal workspace (GNW). A control mechanism that provides the capacity to link decisions into tactics where the outputs of one decision become the inputs of the next. This capacity routs information, task setting, and task sequencing.  It provides flexibility at the cost of serial slow speed. This is known as the global neuronal workspace (GNW)  and provides what we know as consciousness. Bernie Baars conceived of the GNW as a blackboard upon which the light is shined  making the information globally available. It tends to be slow and deliberate and provides analysis and is sometimes called System 2, analytical system, deliberative system, model or conscious…Humans are clearly the most conscious of creatures and that the deliberative system is much newer than the automatic system. It consists of a distributed set of cortical neurons characterized by their ability to receive from and send back to homologous neurons in other cortical areas horizontal projections through long-range excitatory axons.  GNW neurons typically accumulate information through recurrent top–down/bottom–up loops, in a competitive manner such that a single representation eventually achieves a global conscious status. Because GNW neurons are broadly distributed, there is no single brain center where conscious information is gathered and dispatched but rather a brain-scale process of conscious synthesis achieved when multiple processors converge to a coherent metastable state. According to the GNW hypothesis, conscious access proceeds in two successive phases. In a first phase, lasting from ~100 to ~300 ms, the stimulus climbs up the cortical hierarchy of processors in a primarily bottom–up and non-conscious manner. In a second phase, if the stimulus is selected for its adequacy to current goals and attention state, it is amplified in a top–down manner and becomes maintained by sustained activity of a fraction of GNW neurons, the rest being inhibited. The entire workspace is globally interconnected in such a way that only one such conscious representation can be active at any given time. This all-or-none invasive property distinguishes it from the processing network in which, due to local patterns of  connections, several representations with different formats may coexist. (from post the Global Neuronal Workspace)


Consciousness allows us to deploy strategies that can make humans adaptable to quickly changing environments or to make really poor decisions. Interestingly, the mechanisms of conscious access seem to be comparable to those of other decisions, involving an accumulation of evidence toward a threshold with the difference that conscious perception corresponds to a global decision to engage many of the brain’s resources. GNW provides a working memory space that can be temporarily detached from incoming stimuli. Doing multidigit arithmetic in your head is an example for those of us who are not numerical savants. To multiply 30 by 47, I might multiply 30 by 40 and get 1200 and add it to 7 by 30 to get 1410. Like other conscious decisions, this competence involves serial chains of successive decisions each based on an accumulation of evidence that our working memories can handle and a forwarding of the result. Such serial conscious work causes a bottleneck by delaying processing of the most other things. This is typically known as attentional blink.

Those of us who have experienced propofol anesthesia for colonoscopies get a little more insight into consciousness. One “wakes up” usually after a half hour or less not feeling groggy and really surprised to know that the procedure is over. Propofol fragments neural activity. Neural activity remains locally organized, but globally disintegrated–possibly by alpha like rhythm in the prefontal cortex. Self-consciousness is a particular instance of conscious access where the conscious spotlight is oriented toward internal states.

Michio Kaku in his book…points out that we actually have two centers of consciousness separated by the corpus callosum. Looking at cases of people with severed corpus callosums, Dr Michael Gazzaniga has found two conscious selves, the verbal and interpretive left brain and the artistic and holistic right brain. The left hemisphere is dominant. In sharp contrast to its reputation as the deliberative and judicious system, the left hemisphere will abandon analysis and ignore inconsistencies for the purpose of giving us a smooth sense of a single “I”.  The left brain is more than willing to make preposterous excuses for this purpose. If you have an intact corpus callosum, it may not have to do this often, but it will do it if needed to maintain that single “I”.

Amazingly, the brain has the ability to host a ceaseless stream of all-or-none conscious episodes and this ability rests on the integrity of long distance cortico-cortical exchanges. Dehaene found a basic deficit of consciousness perception in schizophrenia.  Words had to be presented for a longer time before schizophrenics reported conscious seeing. “Schizophrenics’ main problem seems to lie in the global integration of incoming information into a coherent whole.” Dehaene suggests that schizophrenics have a “global loss of top-down connectivity. This loss impairs capacity for conscious monitoring, top-down attention, working memory, and decision making. Apparently in schizophrenics, the prediction machine is not making enough predictions. With reduced top down messages, sensory inputs are never explained and error messages remain triggering multiple explanations. Schizophrenics thus see the need for complicated explanations that can lead to the far fetched interpretations of their surroundings that may express themselves as bizarre hallucinations and delusions.

Dehaene is a believer in the Bayesian unconscious. “A strict logic governs the brain’s unconscious circuits–they appear ideally organized to perform statistically accurate inferences concerning our sensory inputs.” Both the unconscious and conscious systems seem to work in a linear fashion (Brunswik’s Lens Model), but the conscious system can redirect.


Dehaene suggests that consciousness allows us to share information with others and that leads to better decisions. Dehaene’s most interesting idea is that our social abilities allow us to make decisions together and that these are better decisions. Although one can argue that language is imperfect and that much of it is used to transmit trivia and gossip, Dehaene provides evidence that our conversations are more than tabloids. This is a point that needed to be made to me. I was tending to believe that there was almost a direct tradeoff between cognitive skills and social skills and even though that tradeoff was adaptive, maybe it was close. Dehaene puts forth the argument that two heads are better than one and that consciousness makes this possible (This is also directly in line with Scott Page’s: The Difference — How the Power of Diversity Creates Better Groups, post Diversity or Systematic Error).

Experiments indicate that the impact of later evidence is reduced when more evidence has been accrued but only for highly visible information. For difficult to perceive information, the information contributes equally to the decision.   Thus consciousness may play a role in decision making by biasing the accumulation of new evidence. Since Salma Hayek has her own neuron, new actresses, who do not, may lose out. (from post Toward a Culture of Neurons)


As one might expect with a system of neurons that encode only relative quantities and noisily, we have what Marcus calls a contextual memory as opposed to a postal code memory like a hard drive.  We pull things out of memory using contextual cues.  We give hints to our brain and see what it comes up with.  Unfortunately, some of the cues are actually related to the environment where the memory was created. So if you are trying to remember the genus and species of a plant that you looked up in the garden, then the garden is the place to go to remember it.

Context-dependent memory does have its strengths.  It prioritizes by bringing most quickly to mind things that are common, things that we’ve needed recently, and things that have previously been relevant in situations that are similar to our current circumstances.  It also seems to work in parallel, searching several threads at a time, to compensate for those rather slow neurons.  The weakness of context dependent memory is reliability.

Anchoring our memories in terms of context and cues, rather than specific pre-identified locations, leads to another problem: our memories often blur together.  This can lead to false memories and makes us vulnerable to being led by the most recent context.  Remembering the gist rather than the detail is bad enough, but we often think we remember the detail.  This is not good in the courtroom.

Every time we access a memory, it becomes “labile,” subject to change, and this seems to be true even for memories that seem especially important and firmly established.

There are things that can be done to improve our poor memories. Take for example, the ancient “method of loci”.  If you have a long list of words to remember you can associate each one with a specific room in a familiar large building: the first word with the vestibule, the second word with the living room, the third word with the dining room, the fourth with the kitchen, and so forth. There are also rhymes and repetitive rote memorization.  Still most of us realize the weaknesses of our memories and will even admit, while we are unlikely to admit any weakness in our judgment.

Marcus includes a quotation from Steven Pinker: “To a very great extent, our memories are ourselves.”  This might be a good reason to have not so good memories.

As Michio Kaku points out in his 2014 book, The Future of the Mind–The Scientific Quest To Understand, Enhance, And Empower The Mind, Doubleday, New York that over time there have been examples of people with special memory abilities or calculation abilities after sustaining damage to the left brain. The savant skill of photographic memory may be initiated by some sort of injury to the left brain leading to right brain compensation. This still does not explain how the right brain can perform these miraculous feats of memory. In 2012 a new study showed that the key to photographic memory may not be the ability of remarkable brain to learn, on the contrary, it may be their inability to forget.

In a study done by scientists at the Scripps Research Institute in Florida working with fruit flies, the fruit flies were exposed to different smells and were given positive reinforcement (with food) or negative reinforcement (with electric shocks). The scientists knew that the neurotransmitter dopamine was important to forming memories. To their surprise they found that dopamine actively regulates both the formation and the forgetting of new memories. Previously it was thought that forgetting might be simply the degradation of memories with time, which happens passively by itself over time. This new study  shows that forgetting is an active process requiring intervention by dopamine and that forgetting may be adaptive. The Men in Black  films figured this out years ago