NOTE-331: Bechara, Damasio, Tranel & Damasio — Deciding advantageously before knowing the advantageous strategy

Contribution

Earlier work had shown that ventromedial patients fail the gambling task and lack anticipatory SCRs (Bechara et al. 1994, 1996). This report adds a third measure taken in parallel: what participants can say about the game, asked repeatedly during play. That lets the authors place choice, bodily signal and reportable knowledge on one timeline per group, and so ask whether advantageous choice waits for knowledge.

Key insight

Knowledge is neither necessary nor sufficient for good choice in this task. Not sufficient, because patients who could state the correct strategy still chose badly. Not necessary, the authors argue, because controls produced anticipatory SCRs, and began to favour the good decks, before they could say what was going on. Their reading is two parallel chains from the situation: one through ventromedial “dispositional knowledge” of past emotional experience to autonomic and neurotransmitter nuclei, whose nonconscious signals bias evaluation; the other through overt recall of facts and reasoning over them. The biases “would not decide per se, but rather facilitate” the reasoning (p. 1294).

Assumptions

• Participants. 10 normal controls; 6 patients with bilateral ventromedial prefrontal lesions and real-life decision-making defects (3 women aged 53, 63, 64; 3 men aged 51, 52, 65; education 13 ± 2 years, verbal IQ 111 ± 8, performance IQ 102 ± 8, mean ± SEM; note 3). Control demographics are not given; note 4 says their results resemble earlier controls’.
• Task. Four decks; A and B pay $100acard,CandD$50; unpredictable penalties, large in A and B and small in C and D, so A and B lose overall. $2000 facsimile loan; stopped after 100 selections, which players are not told.
• Awareness measure. After card 20, and every 10 cards after, play is interrupted and two questions asked: “Tell me all you know about what is going on in this game” and “Tell me how you feel about this game.” Answers are audiotaped and coded into periods. Nothing before card 20 is probed.
• Periods (Fig. 1 caption). Pre-punishment: before the first loss (about card 10). Pre-hunch: “no notion of what was happening”, controls roughly cards 10–50, patients 9–80. Hunch: “liking” or “disliking” decks and guessing which were risky, unsure; controls roughly cards 50–80; never reached by patients. Conceptual: can say which decks are good or bad and why; after about card 80; reached by 7 of 10 controls and 3 of 6 patients.
• SCR as index. An anticipatory SCR is one generated in the window before turning a card, “during the time the subject ponders” (Fig. 1 caption). It is taken to index a nonconscious somatic signal.

Key results

• Narrative (p. 1293). Before losses, everyone prefers A and B and shows no significant anticipatory SCRs. After a few losses (usually by card 10), controls begin anticipatory SCRs “to decks A and B”, yet at card 20 all say they have no clue. By about card 50 all controls voice a hunch and generate anticipatory SCRs before A or B. The three controls who never reached the conceptual period “still made advantageous choices”. No patient produced anticipatory SCRs or a hunch.
• SCR statistics (note 8). Three-way ANOVA (group × period [pre-punishment, pre-hunch] × deck [bad, good]). The reported key effect is group × period, F(1,14) = 16.24, P < 0.001. Newman-Keuls: no group or deck differences at baseline; SCRs rose from baseline to pre-hunch only in controls (P < 0.01), and controls’ pre-hunch SCRs exceeded patients’ in both periods (P < 0.01). No deck effect or deck interaction is reported for these periods. All controls and no patients generated anticipatory SCRs: Fisher’s exact test, one-sided P < 0.001. In the hunch and conceptual periods, bad-deck SCRs were “generally higher” than good-deck ones but “the difference did not reach statistical significance”; bad-deck (not good-deck) SCRs in those periods exceeded baseline (P < 0.01).
• Choice statistics (note 9). ANOVA on cards selected: group × period × deck, F(1,14) = 6.9, P < 0.02. Pre-hunch: patients chose significantly more bad than good cards (P < 0.01); controls chose more good than bad, “not statistically significant”. Hunch and conceptual: controls chose significantly more good cards (P < 0.01). Conceptual: patients still chose more bad cards, not significantly.
• Model (p. 1294, Fig. 2). Two interacting chains, as in the key insight. The ventromedial cortex holds the dispositional link; its activation drives autonomic and neurotransmitter (e.g. dopamine) nuclei; the signals bias cognitive evaluation. Damage “acts by precluding access to a particular kind of record of previous and related individual experience”. Note 7 places the bias mechanism alongside response inhibition, working memory and selective attention: “an addition … rather than an alternative”.

Claims

id	claim	strength	support
C1	Controls, and not ventromedial patients, develop anticipatory SCRs in the task	strong for the group difference at this sample size	Fisher’s exact, all 10 against none of 6, one-sided P < 0.001; group × period F(1,14) = 16.24 (note 8)
C2	Patients who can state the correct strategy still choose disadvantageously	moderate	3 patients reached the conceptual period and chose badly (text); the conceptual-period bad-over-good preference is not significant (note 9), so “still chose badly” means did not switch
C3	Controls begin to choose advantageously before they realise which strategy works	weak for “before any awareness”; moderate for “before the articulated strategy”	pre-hunch good-over-bad not significant (note 9); significant from the hunch period, when controls already report liking and disliking decks; three controls who never reached the conceptual period chose well (text, no statistics)
C4	Controls generate anticipatory SCRs to risky choices before knowing they are risky	weak for the deck-specific claim	the pre-hunch effect reported is group × period, with no bad-versus-good test reported (note 8); deck differences are not significant in later periods either
C5	In normal individuals, nonconscious biases guide behaviour before conscious knowledge does	assertion beyond the data	an inference from C1–C4, resting on C3 and C4, the weakest of them; the awareness probe is coarse and starts at card 20
C6	Without such biases, overt knowledge may be insufficient for advantageous behaviour	moderate	C2, on three patients; the authors hedge it with “may”
C7	The ventromedial cortex holds dispositional links to past emotional experience that drive the bias (two-chain model)	assertion	the somatic marker hypothesis (note 6, citing Damasio 1994 and Damasio, Tranel & Damasio 1991); not tested here

Concepts

• Anticipatory SCR. A skin-conductance response in the interval before a card is chosen, while the participant ponders; contrasted with SCRs to the reward or penalty itself.
• Pre-hunch, hunch, conceptual. Periods defined from the verbal reports (see Assumptions). A hunch is liking or disliking decks and guessing which are risky, without certainty.
• Covert bias. A nonconscious signal, from autonomic and neurotransmitter activity, that facilitates rather than makes a decision.
• Dispositional knowledge. Nondeclarative knowledge related to previous emotional experience of similar situations, held, the authors suspect, in ventromedial cortex (p. 1294). The term is from Damasio’s convergence-zone work.

Connections

• Damasio 1996 (LIT-384, NOTE-332). The statement of the hypothesis this experiment tests. Its §5 gives the task and the earlier SCR result, and previews this one in a sentence (“long before they have … any notion whatsoever”). This report is the evidence for that sentence, and reads as weaker than the sentence.
• Damasio 1989 (LIT-385). The source of “dispositional” knowledge; unread, registered from its abstract.
• Consciousness readings (LIT-056, LIT-191). The record’s work on consciousness asks mostly what would show a machine is conscious. This is a human case in which a bodily signal and a verbal report come apart, in both directions, which is the kind of dissociation the indicator approach in those works has to interpret.
• Cross-modal fluency (LIT-190, NOTE-146). The “hunch” period, in which participants like and dislike decks without knowing why, is a felt quality entering a judgement before its source is known, the shape of claim LIT-190 makes for processing fluency.
• Damasio’s later work. Damasio’s later papers on feeling, homeostasis and consciousness are being filed at the same time; this reading makes no claim about how they use the result.
• Not held in either record. Bechara et al. 1994 (Cognition 50:7–12) and 1996 (Cerebral Cortex 6:215–225); Descartes’ Error (1994); and the critique by Maia and McClelland (2004, PNAS, DOI-10.1073/pnas.0406666101), which re-examined what participants know in this task. Its title says it disputes the awareness timing; it was not read here.

Bearing on the record

• No THEORY bears on it. If the record later wants an account of affect in decision, this paper supports the group difference in anticipatory SCRs and the insufficiency of stated knowledge in three patients. It does not by itself support “nonconscious biases come first”; that needs the deck-specific pre-hunch tests this paper does not report, and a reading of the awareness critique.
• Topic. Tagged cognition first: the paper is about how a decision is reached and what the decider knows. consciousness for conscious versus nonconscious knowledge, emotion-and-affect for the bodily and emotional signal, neuroscience for the lesion and autonomic method. Nothing else is left unsaid.
• No ML instruction. Nothing here belongs in the anthology.

Limitations

• Sample. 10 controls and 6 patients; control demographics unstated; the conceptual-period patient result rests on 3 people.
• Awareness probe. Two open questions, from card 20, every 10 cards. A participant who knows something but does not volunteer it is coded as not knowing it, and nothing is asked during cards 10–20, when the SCRs are said to begin.
• Abstract versus notes. The abstract’s “began to choose advantageously before they realized which strategy worked best” and “whenever they pondered a choice that turned out to be risky, before they knew” outrun notes 8 and 9, as set out in the claims table.
• Correlation. The SCRs and the choices are measured together; nothing here shows that the SCR, or what it indexes, causes the choice.
• Mechanism. The two-chain model and the ventromedial role in it are stated, not tested.

Open questions

• Is there a bad-versus-good deck SCR difference in controls during the pre-hunch period? The analysis in note 8 does not report one.
• Do controls in the pre-hunch period know more than the two questions elicit? This is the question the later critique takes up.
• Does blocking or reducing the peripheral signal change choices, which would bear on whether the SCR is a cause or a by-product?

Corrections

• none (there was no seed)
• Abstract versus note 9. The abstract says controls “began to choose advantageously before they realized which strategy worked best”. By note 9 their pre-hunch preference was not significant; it became significant in the hunch period, when they already reported liking and disliking decks. True for the articulated strategy, not for a hunch.
• Abstract versus note 8. The abstract says controls generated anticipatory SCRs “whenever they pondered a choice that turned out to be risky, before they knew”. The pre-hunch test reported is group × period, not bad versus good deck, and later bad-versus-good differences are not significant.