When calculating expected cost/benefit, you take the value of the outcome multiplied by the probability of that outcome. So the expected cost to play Taleb's game of Russian Roulette is 1/6 * $6m = $1m, if you value your life at $6m.

Note that's a bit higher than the expected benefit of $0.833m, I kept the numbers round for brevity. The breakeven point would be:

5/6 * C_win = 1/6 * C_lose

C_win = 1m

C_lose = 5/6 * 6/1 * 1m = $5m

My point was that Taleb omitted this calculation entirely, which makes it an incomplete cost/benefit analysis.