Agency, under uncertainty

Acting under
uncertainty

Every month brings a new forecast about which jobs AI will take, and every month some of last month’s forecasts look wrong. You can spend the year trying to guess right. Or you can arrange your work so that you do not need to.

That second option has a name and an author. Nassim Nicholas Taleb spent five books, from Fooled by Randomness (2001) to Skin in the Game (2018), on one practical question: how do you act when you cannot know what will happen? His answer is not a better forecast. It is a way of standing so that surprises cost you little and can still pay off.

On this site Taleb is the third of three working sources. The Hermetic texts give direction: what you attend to and what you are for. Transurfing gives a vocabulary for attention and for the weight you put on outcomes. Taleb gives the rules for acting without knowing the ending. None of the three is proof of the others; each covers a different step of the same loop.

The loop

Notice, choose, act, observe, adapt. The first two steps are about attention and direction, and the rest of this site covers them: five questions before you ask AI is a tool for noticing and choosing. This page is about the last three: how to act when the result is uncertain, how to read what comes back, and how to change course without having bet everything on the first try.

Five tools, in our words

1. Avoid ruin first. An average over many people is not what happens to you over time. If one step can end the game, a lost job you cannot replace, a reputation, your health, then “on average it pays” stops being an argument. Taleb develops this in Skin in the Game (2018) and, with co-authors, in a 2014 paper on the precautionary principle. For AI this means: no single tool, platform or prediction should be able to take out the thing you cannot rebuild.
2. Keep options. An option is a right without an obligation: small, cheap tries whose loss is capped and whose gain is open. Antifragile (2012) borrows an old example from Aristotle’s Politics: Thales renting olive presses cheaply in advance and profiting when the harvest came in large. With AI, the cheap try is an afternoon with a new tool on a real task, not a quarter rebuilt around it.
3. Use a barbell. Most of your work very safe, a small part openly experimental, nothing in the risky-looking middle. Taleb describes the strategy in The Black Swan (2007) and generalises it in Antifragile. In practice: keep the core of your craft done the way you trust, and put a fixed slice of time into experiments you would not mind losing.
4. Subtract before you add. Via negativa, a term from negative theology, in Taleb’s use means improving by removal: what to stop, what to avoid. When everything promises more output, the higher-leverage move is often to cut one input: a feed, a tool, a meeting, a dependency.
5. Keep skin in the game. The one who decides should bear the cost of being wrong. Taleb and Constantine Sandis made it an ethical rule in 2014. A model that recommends bears no cost when the recommendation fails; you do. So the model can supply options and counter-arguments, and the decision with consequences stays with the person who will live with them. This is the same line Human agency in the AI age draws between delegating the input to judgement and the output.

One more idea from Antifragile helps with the question of which skills to keep. For things that do not perish, ideas, books, techniques, the longer something has lasted, the longer it is likely to last. Taleb calls it the Lindy effect. Careful reading, clear writing and judgement under pressure have lasted a long time. A prompt trick from last month has not.

Four moves for this week

The arrangement is ours. Each move says what it should do and where it stops.

1. Name your ruin. Write down the one or two things you must not lose: income you cannot replace, trust of a key client, your health. Check this week’s use of AI against that line only. Anything that cannot reach it is allowed to fail.
2. Run one capped experiment. Pick one new tool or one new way of working, give it a fixed budget, two hours or one task, and write before you start what would count as worth keeping. Most tries will not be. That is the point: the loss is small and known, the gain is not.
3. Remove one input. Before adding anything, take one thing away: a newsletter you skim, a tool you open out of habit, a recurring meeting. Note after a week whether you missed it.
4. Ask who pays if it is wrong. For each AI recommendation you act on this week, write one line: who bears the cost if this is wrong? If the answer is “someone else”, slow down and check it yourself.

Where this comes from

• Source. Taleb’s Incerto: Fooled by Randomness (2001), The Black Swan (2007; second edition 2010 with an essay on robustness and fragility), The Bed of Procrustes (2010), Antifragile (2012) and Skin in the Game (2018), plus two papers from 2014 that are free to read. The ideas were built on financial markets and risk.
• Our reading. Carrying them over to work with AI is our step, and it is an analogy, not a finding. Taleb does not write about Hermeticism, and we do not present him as part of that tradition. He is here because he answers a question the older texts do not: what to do when you cannot know the outcome.
• Practice. The four moves are our arrangement, not Taleb’s.

What this does not settle

Whether these tools fit your situation. Reviewers have pushed back hard on Taleb: some found his claims overstated outside finance, others rejected antifragility as a general recipe. A personal life is not a portfolio, and some losses, of a relationship or of time, do not come with a capped downside you can choose in advance.

Nor does it predict what AI will do to your field. The whole approach starts from admitting that nobody can. It only claims that you can arrange to be less hurt by the surprise and more able to use it.

Where to go next

For the noticing and choosing steps of the loop: five questions before you ask AI and Human agency in the AI age. For attention and the weight you put on outcomes: importance in Transurfing, with its limits. And the letter, As Above, one source and one question at a time: get the letter.

Fair questions

What does antifragile mean?

Taleb’s word, from Antifragile (2012), for things that gain from disorder and stress rather than only surviving it. Fragile things lose from volatility, robust ones are roughly unaffected, antifragile ones benefit.

How do you make decisions under uncertainty?

In Taleb’s terms: first make sure no single step can ruin you, then prefer small tries with a capped loss and an open gain, improve by removing before adding, and keep the people who decide exposed to the cost of being wrong.

What is the barbell strategy?

Putting most of your resources somewhere very safe and a small part into openly speculative bets, with nothing in the middle. Taleb describes it in The Black Swan (2007) and generalises it in Antifragile.

Is Taleb part of the Hermetic tradition?

No. He does not write about Hermeticism. We use his tools for one step of a practical loop: acting when the outcome is unknown.

Sources & further reading

• Nassim Nicholas Taleb, Fooled by Randomness (Texere, 2001); The Black Swan (Random House, 2007; 2nd ed. 2010); The Bed of Procrustes (Random House, 2010); Antifragile (Random House, 2012); Skin in the Game (Random House, 2018).
• Taleb, N. N. and Sandis, C. (2014). The Skin In The Game Heuristic for Protection Against Tail Events. Review of Behavioral Economics 1(1-2). arXiv:1308.0958
• Taleb, N. N., Read, R., Douady, R., Norman, J. and Bar-Yam, Y. (2014). The Precautionary Principle (with Application to the Genetic Modification of Organisms). arXiv:1410.5787
• Aristotle, Politics I.11: the story of Thales and the olive presses, retold in Antifragile.
• The application to work with AI and the four moves are our own arrangement.