Run an AI portfolio for a year and see its risk-adjusted return
Business teams and vendors bring you AI systems. You decide which systems go into use and under which conditions, with a limited budget and limited time. The game is free and runs in your browser.
Your year as Chief AI Officer
Business teams and vendors bring you AI systems to approve. You decide what goes into use and under which conditions, with a limited budget and limited time each quarter.
Your score
The real risk of each system and the reliability of each vendor’s claims are set at random for every game and stay hidden until the end. Your score compares the expected value of your decisions with the best result that was possible. Good or bad luck with incidents does not affect it.
How the game works
Each level is one year in four quarters. Requests come from business sponsors and from vendors, and each request shows the value of the system and the losses that the organisation expects from it. The map beside your desk shows each system by its yearly value and its expected yearly loss, so you can see the full portfolio while you decide on one request. The AI value-to-risk ratio compares the same two figures for each system. On the later levels, some AI tools are already in use before anyone asks you, which is the problem of shadow AI.
Decide
Each request has four options, from approval as proposed to a decision against the system. Safeguards cost budget, and some of them delay the start.
Check
You can have an in-house system tested, request a vendor’s evidence, or speak to the vendor. A check uses one of your actions for the quarter, so you cannot check everything.
Run
A system in use earns money each quarter and can cause incidents. Each quarter of use also tells you more about its real risk, and its range on the map becomes narrower.
Review
At the end of the year, the board review shows the real risk of each system, the quality of each vendor’s evidence, and the decision that was best.
The score
Your score is the expected value of your decisions as a share of the best result that was possible with full information. Luck does not change the score. A year without incidents does not increase it, and an unlucky incident does not decrease it. Each level has a pass mark. Each new game takes different systems from a pool of more than 100 cases, so a level is different each time you play it. The first two levels are open to everyone, and the later levels open when you give us your work email.
The player’s decisions had an expected value of CHF 1.1M. The best possible decisions had an expected value of CHF 1.3M.
The method behind the game
The game uses the same calculation as risk quantification on the Modulos platform. The expected loss of a risk is the number of incidents per year, multiplied by the loss per incident, and reduced by the share that the safeguards stop. A claim about a safeguard counts in full only when independent tests support it, and a claim that is supported only by marketing material counts for very little. The game is a simplified model, but it asks the questions that a governance team must answer for each real system: the value of the system, the possible cost, and the quality of the evidence.
On the Modulos platform, these estimates come from your own AI inventory, your controls and your evidence, and the platform repeats them on a schedule. The sample portfolio shows the result for a full set of AI systems. All vendors, products and people in the game are invented.
Get a demoThe grey figure is not verified. Against CHF 2.5M of value a year, Fennel returns between −CHF 0.5M and CHF 1.1M as proposed.