The Trust Curve
Seven questions I’ve used to calibrate my relationship with AI
helped me build a company.
It also confidently fabricates facts.
That is a difficult combination for us to hold at once. With humans, when our trust is broken, we step back. With AI, one session expands what we think it can do for us and the very next reminds us never to let our guard down.
Research on algorithm aversion shows that seeing a forecasting algorithm make a mistake can quickly erode people’s confidence in it. If I let the bad session decide I don’t trust AI anymore, I lose a critical collaborator. If I let the brilliant session make me over-rely, I risk hurting my credibility.
Here are seven questions I’ve used to calibrate my trust in AI. These questions are a working map, shaped by my own experience (building an AI-native company) and by experiences of CXOs experimenting with AI firsthand. These are not seven linear stages to graduate through as you learn to collaborate more heavily with AI.
Trust in AI can be a climb, a slope or a loop, all on the same day. The one thing I can tell you is the day after a bad AI day is a more cautious, better AI day!
Ability: Can it do something useful for me?
I’m just going to say it: most of us started with a recipe look-up or the ‘3 days in Paris’ itinerary. That won’t tell you much about what AI can unlock for you in your work life. Give it something you would have done yourself or delegated to your most trusted lieutenant. Choose nuanced work you understand deeply so you can interrogate it to your heart’s content!
Reliability: Can I count on it for this task consistently?
A good answer once is exciting. A good answer on the next assignment is more useful.
AI can disappoint you when you start a fresh conversation and expect it to somehow remember what made the last result good. I have had to learn to give it more than a prompt. It needs examples of the standard I’m asking for, along with the corrections I made and why.
Integrity: Is it telling me the truth?
This one cost me a week of rework. Without the right guardrails, AI presented invented data as research insights rather than reporting what it found. I had to start clean. AI can give a correct answer and an invented one in the same assured voice. Asking “Are you sure?” does not establish which is which. Nor does a beautifully written source list if you never open the sources.
AI can give a correct answer and an invented one in the same assured voice.
Resonance: Can it bring me unique, complementary ideas?
At some point, the work becomes more than getting through a task. This one was a game changer for me, using AI to extend into horizons I haven’t had formal instruction or experience in.
Resonance isn’t the same as being impressed. It’s curating concepts you wouldn’t have thought of on your own.
Mirror: Does it understand how I work?
The real unlock is when you can actually get it to operate at your level for certain tasks. By now you know which tasks those could be. But it comes from a lot of curating, teaching, correcting and still owning the last mile. Almost like building your own personal context graph.
Agency: Can it act on my behalf?
Building an “agent” is not the milestone. Getting back bandwidth as a CXO is. I believe this is the most profound shift we are living through. What can AI truly take off your plate so you can focus on higher-order things?
Building an “agent” is not the milestone. Getting back bandwidth is.
Transcendence: What new horizon does it unlock for me?
I see a groundswell of experienced professionals taking a chance on themselves, powered by AI. Building companies and projects they never imagined for themselves. A doctor building a holistic wellness platform, a teacher building a global storytelling platform, a consultant building a product (yup, sometimes you just pat your own back. Especially if you’re a new founder!).
Trust in AI is a deeply personal journey. What I may trust it to do may not be what you feel comfortable with. The only way to know is to get your hands dirty and calibrate. I still have brilliant AI days and bad AI days. The bad ones no longer tell me to walk away; the brilliant ones no longer give me permission to stop checking. I know more clearly what to ask it to do, what to inspect myself and when to just do it on my own.
Happy calibrating!