Human-AI Interaction & Thinking
Think With AI: thinking, verifying and deciding with AI. The formats come from EIL research and go beyond prompt engineering.
Status: pilot. The three formats are drafts derived from EIL research. They have not yet been tried with external groups, and there are no results yet.
What the formats work on
- Human-AI Thinking
- Cognitive collaboration
- Critical reasoning
- Perspective taking
- Hypothesis testing
- Complex problem solving
- Longitudinal interaction
- Agentic thinking
- Cognitive independence
- Verification and human oversight
Three formats to test
Compact team workshop
- For whom
- Teams of 6 to 12 people who use AI daily for analysis or decision work.
- Learning goals
- Recognize one’s own checking habits toward AI answers. Develop a verification protocol for one’s own daily work.
- Exercises
- Verification drill: support, refute or leave open statements by an AI.
- Perspective taking: have the AI present the same situation from three roles and name the differences.
- Counter-hypothesis: deliberately test one’s own assumption against the AI.
- How success is seen
- Each participant leaves with a written verification protocol. Before and after, participants rate how sure they are about AI statements and how often that confidence was right.
- Delivery
- Half to one day, materials and a short results report.
In-depth research and interaction program
- For whom
- Small cohorts from research, consulting and development who work with AI over weeks.
- Learning goals
- Observe and describe one’s own use of AI over several weeks. Deliberately preserve cognitive independence, for example through tasks without AI help as a baseline.
- Exercises
- Interaction log: document selected dialogues and analyze them together.
- Tasks with and without AI in alternation, comparing the results.
- Agentic work: split a task, delegate parts and check the handoffs.
- How success is seen
- A personal trajectory report per participant and a joint analysis of the cohort. Duration and scope are fixed only after the pilot.
- Delivery
- Several weeks, accompanying, with an interim analysis.
Interactive website demonstration
- For whom
- Anyone who wants to try in a few minutes what verifying with AI looks like.
- Learning goals
- Practice the difference between supported, unsupported and needs-checking.
- Exercises
- Rate three example statements and read the reasoning. The exercise runs in the browser; no input is stored.
- How success is seen
- The exercise shows the principle. It measures nothing and does not replace training.
- Delivery
- On this page, see below.
Try it: supported or not?
Rate each statement. The reasoning follows.
The AI says: “Studies show teams with AI always decide faster.” No study is named.
The AI summarizes your own document and quotes two paragraphs verbatim.
You calculate 12 × 15 yourself and get 180. The AI also says 180.
Ask for a pilot
If your team wants to test a format, write to us. Price and scope are set for your case after a conversation.
CONTACT
Work with the Lab
Research collaboration, methodology question or applied-research enquiry.
