Co-Intelligence

You ask an AI for a project proposal. Seconds later, a clearly structured text appears, persuasively written and remarkably complete. It is tempting to forward it immediately. On closer inspection, however, two figures are wrong, one source does not exist and an important objection is missing.

In Co-Intelligence , Ethan Mollick describes generative AI as a new form of collaboration. He invites us to explore these systems actively—while remembering that they can produce convincing-sounding errors. Co-intelligence does not emerge because AI thinks for us, but because we deliberately design tasks, standards and oversight.

1. Experiment with real tasks

AI literacy does not develop through feature lists alone. A limited use case is more helpful: drafting an outline, gathering counterarguments, adapting a text for another audience or generating variations of an idea. Only concrete work reveals where a system saves time, stimulates new perspectives or creates additional checking work.

A clear role description is worthwhile. Should the AI act as a critical countervoice, an explanatory partner or a source of ideas? The more precisely context, audience and quality criteria are specified, the easier it is to assess the result. The role metaphor helps with instruction—it does not mean the system possesses consciousness or human understanding.

2. Do not confuse fluent language with truth

Generative AI predicts the statistically appropriate next element. This can produce helpful drafts, but also invented facts, one-sided accounts or errors that are barely noticeable in the language. Every task therefore needs an appropriate checking process.

For brainstorming, a plausibility check may suffice. For medical, legal, financial, scientific or personal-data-related statements, robust original sources and expert review are indispensable. Good collaboration therefore also means distinguishing tasks by risk and giving AI no more authority than its output deserves.

3. Keep human responsibility visible

When an output is produced together with AI, it must still be clear who decides and bears responsibility for what. People determine the goal, recognise social contexts, weigh values and bear the consequences. This also includes asking what data may be entered at all and whether affected people should know that AI was used.

The strongest use often lies not in complete delegation, but in a loop: a person formulates intention and criteria, AI generates options, the person checks and adds to them, and AI helps with a further version. Our own thinking thus shapes the entire process rather than serving only as a final checkpoint.

Your exercise: a controlled AI experiment

Choose a manageable task without confidential data. Spend three minutes preparing:

1. Minute 1 – The brief. Specify the goal, context and desired role of the AI. Also state what it should not decide.
2. Minute 2 – The standard. Note three criteria for a good result, such as factual accuracy, suitability for the audience and visible uncertainty.
3. Minute 3 – The check. Decide which statements you will check against which original sources or your own expertise.

Then compare the AI draft with a short version of your own: what improved, what became weaker and which step remained clearly human?

What we take from this

Co-intelligence combines technical experimentation with critical judgement. In WEITER7 , it becomes a clearly bounded experiment: choose an application, set a quality standard, check the output and deliberately adjust your way of working.

For which task could AI broaden my perspective—and which decision do I explicitly not want to delegate to it?

Context: This insight summarises selected ideas from Mollick’s practice-oriented book in our own words. The exercise and WEITER7 connection are our own applications. The capabilities, risks and terms of use of generative AI change rapidly. This approach does not replace data-protection review, expert oversight or internal organisational rules. Human responsibility is indispensable for decisions with significant consequences.

Sources and further reading: Official publisher’s page ↗ · Knowledge at Wharton: conversation with Ethan Mollick ↗