his summer, AI will be framed in two different ways. At the G7, it will be treated as a governance problem and at VivaTech, as an opportunity. These approaches are not yet aligned, and that misalignment is beginning to shape how AI is being deployed in practice.
We are approaching the implementation of AI across societies through fragmented and often siloed approaches. At Saviesa, we see extensive discussion about what the future of AI could or should be, but far less about the conditions required to get there. There are early attempts to address this gap, but none can yet claim success, and each remains incomplete. The practical task now is to shift the focus of debate away from competing visions of AI, and towards the conditions that make its use accountable in practice.
The EU AI Act classifies risk and attaches obligations, with certain uses prohibited and others subject to requirements on transparency, human oversight, and accountability but its credibility will depend on whether it can hold under commercial pressure and rapid technological change, particularly as parts of the framework are still being implemented. UNESCO’s AI ethics framework has been widely adopted because it provides a common language. However, adoption does not guarantee implementation, and evidence of consistent application across countries remains limited.
This is an opportunity to rebuild consensus. Not in agreeing on a shared vision of AI, but in establishing agreement on the conditions under which it is used. From the work we are doing in education, creativity and leadership, one pattern is already clear: Practice is outrunning principle. In education, AI systems are being used as a matter of course by students and institutions, often ahead of formal guidance. Pilot work by one of our UK partners, Big Education, illustrates how educators are navigating a rapidly shifting landscape with incomplete and sometimes conflicting information.
In our research on leadership, we are also seeing the limits of assumptions that have been treated as wisdom for decades:
- “More data leads to better decisions.” In practice, it often produces greater confidence without greater understanding, and in some cases, confidence in the wrong conclusions.
- “Ethical questions can be addressed after deployment.” In reality, organizations are finding it difficult to retrofit governance and accountability into systems that are already embedded in operations.
- “Efficiency is a sufficient measure of progress.” Increasingly, speed is at the expense of judgement, reflection, and long–term capability.
None of these assumptions has collapsed entirely, but all of them need to be reconsidered. The emerging needs are demanding ones:
- Human oversight must move from principle to requirement, not because it is philosophically attractive, but because systems that cannot be questioned or refused or have no responsibility are difficult to justify once deployed.
- AI literacy needs to evolve. It is no longer primarily about use, but about judgement: understanding limits, recognizing bias, and knowing when reliance is misplaced.
The question now is not whether new visions of AI will emerge but whether we are prepared to set the conditions that make those visions workable.
a global affairs media network
Beyond visions, AI needs working conditions
Image by Dan Cristian Pădureț via Pexels.com
June 10, 2026
AI practice is outrunning principle. Focus must shift to the conditions that make use accountable, writes Leonor Diaz Alcantara.
T
his summer, AI will be framed in two different ways. At the G7, it will be treated as a governance problem and at VivaTech, as an opportunity. These approaches are not yet aligned, and that misalignment is beginning to shape how AI is being deployed in practice.
We are approaching the implementation of AI across societies through fragmented and often siloed approaches. At Saviesa, we see extensive discussion about what the future of AI could or should be, but far less about the conditions required to get there. There are early attempts to address this gap, but none can yet claim success, and each remains incomplete. The practical task now is to shift the focus of debate away from competing visions of AI, and towards the conditions that make its use accountable in practice.
The EU AI Act classifies risk and attaches obligations, with certain uses prohibited and others subject to requirements on transparency, human oversight, and accountability but its credibility will depend on whether it can hold under commercial pressure and rapid technological change, particularly as parts of the framework are still being implemented. UNESCO’s AI ethics framework has been widely adopted because it provides a common language. However, adoption does not guarantee implementation, and evidence of consistent application across countries remains limited.
This is an opportunity to rebuild consensus. Not in agreeing on a shared vision of AI, but in establishing agreement on the conditions under which it is used. From the work we are doing in education, creativity and leadership, one pattern is already clear: Practice is outrunning principle. In education, AI systems are being used as a matter of course by students and institutions, often ahead of formal guidance. Pilot work by one of our UK partners, Big Education, illustrates how educators are navigating a rapidly shifting landscape with incomplete and sometimes conflicting information.
In our research on leadership, we are also seeing the limits of assumptions that have been treated as wisdom for decades:
- “More data leads to better decisions.” In practice, it often produces greater confidence without greater understanding, and in some cases, confidence in the wrong conclusions.
- “Ethical questions can be addressed after deployment.” In reality, organizations are finding it difficult to retrofit governance and accountability into systems that are already embedded in operations.
- “Efficiency is a sufficient measure of progress.” Increasingly, speed is at the expense of judgement, reflection, and long–term capability.
None of these assumptions has collapsed entirely, but all of them need to be reconsidered. The emerging needs are demanding ones:
- Human oversight must move from principle to requirement, not because it is philosophically attractive, but because systems that cannot be questioned or refused or have no responsibility are difficult to justify once deployed.
- AI literacy needs to evolve. It is no longer primarily about use, but about judgement: understanding limits, recognizing bias, and knowing when reliance is misplaced.
The question now is not whether new visions of AI will emerge but whether we are prepared to set the conditions that make those visions workable.
