목차
Title page 1
Contents 3
About this toolkit 4
Introduction 5
What are hidden risks? 8
Limitations of existing approaches to AI safety 11
A new approach: surfacing 'hidden' risks 13
The Mitigating 'Hidden' AI Risks Framework 15
1. Quality Assurance 18
2. Task-tool mismatch 20
3. Perception, Emotions and Signalling 22
4. Workflow and Organisational Challenges 25
5. Ethics 28
6. Human Connection and Technological Overreliance 30
Step-by-step: How to use the Mitigating Hidden AI Risks Framework 32
Step 1. Set up a multidisciplinary and diverse working group 33
Step 2. Surface potential hidden risks for your tool 35
Step 3. Review and prioritise risks 37
Step 4. Monitor and develop mitigation strategies for your risks 38
Step 5. Implement ongoing monitoring and review mechanisms 42
Tips for Teams 44
Scope and limits of this guide 46
Acknowledgements 48
References 50
Figures 12
Figure 1. Examples of challenges which show why 'human-in-the-loop' is not a fixall solution 12
Figure 2. To prevent unintended consequences, we have to understand the mechanisms which create 'hidden' risks and which can lead to negative consequences and outcomes 14
Figure 3. Six categories of 'hidden' risks arising from organisational AI roll outs 16
Figure 4. Three principles to ensure that humans have the right conditions to be "in the loop" 39
