# Group 1 Notes
###### tags: `TAS Hub`


:::info
**Resources**
- Link to case study: https://drive.google.com/file/d/12Z11YqzoKvWM3HyVN8Rtdl3ux2_6oUhp/view?usp=sharing
- Link to Assurance Platform: https://turing-assurance-platform-dev.azurewebsites.net/
- Link to workshop slides: https://drive.google.com/file/d/1RZypnxSwEsKD2riZilXd5gKguMkrhvS2/view
:::

:::warning
**Objectives**
1. Decide on a goal for your assurance case.
2. Ensure the goal's scope includes ethical considerations.
3. Formulate a set of interlinked claims about properties of the system or project that are relevant for establishing the goal.
4. Identify the evidence that justifies how these properties have been established within your project or system.
:::

## Assurance Case Notes

### Notes on Goal(s)

- The system is continuously monitored to minimise the harms that could arise from unintended consequences
- Votes: ***

- The AI system supports healthcare professionals deliver on their professional duty to ensure patients are informed about and understand the decisons being made
- Votes: **

- the clinical efficacy of the AI system is assessed through a double blind test so is better or as good as a human equivalent - so it can be "turing" tested! This is based on the idea that it is a training tool rather than something for an experienced clinician to use.
- Votes: *

- Real world monitoring for bias, drift and clinician feedback loops enables trustworthy and accountable use
- Votes: ****

- AI system provides decision support and knowledge aquisitio of clinician to ensure development and retention of expertise
- Votes: *

- The NLP algorithm makes clinically plausible recommendations
- Votes: *

- Long-term use of the system improves professional judgement and clinical decision-making for newly qualified psychiatrists
- Votes: ******

- The NLP algorithm draws a clinician's attention to patient speech characteristics which are easily missed by humans

### Notes on Property Claims

- Confidence / probability of recommendations
- feedback loops when clinician disagreement occurs data is captured and outcomes measured for learning
- knowledge aquisition supported to develop expertise
-

### Notes on Evidence

- active monitoring of an AI system by AI users improves their knowledge of an AI system's limits and their own system-independent diagnostic abilities
- feedback loop - data provided of user responses and trends of disagreement/agreement and final outcomes

### Unordered Notes
- appropriateness for different training levels of psychiatrists
- Are Recommendations aligned to specific clincial guidelines (e.g. NICE guidelines)
- Uncertainty about efficacy of NLP for people with strong accents or from minorty communities (discriminatory impact of tool)
- Ability for tool to make accurate recommendations for all neurodiverse individuals
-