1/14
Looks like no tags are added yet.
What is AI?
AI: Computer programs that can complete cognitive tasks typically
associated with human intelligence
E.g. Youtube algorithm
what are the 3 categories of AI used in medical science?
➢Machine Learning – Analyse data to make decisions or predictions
Ability to learn patterns and be trained
➢Deep Learning – Multi-layered neural networks to learn complex patterns
This has advanced ai in the past few years
Neural networks – Layers of interconnected nodes
∴ Taking into account multiple layers of information processing
➢Generative AI – AI that can generate new content (text, images, music etc.)
*Med science been using other forms of AI for years in lab data analysis
Large Language Models
Deep learning model trained on Massive data set
NLP - Generates & understands human language
Why is AI creating a challenge
➢Education
Reliance on AI hampering understanding
Cheating (Higher education sectors had to change all assessments to avoid this)
➢Medical Research
AI generated publications, even journals which are meant to be accurate,
Conclusions drawn on hallucinations or out of date info
➢Healthcare
Self-diagnosis increase, similar to people googling their symptoms in previous years, leading to bias from patient making professional diagnosis difficult
Medication adherence e.g. looking up side effects
How is AI effecting post grad jobs?
Graduate entry jobs are down
Opportunities for vaccines requiring AI/Machine learning tools
PhySoc report on sepsis treatment.
AI was trained on previous treatment decisions, including suboptimal ones, and tested in shadow mode alongside clinical evaluation.
How can AI support sepsis treatment?
Analyses patient data & suggests dose of IV fluid + vasopressor doses
What are the potential benefits of using AI in medical treatments?
Allows hospitals to work through more patients with reduced staff which is critical during outbreaks.
Why is ML not replacing the medical scientist?
Its been around for a while and human element still required to: train programs and the model, write scripts and assess/analyse data
What are the dangers of AI in medical science
➢ Knowledge cut off, trained up to a point then training stops, frequent updates required
Limited to the knowledge that was present at time
of ML
Results in hallucinations where AI makes up based on a prediction
➢ Harmful biases
Look at the demographic in the AI generated images: perpetuating stereotypes
➢ Poor at producing accurate graphical representations, specific to science
What are the 3 actions that PhySoc want taken into account?
Establish a Physiology & AI framework to set improved guardrails for AI in health
they learn patterns by taking in all information even if some useless, needs to be centred around this, money going to the saftest and best models
Needed to ensure safety but also acceptance by intended users
Ensure that research funding mechanisms prioritise physiologically plausible AI tools
Embed physiological evidence in regulatory approval of AI tools
Include physiological evidence/insight as a requirement for regulatory approval
Assessment
What makes medical scientists ideal AI pilots?
The better the input, the better the output. The more you know about a task the better you can subtask
How can you prevent hallucinations?
Inclusion of sources
What is sub-tasking?
breaking down a task into sub-tasks is easier with an in-depth knowledge of the task/procedure
How can AI be used for science communication?
difficult to explain topics to people on the street as a specialised scientist so AI can be used
➢ “Summarise the paper for a non-scientific audience?”
Wha is the human in the loop approach?
when AI provides recommendations or decisions, but a human (e.g. a doctor) reviews and approves them before they are acted upon.