AI

Question 1

There are 3 types of AI according to the article. What does the following mean?
- Supervised

Answer 1

The goal is a prediction of a known outcome. A model is trained on a dataset that includes features (variables) and labels (outcome or class of interest). The algorithm generates a function, which maps features to labels and then uses it to predict the labels of new unlabelled data. For instance, if the objective is to predict mortality following allogeneic haematopoietic stem cell trans- plantation (HSCT), the model will be trained on a dataset containing information on patient, disease and transplanta- tion characteristics as well as the associated outcome (e.g survival for each individual).

Question 2

There are 3 types of AI according to the article. What does the following mean?

• Unsupervised

Answer 2

Unsupervised learning is not about predicting a specific output. Instead, the algorithm attempts to identify patterns or groupings within the data. This is inherently a more chal- lenging task to judge, and often the value of groups ‘learned’ through unsupervised learning is evaluated by performance in a subsequent supervised learning task, assessing whether these groupings are biologically or clinically useful. Using the unsupervised learning technique of hierarchical clustering on gene expression data, Alizadeh et al. demonstrated that dif- fuse large B-cell lymphoma could be classified to two main
16 patterns – germinal centre and activated B cell. These
assignments were subsequently found to be predictive of response to treatment.

Question 3

There are 3 types of AI according to the article. What does the following mean?

• Reinforcement learning

Answer 3

Reinforcement learning is a newer class of learning and represents a hybrid of supervised and unsupervised learning. In reinforcement learning, the algorithm maximizes accuracy by trial and error. Feedback from the consequences of real
and stimulated decisions on the training set shape the model. Using reinforcement learning techniques, Komorowski et al. developed a model to recommend dynamic fluid and vasopressor management in patients admitted to the intensive care unit. 

These algorithms are inherently ‘data-hungry’, benefiting from comprehensive data across tens or hundreds of thousands of cases, until now limiting their application in the haematological arena.

Question 4

What are some advantages of AI? - 8

Answer 4

Reduction in human error

Takes risks instead of humans

Available 24/7

Helping in repetitve jobs

Digital assistance

Faster decisions

Daily applications

New inventions

Question 5

What are some disadvantages of AI? - 5

Answer 5

High costs of creation

Making humans lazy

Unemployment

No emotions

Lack of out of box thinking