The use of big data analytics in research (e.g. to develop decision aids for doctors) differs from traditional methods of research in several ways: ^(24,25)

(Click on each tab for further elaboration)

1. Source of data

Data may no longer be consciously provided by subjects, but automatically generated or observed (e.g. by sensors on devices, or through online cookies).

2. Amount of data

Maximum / excessive amount of data is collected in big data research (n = the whole population) as compared to the minimum amount necessary (n = a sample of the population). This is to maximise the possibility of finding useful correlations within the data.

3. Purpose of data collection and analysis

In some big data research, the aim of processing a set of data, and how the findings can be translated into practical use, may not be clear from the outset.

The initial purpose of data processing shifts from the traditional ‘hypothesis testing’ to ‘searching the data for any interesting correlations or findings’.

4. Sharing or disclosure of data

Big data researchers are more likely to outsource the complex task of data processing to third parties with expertise in AI. Data sharing and linking between organisations also become more common with the aim to create even bigger datasets.

5. Retention of data

Data is retained for much longer lengths of time due to the ever increasing capacity of storage systems (e.g. cloud storage).

ICO. Data Protection Act and General Data Protection Regulation: Big data, artificial intelligence, machine learning and data protection [Internet]. UK Information Commissioner’s Office; 2017. Available from: https://ico.org.uk/media/for-organisations/documents/2013559/big-data-ai-ml-and-data-protection.pdf

Balthazar P, Harri P, Prater A, Safdar NM. Protecting Your Patients’ Interests in the Era of Big Data, Artificial Intelligence, and Predictive Analytics. Journal of the American College of Radiology. 2018 Mar 1;15(3, Part B):580–6. Available from: http://www.sciencedirect.com/science/article/pii/S1546144017315995

What are the ethical challenges posed by research using big data and AI?

Feedback

Challenge 1

Informed consent

People may not realise that their data could be used in research when they provided them in the first place, even if they have checked a box to say ‘I have read and agreed to the terms and conditions’, because ‘that has been described as “the biggest lie on the web”’ (ICO, 2017, pg. 63).

Besides, when big data analytics is used for exploratory purposes rather than testing of a known hypothesis, researchers may not be able to tell subjects why they need to collect seemingly excessive amount of information, or how research findings (if any) may be used and what risks it would bring.

Challenge 2

Privacy and confidentiality

The long-term storage and frequent sharing of large quantities of data will increase the risks of unauthorised access.

The practice of data linking to create richer datasets increases the risk of re-identifying previously anonymised subjects.

Concerns are also raised that personal data may be sold or shared with third parties for purposes that have not been consented for.

Challenge 3

Potential harm of predictive analytics

Information within databases may contain errors and hidden human biases when they were recorded in the first place. If computers are ‘trained’ using inaccurate or biased datasets, their predictions may inherit such inaccuracies or biases.

However, validating these predictions becomes difficult when they result from machine learning that is too complex for human understanding. This is known as the ‘black box effect’ whereby we could not explain how / why a computer makes a particular conclusion (e.g. that someone is at high risk of breast cancer based on records of her sleep pattern).

‘Profiling’ occurs when predictions made by computers are translated to differential treatment of individuals. While accurate profiling can be a form of personalised or precision medicine, inaccurate profiling due to inaccurate predictions may lead to discrimination and harm.

The UK Information Commissioner’s Office (ICO) document on ‘Big data, artificial intelligence, machine learning and data protection’ (2017) provides further guidance on how privacy can be safeguarded in the era of big data, drawing interesting examples from various disciplines.