Trusting Data Science and Machine Learning

From Machine Learning
Jump to: navigation, search

Public awareness of the need for trust and privacy is ever increasing, with regards to both the online and offline worlds (the latter becoming increasingly small!).

"If we remain mindful of ethics then we can be sure, and the public can be reassured, that we are not moving towards Orwell's 1984." - from Public trust in data science from the ministry of defence.

How can I trust a machine?

Machines will do what they are programmed to do, and as such are neither trustworthy or untrustworthy. You can trust the results of an algorithm as far as you trust the person or organisation that created it.

Sometimes the way a machine is programmed means that it is inherently opaque and it is very difficult to discern how it has produced its results. In machine learning, neural networks are a type of "black box" algorithm where the interactions between nodes in the hidden layers are not easy to find out. In these situations, trust in the machine is still trust in the developer. A "sufficient understanding of how and why the output is produced"[1] is required on the developer's behalf, and trust should be based on an explanation of this understanding.

For all that the world is becoming increasingly digitised, and tasks becoming automated, the world of machines is a very human-centric one. The inputs and outputs are determined and interpreted by people, and machines are only as efficient/effective/accurate as the people that make and program them.

Catapult Network

The Digital Catapult has a publication on the implementation of a personal data receipt for consumer based applications. They have highlighted the fact that being transparent is one of the first steps towards being trusted.

Legalese

The ever increasing public conscience of the importance of proper data handling as well as the amount of trust being placed in automated processes and objects has made it necessary for procedures and legislation to be put in place to protect the people allowing access to their data and the people using "black box" machines.

Relevance of the GDPR

The GPDR (General Personal Data Regulation) as its name implies regulates the handling and storage of personal data. In the energy sector it is not often necessary to store individual's data, and so the regulations mostly do not apply.

The GDPR is a good example of well laid out, clear communication. It is accessible in simple language, something that is often missing in the technical world! When it comes to gaining trust, being clear and transparent is always the start and this page explaining the basic concepts of the GDPR is a good example of clear communication relating to the ever changing nature of personal data content and storage issues.

UK Government Guidelines for data handling

This 2019 guidance from the government provides a "Code of Conduct for data-driven health and care technology"[2]. The healthcare industry is one where being sensitive to data protection is a very high priority, and the ethical guidelines and responsibilities laid out can be applied to other industries where it is still important, but perhaps not as critical.

  • Understand users, their needs and context
  • Define the outcome and how the technology will contribute to it
  • Use data that is in line with appropriate guidelines for the purpose for which it is being used
  • Be fair, transparent and accountable about what data is being used
  • Make use of open standards
  • Show what type of algorithm is being developed or deployed
  • Generate evidence of effectiveness for the intended use and value for money
  • Make security integral to the design
  • Define the commercial strategy

References