As with any technology, the use of Machine Learning in the analysis of big data comes with certain risks. Businesses will need to consider if and how they wish to mitigate these when investing in AI technology to enhance their productivity.
1. Data protection
Perhaps the biggest consideration when it comes to big data and the use of ML for analysis is the protection of personal data. This is particularly pertinent for European businesses, who will be subject to the more stringent General Data Protection Regulation (GDPR) from May 2018.
Using ML means the data held by companies can be repurposed and conclusions can be drawn about individuals that may not have been possible previously. This could in turn lead to unwelcome or discriminatory services. In a recent report on big data, for example, the FTC identifies an instance of people's credit limits being reduced because others who shopped at the same stores had poor repayment histories.
Firms collecting and storing large amounts of data should therefore consider the effect of inferences made by ML on the individuals and social groups concerned. Such consideration could be made in the form of a privacy impact assessment or by combining ML with human input to avoid unfairly targeting or impacting people.
2. Accumulation of technical debt
As systems that do not self-replicate or self-optimise, ML systems are prone to the accumulation of technical debt that cannot be solved in the traditional ways. Hidden debt can build up due to the complexities at system level, which can result not only in a series of undesirable outcomes such as entanglement and undeclared consumers. This means that it's surprisingly easy for system maintenance costs to build up.
Businesses looking to invest in ML must therefore understand the implications of technical debt and weigh up the risks of accruing moderate levels of such debt in order to progress. Employing competent software engineers, technical staff and data scientists is necessary to mitigate this risk.
3. Changes in the external world
Machine Learning is intriguing for its ability to interact with the external world. This world, however, is ever-changing and can become a source of technical debt. As an example, if a decision threshold is set manually and an update based on new data occurs, the existing threshold may become invalid. This might be manageable for one model and one threshold, but if there are many, manually revising them can prove time-consuming.
A data revolution?
Many businesses will be reviewing the pros and cons of ML in the coming months: Deloitte's 2016 Global CIO Survey showed that 64% of its respondents were looking to invest significantly in what the report calls 'Machine Intelligence' (MI) during 2017 and 2018. This suggests a more rapid uptake of the technology across a number of sectors as businesses compete to stay ahead.
This article is a sponsored post.
Discuss in our Forums
See what other readers are saying about this article!
Click here to read & post comments.
2 posts so far.