The study of the ethics of artificial intelligence is emerging as a hot topic of debate in computer and social sciences. The term Ethical AI refers to two related concepts, namely roboethics, the ethics of the humans involved in creating AI systems, and machine ethics, the study of the ethical behaviour of self-aware machines.

   Although the ethics of artificial intelligence is explored in works of science fiction, Business Intelligence technologies and human factors underpin the ethical learning and, therefore, the dilemmas that can arise from the initial conditions under which the programs and algorithms were developed

“The rise of powerful AI will either be the best, or the worst thing, ever to happen to humanity. We do not yet know which.” Stephen Hawking

   Initiatives (For example Machine Intelligence Research Institute or Full AI )aim to help ensure that smarter-than-human artificial intelligence has a positive impact on the world and that human self determination does not become limited by this intelligence. By bringing debate, insight, governance and guidelines there is a growing awareness of the ethical implications of Artificial Intelligence. As with many branches of technology, the aims of these bodies are simply stated, but the debates become clouded by the intricacies of the machines and software created as well as the power and political interests surrounding their introduction.

   In a similar way to the rise of Cybersecurity, where access flaws and concerns about control were studied for decades before the internet created the environmental conditions that led to the mistrust in pervasive computing environments, the rise of ethical concerns in Artificial Intelligence may seem to be a distant problem now, but the seeds of tomorrow’s computing concerns are being sown in the activities that we undertake today.

   The crux of many reported ethical problems in business computing is in the embedded inequality of outcomes. These are based in part on the unconscious bias of the creators of such systems. In applying brute machine learning logic to data, can deepen the inequalities, reinforce the assumptions, and embed in actions the unconscious biases of the system developers

Knowing the Future

There are different approaches to planning (“Knowing the Future”) on the two sides of the supplier-customer relationship.

Businesses will often take a resource based view on the data, i.e. which products over time perform best, and of these products, which markets did it sell into. The business then builds a model that will (usually) simply extrapolate these results using linear or log regression to apply the results to supposed future data sets. This is the essence of using simple AI to predict business planning based on the past performance. Businesses aim to increase their return on their data investment by promoting customer engagement.

On the other side, customers (and customer preferences) are subjective. As well as customer preferences that shift, customers themselves shift from demographic to demographic. Savvy customers will also be involved in their own ‘Big data’ initiatives, like using comparison websites to secure their own best deals and product searches. Customers seek to get the best deal for themselves, but also take into account affective and trust cues, even when dealing with the same products online.

This sets up a dynamic that a plan has to take into account not only the current state of the customer relationship, but also takes into account any previous behaviours without prejudice to the customer and arrive at an educated guess as to what they would like to do next. As data on purchases is fixed at the point of service supply in the transaction record which indicators and behaviours are likely to point to the likelihood of additional purchases? This is, after all, the point of the business forecasting exercise.

In a business context, the development of AI follows the model pathway detailed below:


   Each level of analysis adds additional interpretation. Adding transaction data to Business Intelligence uses groupings and characterisations of customers and customer types. Modified (‘wrangled’) data from the Business Intelligence layer is used to seed the Machine Learning algorithms and models, and these models are used to guide the AI rule judgements and decisions that the machines make on the business behalf.

   The key ingredient is asking the correct questions of the data that has been collected, and ensuring that the answer arrived at does not do so in a manner that has become biased by the operational inputs.


   The introduction of bias is generally made unconsciously. This can occur as a result of system programmers making assumptions about the types of customer they are grouping, by being embedded into the logic of the Machine Learning, or through the decision and rule algorithm outputs surfaced through the Artificial Intelligence systems. Human judgement, subjective, incorrect or biased data at any point in the process will affect the cognitive output of the Artificial Intelligence decision machine.

   Businesses should be wary of, and take action to ensure that the data they are working with are not being skewed through the analysis process.  In grouping and classifying it is necessary to ensure that the process is carried out within a governance and ethical structure to produce information and actions that are not prejudicial to different customer groups.