Designing Value-sensitive AI

There has been a productive period of incredible progress and innovation in the field of AI.  AI developers have started to create sophisticated systems that mimic human intelligence and is already capable of learning in and coping with highly uncertain situations.

On the other hand, embedding ethical frameworks into AI technologies including robots is still not an easy task. Calculating the utility of every course of actions would be an impossible demand as a first-order logic is prevalent among AI technologies when being exposed to complicated decisions. Assigning values to human values would certainly be not a suitable method as well as integrating a moral framework without respect for important human values.

One of the most difficult questions to be answered is what it means to be moral. When it comes to developing such technologies, it is crucial to understand how morality differs from ethics. Should our approach to morality have its classification of right and wrong when it comes to AI technologies?

The development of AI technologies cannot be based merely on using empirical methods. One of the most complex aspects of designing these technologies is that they should be, first and foremost, as if they were moral beings. While the emotional decisions are not considered a necessity when it comes to making moral decisions, it is an important component because it often clouds our decision making and is something to take into account when empirically analyzing how humans and nonhumans interface with AI technologies, and perhaps how they reciprocate actions without having emotionally grounded social behaviors.

An ethical framework may not necessarily need to be a perfect approach that can solve any engineering issue or account for every possible future consequence, but it can surely help to minimize the effects of several issues before they arise.

In case of something going wrong, it can be modified and accommodated in future iterations. The development of such an ethical framework should start from the central premise that technology is not value-neutral. In other words, each technology has some values embedded that are of ethical importance to individuals and society such as freedom, equality, trust, autonomy, or privacy. As value-related issues are connected to the application of technology within a social context, an ethical framework should purport to incorporate value solutions into the design and address any issues that may emerge during the early design phases before ubiquitous rollout.

AI as both an emerging and converging technology will almost certainly entail the emergence of new ethical and societal issues, as well as the exacerbation of current issues associated with its development. Integrating an ethical framework with current practices could prove beneficial if the resulting amalgamation enhances existent practices:

  • Emerging technology should be sufficiently evaluated for design;
  • Such a methodology should also address the activities of designers and developers, for instance by taking into account potential surveillance techniques of AI designers to reduce the likelihood of an infeasible design.

Despite the difficulty of the development of a moral framework for AI, a focus on solely universally held values may also make such frameworks deviate from rights-based approaches to responsibility. Similar to conveying to children a concept of what is right or wrong, it is certainly critical to provide a rigid framework of dos and don’ts for AI. However, this should not be limited to a finite set of conditionals as this may result in a complicated mess of logic which breeds accusatory actions when something goes wrong.

Data Driven Investor’s Ayse Kok wrote this article. The original article is here.

The views and opinions expressed in this article are those of the author and do not necessarily reflect those of CDOTrends.