Building Trustworthy AI Systems

Building Trustworthy AI Systems
Talal Abu-Ghazaleh
As Artificial Intelligence (AI) systems proliferate across the world, the need to build such systems that are trustworthy is paramount in order to better promote safety, transparency and auditability, particularly as they start to play a greater role in helping humans with analysis and decision making. The potential this technology holds for  improving our lives is immense and  is already seeing greater adoption across many sectors. As with any technology, the potential for abuse also exists and many AI researchers have voiced their concerns about rapid, irresponsible development, which could harm rather than benefit.
Building trustable AI systems is essential as it starts to play a greater role in our lives with greater demands being placed on such systems from technologies such as robotics, 5G and IoT. The amount of big data generated from these systems can only fathomably be processed using intelligent AI machines to help us make sense of this information in a meaningful manner. The falling costs, availability and maturity of AI technologies is providing value that otherwise would be nearly impossible to derive using traditional methods as the variables and interactions are far too many for humans to analyze.
This in effect means that AI systems will help play a greater role in enabling us manage important systems and networks which must be done in a manner that is safe, reliable and that can be explained. For example, the logic as to why an AI system has taken the decision to intelligently reroute electricity in a power plant must be clear so that there is a proper audit trail. This obviously is more critical in settings such as healthcare or in the case of nuclear power plant management where the consequences could be catastrophic, but is also valuable to online marketers for example to know why an AI system has recommended a specific product to a certain user on the internet.
In order to build trustworthiness in AI systems, I see that there are three areas that must be focused on, namely:
•Developing explainable and safe AI systems
•Training AI systems using unbiased dataset
•Improving the security of AI technology
Allow me tackle each one of these areas in brief.
AI systems must be developed with explainability in mind. The increasing reliance on AI systems to analyze and process data to deliver better and faster insights will mean that they will come under scrutiny to ensure they are functioning correctly and transparently; which is the goal of explainable AI (XAI). This poses a challenge as AI systems are complex black boxes which really only a few  understand, and that  brings about the need to have the ability to backtrack and understand why AI systems came to the conclusion they did. 
As well as explainability, they should be developed with hard coded safety features to ensure that there are guarantees around human safety that cannot be violated regardless of the conclusions reached. 
The second part to the AI trustworthiness puzzles is the need to train such systems using vetted, unbiased information. Any AI system needs to be trained on sets of data in order to build correlations, which it then uses to makes predictions about future situations it is presented with. If AI systems are fed with datasets that are  biased or contain prejudice, the resulting outcome will be a system that reflects the same bias. Checks and balances need to be in place to ensure that critical AI applications in particular are trained with vetted, unbiased information.
The final part to achieve trustworthiness is to ensure that AI systems are properly secured using evolving AI based security systems. The field of cyberwarfare is rapidly changing and it would be disastrous to have such systems compromised. Malicious actors are becoming more sophisticated in their attacks and are building AI based attacks which we must have answers for. 
Building trustworthy AI systems must occur by governments working hand in hand with the private sector to properly understand this technology to develop national policies and capacity in this area so that a holistic ecosystem can be built in which this technology can thrive.