The Accountability of AI

Co-authored with Dr. Yvonne, Ai-Chi Loh

We have asked ourselves since the time of Plato how do we attribute moral responsibility for our own behaviour and how do we hold others and ourselves accountable for certain actions (Stanford Encyclopedia of Philosophy, 2019). Commonly, aversion to accountability is attributed to reasons such as limited involvement in decision making processes, fear of failure, and lack of confidence. This underlying understanding of human behaviour is important if we truly aspire to create a pragmatic and robust framework for AI accountability.

Do humans inherently seek or avoid accountability, and is our pursuit of more developed AI systems in truth a subconscious mask for shifting responsibility from the human to the AI.

In research involving slightly over 160 participants, researchers from North Carolina State University found that participants were more likely to blame the AI system for accidents if they thought that the robot was autonomous, i.e. not ultimately controlled by a human (Gurlough, Stokes & Gillan, 2019). These findings, while intuitive, present a disturbing reality. It has become convenient for humans to transfer culpability to a machine and as succinctly said by the study’s co-author, Douglas Gillan, AI may be “making it more difficult to hold human employees accountable” (Shipman, 2019).

Similarly, industry have been recognising the rising risks stemming from unclear accountability and have been increasing the call for tighter AI regulations as a proposed solution, stating that the current light touch approach is insufficient (Nuttall, 2020). Arguably, seemingly-morally cognizant tech bosses, such as Sundar Pichai, are increasingly aware of the techno-deterministic attitudes amongst technophiles to the point when developers and leaders are showing less and less social accountability (Hicks 2018).

We find that some jurisdictions are already proposing laws to hold individuals and companies accountable/responsible for harmful content produced by the algorithms of their information platforms (Bartlett, 2020). It is important to highlight, as identified in Yeung (2019), that any proposed measures need to attend to both the *prospective *and *historic *allocation of responsibility. In other words, while establishing obligations and duties associated with roles and tasks that look to the future, directed towards the production of good outcomes and the prevention of bad outcomes there is a need to concurrently look backwards and allocate responsibility for adverse conduct and events that occurred in the past.

One may challenge whether it is truly feasible to reap the benefits of AI while concurrently holding people accountable for the outcomes of these AI systems (Doshi-Velez et al, 2017). However, we continue to hold the view, as stated in Bartlett (2019), that ultimately humans cannot shift the accountability of their creation. Proposed accountability frameworks should encourage human adoption of accountability and thus be based on the a) philosophical theses of moral responsibility, as well as the b) social-psychological dimensions of responsibility attributions.

References

• Barlett, M. (2019),* Solving the Accountability Ga*p
• Broussard, M (2018), Artificial Unintelligence: How computers misunderstood the world. MIT Press
• Doshi-Velez, Finale & Kortz, Mason & Budish, Ryan & Bavitz, Chris & Gershman, Sam & O’Brien, David & Schieber, Stuart & Waldo, James & Weinberger, David & Wood, Alexandra. (2017). Accountability of AI Under the Law: The Role of Explanation. SSRN Electronic Journal. 10.2139/ssrn.3064761.
• Furlough, C., Stokes, T., & Gillan, D. J. (2019). Attributing Blame to Robots: I. The Influence of Robot Autonomy. Human Factors.
• Hicks, M. (2018). Fixing Tech’s Built-in Bias, American Scientist (106:5), p. 314.
• Nuttall, C. (2020),* Google chief calls for AI regulatio*n, The Financial Times London, UK
• Shipman, M. (2019), Darn You, R2!* When Do We Blame Robots*?
• Singapore Legal Advice (2019), Employers’ Vicarious Liability for Employee Negligence in Singapore.
• Stanford Encyclopaedia of Philosophy (2019),* Moral Responsibilit*y
• Yeung, K. (2019),* Responsibility and AI: Prepared by the Expert Committee on human rights dimensions of automated data processing and different forms of artificial intelligence (MSI-AUT)*, Council of Europe, September 2019