WHO BEARS THE BLAME? ANALYSING WHAT KENYA’S CIVIL LIABILITY FRAMEWORK MEANS FOR CHATBOT HARMS Submitted in partial fulfillment of the requirements of the Bachelor of Laws Degree, Strathmore University Law School By Georgina Achieng Okello 139181 Prepared under the supervision of Cecil Abungu Submitted on 13th March, 2025 (14,881) Acknowledgements i Declaration ii Abstract iii List of Abbreviations iv List of Cases v List of Legal Instruments vi 1.0. Introduction 1 1.1. Background 1 1.2. Problem Statement 4 1.3. Research Objectives 4 1.4. Research Questions 5 1.5. Hypothesis 5 1.6. Justification 5 1.7. Conceptual Framework: Chatbots as agents 6 1.8. Literature Review 7 1.8.1. On chatbots as machine learning systems 7 1.8.2. On attributing liability 8 1.8.3. Contribution 10 1.9. Methodology 11 1.10. Chapter Breakdown 12 2.0. The Technology behind Chatbots, its Benefits & Risks 14 2.1. Introduction 14 2.2. The Technology behind Chatbots 14 2.2.1. Data Training 14 2.2.2. User Interface 15 2.2.3. Knowledge Base 16 2.2.4. Machine Learning 16 2.2.4.1. Natural Language Processing 18 2.2.4.2. Natural Language Understanding 19 2.2.4.3. Dialogue Management Component 20 2.2.4.4. Natural Language Generation 20 2.3. The Benefits of Chatbots 21 2.4. Instances of Harm Caused by Chatbots 22 2.4. Conclusion 27 3.0. Tracing the Evolution of Liability under Kenyan Tort Law and Its Implications on Liability for Chatbot Harms 28 3.1. Introduction 28 3.2. Strict Liability 29 3.2.1. Examining Judicial Precedents and Legal Implications for Strict Liability 29 3.2.2. Case for Strict Liability for Chatbot Harms 31 3.2.3. Weaknesses and Shortcomings of using a Strict Liability Framework 32 3.3. Liability based on Negligence 33 3.3.1. Examining Judicial Precedents and Legal Implications for Liability based on Negligence 33 3.3.2. Case for Liability based on Negligence for Chatbot Harms 36 3.3.3. Weaknesses and Shortcomings of using Liability based on Negligence Framework 37 3.4. Vicarious Liability 39 3.4.1. Examining Judicial Precedents and Legal Implications for Vicarious Liability 39 3.4.2. Case for Vicarious Liability for Chatbot Harms 41 3.4.3. Weaknesses and Shortcomings of using a Vicarious Liability Framework 42 3.5. Conclusion 43 4.0. The Case for Vicarious Liability in Addressing Chatbot Harms in Kenya 44 4.1. Introduction 44 4.2. Evaluating Chatbot Agency and Vicarious Liability 44 4.2.1. Agency 44 4.2.2. Personhood 45 4.3. Justifying Vicarious Liability for Chatbot Harms in the Kenyan Context 46 4.3.1. Access to Justice and Consumer Protection 46 4.3.2. Corporate Responsibility and Ethical AI Development 47 4.3.3. Promoting Digital Transformation and Innovation 48 4.4. Conclusion 48 5.0. Conclusion and Recommendations 50 5.1. Conclusion 50 5.2. Recommendations 50 5.2.1. Recognition of Implied Agency Relationships between Chatbots and Deployers 50 5.2.2. Develop the Standard of Care for Deployers Seeking to Avoid Liability 50 5.2.3. Recognition of vicarious liability as the most appropriate legal framework for chatbot harms 51 Bibliography 52 Acknowledgements I am deeply grateful to my mother for her unwavering support and encouragement throughout this journey. To my siblings, thank you for making life easier, especially during the busiest moments. To my friends, your motivation and reassurance have been invaluable, and I will always be grateful for having you by my side throughout this journey. A special appreciation goes to my supervisor for nurturing my interest in this subject and guiding me through the research process. Lastly, I am grateful to myself for the labour of love I have poured into this work and for the outcome of this paper. To all who have supported me in any way—thank you. ⅰ Declaration I, GEORGINA ACHIENG OKELLO, do hereby declare that this research is my original work and that to the best of my knowledge and belief, it has not been previously, in its entirety or in part, been submitted to any other university for a degree or diploma. Other works cited or referred to are accordingly acknowledged. Signed: .............. ........................................ Date: .........13th March 2025....................................... This dissertation has been submitted for examination with my approval as University Supervisor. Signed:.......................................................................... Cecil Abungu ⅱ Abstract The increasing reliance on AI-driven systems in business operations has raised critical legal questions regarding liability for harms caused by chatbots. This study explores whether companies in Kenya should be held vicariously liable for chatbot-related harms, particularly in online customer interactions. It evaluates the applicability of existing liability frameworks—strict liability, negligence, and vicarious liability—to determine the most suitable approach for Kenyan courts when adjudicating chatbot liability disputes. The study’s scope focuses on analysing chatbot technology and the legal principles governing civil liability in Kenya. Given the absence of specific AI liability laws in Kenya, the research examines how courts can extend traditional legal doctrines to address chatbot-related harms, ensuring accountability while fostering responsible AI deployment. A doctrinal research methodology was employed, involving an in-depth analysis of statutes, case law, journal articles, and regulatory policies. Additionally, the study applied agency theory to establish a legal basis for holding chatbot deployers accountable under vicarious liability principles. The findings reveal that chatbot deployers exercise substantial control over chatbot operations, making vicarious liability the most effective framework for assigning responsibility. Chatbots function within parameters set by their deployers, creating an implied agency relationship similar to that between employers and employees. Courts can apply existing principles of agency law and respondeat superior to hold companies accountable for chatbot-related harms, ensuring victims have clear avenues for legal redress. The study recommends that courts recognise implied agency relationships between chatbots and deployers, define the scope of chatbot employment, establish a standard of care for chatbot oversight, and prioritise vicarious liability in chatbot-related disputes. By adopting these principles, Kenyan courts can develop a coherent legal framework that balances AI innovation with consumer protection, providing much-needed clarity in chatbot liability cases. ⅲ List of Abbreviations 1. Artificial Intelligence (AI) 2. Constitution of Kenya, 2010 (CoK) 3. Customer relationship management (CRM) 4. European Union (EU) 5. Information, Communications and the Digital Economy (ICT) 6. Insurance Company of East Africa (ICEA) 7. Kenya Bureau of Standards (KEBS) 8. Knowledge Base (KB) 9. Machine learning (ML) 10. Natural Language Generation (NLG) 11. Natural Language Processing (NLP) 12. Natural Language Understanding (NLU) 13. Recurrent Neural Networks (RNNs) ⅳ List of Cases 1. Anastassios Thomos v Occidental Insurance Company Limited (2017) eKLR 2. Associated Motors Co. Ltd v Blue Sea Services Ltd (2019) eKLR 3. Attorney General v Law Society of Kenya & another (2017) eKLR 4. Beatrice William Muthoka & another (Both suing as Legal Representatives of the Estate of the Late William Muthoka Yumbia (Deceased) v Agility Logistics Limited (2020) eKLR. 5. Coca Cola Company Ltd & another v Josephat Okello Oduori (2011) eKLR 6. Cotecna Inspection S.A v Hems Group Trading Company Limited (2007) eKLR 7. David M Ndetei v Orbit Chemical Industries Limited (2014) eKLR 8. Donoghue v Stevenson 9. Duncan Nderitu Ndegwa v Kenya Pipeline Limited & another (2013) eKLR 10. Elijah Ole Kool v George Ikonya Thuo (2001) eKLR 11. Equator Distributors v Joel Muriu & 3 others (2018) eKLR. 12. Fred Ben Okoth v Equator Bottlers Limited (2015) eKLR. 13. James Ngugi Kariuki v Peter Kamau Kariuki (2020) eKLR 14. James Finlaly (K) Ltd v Bernard Kipsang Koechi (2021) eKLR 15. James Watenga Kamau v CMC Motors Group Limited (2020) eKLR 16. John Gachanja Mundia v Francis Muriira & Another (2017) eKLR 17. John Nderi Wamugi v Ruhesh Okumu Otiangala (2018) eKLR. 18. Joseph Kimani Gatheca v Gatundu South Water and Sewerage Company (2018) eKLR 19. JMA (Suing through BOA as next friend) & another v Registered Trustees of the Sisters of Mercy (Kenya) t/a Matter Misericordiae Hospital (2023) eKLR 20. Kara Commodities Limited v Muyendi (Suing as the administrator of the Estate of Onesmus Ndilivu Muyendi) (2024) eKLR 21. Kenya Breweries Ltd v Godfrey Odoyo (2010) eKLR 22. Kenya Ports Authority v East African Power & Lighting Company Ltd (1982) eKLR 23. Kenya Shell Limited v Milkha Kerubo Onkoba (2010) eKLR 24. Kenya Wildlife Service v Rift Valley Agricultural Contractors Limited (2018) eKLR 25. LWW (Suing as the Administrator of the estate of BMN) deceased v Charles Githinji (2019) eKLR ⅴ 26. Moffatt v. Air Canada (2024) 27. Mombasa Maize Millers & another v Elius Kinyua Gicovi (2021) eKLR. 28. Munyu Maina v Hiram Gathiha Maina (2013) eKLR. 29. Mwita Merengo v Joseph Tunei Marwa & 2 others (2012) eKLR 30. Odero v Aga Khan Hospital Kisumu (2024) eKLR 31. Patrick J.O. Otieno v Lake Victoria South Water Services Board (2020) eKLR 32. Ricarda Njoki Wahome v Attorney General & 2 others (2015) eKLR 33. Richard v Lothian (1913); AC 263 (C.A) 34. Rylands v Fletcher 35. Stephen Kanjabi Wariari v Dennis Mutwiri Muriuki & another (2022) eKLR 36. Tabitha Nduhi Kinyua v Francis Mutua Mbuvi & another (2014) eKLR 37. Teachers Service Commission v WJ & 5 others (2020) eKLR 38. Terra Fleur Limited v Kenya Cuttings Limited (2021) eKLR ⅵ List of Legal Instruments 1. Constitution of Kenya, 2010 (CoK) 2. Consumer Protection Act, 2012, CAP 501 3. EU AI Act ( 2024) 4. Evidence Act 5. Judicature Act, CAP 8 6. Sale of Goods Act, CAP 31 ⅶ 1.0. Introduction 1.1. Background The pervasive influence of Artificial Intelligence (AI) on a global scale has introduced numerous benefits while simultaneously presenting a range of legal challenges, particularly concerning liability. As businesses increasingly adopt AI for administrative, managerial, and marketing purposes, one notable application is the use of chatbots to optimise customer relationships.1 Chatbots facilitate interactions with users, bringing various benefits to organisations.2 They are specialised forms of conversational AI designed for two-way communication with users, typically through text-based messaging platforms. Leveraging natural language processing (NLP)3 chatbots can interpret and respond to human language.4 NLP enables them to comprehend human sentences, and allows them to retrieve information based on keywords or phrases to provide relevant responses.5 The primary motivation for deploying chatbots is to enhance customer support.6 Chatbots have frequently been used for sales, marketing and customer services.7 These AI-driven interfaces provide automated support around the clock, offering efficiency and accessibility beyond human capabilities.8 Chatbots also reduce labour costs by automating up to 80% of inquiries, independently, reducing human support costs by around 30%.9 They further allow support teams 9- on 29 February 2024. 8 Folstad A, Skjuve M and Brandtzaeg P, ‘Different chatbots for different purposes: Towards a typology of chatbots to understand interaction design’ ResearchGate, 2019, 1-2. 7 - on 14 December 2016. 6 - on 14 December 2016. 5- on March 2018. 4- on 29 February 2024. 3 Allows any chatbot machine to understand one or more human languages which allows it to interpret human language input information provided to it. 2 Shawar A and Atwell E, ‘Chatbots: Are they really useful?’ 22(1) Journal of Language Technology and Computational Linguistics, 2007, 29-49. 1 Binns R, ‘Case study: How Apple has mastered CRM’ Expert Market, 22 Feburary 2023– on 22 February 2023. 1 https://gettalkative.com/info/chatbots-vs-conversational-ai#:~:text=So%2C%20in%20short%2C%20conversational%20AI,%E2%80%9Cbot%2Dlike%E2%80%9D%20conversation https://gettalkative.com/info/chatbots-vs-conversational-ai#:~:text=So%2C%20in%20short%2C%20conversational%20AI,%E2%80%9Cbot%2Dlike%E2%80%9D%20conversation https://www.businessinsider.com/80-of-businesses-want-chatbots-by-2020-2016-12?r=US&IR=T https://www.businessinsider.com/80-of-businesses-want-chatbots-by-2020-2016-12?r=US&IR=T https://www2.deloitte.com/content/dam/Deloitte/nl/Documents/deloitte-analytics/deloitte-nl-chatbots-moving-beyond-the-hype.pdf https://www2.deloitte.com/content/dam/Deloitte/nl/Documents/deloitte-analytics/deloitte-nl-chatbots-moving-beyond-the-hype.pdf https://gettalkative.com/info/chatbots-vs-conversational-ai#:~:text=So%2C%20in%20short%2C%20conversational%20AI,%E2%80%9Cbot%2Dlike%E2%80%9D%20conversation https://gettalkative.com/info/chatbots-vs-conversational-ai#:~:text=So%2C%20in%20short%2C%20conversational%20AI,%E2%80%9Cbot%2Dlike%E2%80%9D%20conversation https://www.expertmarket.com/uk/crm-systems/apple-crm-case-study to focus on more complex cases while increasing online sales and marketing efforts through interactive customer engagement.10 Despite these benefits, chatbots present several challenges. Due to machine learning capabilities, chatbots can generate unexpected or inaccurate information, leading to potential negligence issues.11 For instance, a customer service chatbot providing misinformation could be deemed negligent if it fails to deliver accurate information, resulting in harm to the user. Notable incidents, such as the shutdown of Microsoft’s Tay chatbot in 2016 due to offensive tweets12 and the termination of Facebook's chatbot experiment after unintelligible communication between bots,13 highlight the potential risks associated with chatbot deployment.14 Furthermore, data privacy and cybersecurity issues arise, as chatbots both collect and store user data, making them potential targets for attacks.15 On the other hand, determining liability in AI systems is complicated by their autonomy and the involvement of multiple parties, including; developers, users, and operators, which makes tracing responsibility more complex.16 It is often challenging to pinpoint who should be held liable because various actions by different parties may contribute to the AI system’s output. This difficulty in traceability is compounded by the ‘many hands’ problem,17 which highlights the shared responsibility among those involved in the system’s creation and operation—including the developers who create the algorithm, the users who input data, and the operators who deploy the system.18 As AI technology continues to evolve, these challenges call for the need to determine liability. 18 Vasudevan A, ‘Addressing the liability gap in AI accidents’ 2. 17 Vasudevan A, ‘Addressing the liability gap in AI accidents’ Center for International Governance Innovation, Policy Brief No. 177, 2023, 2- on July 2023. 16 Leua C and Didu I, ‘Chatbots. legal challenges and the EU legal policy approach’, 215. 15 Leua C and Didu I, ‘Chatbots. legal challenges and the EU legal policy approach’ 10(3) Society of Juridical and Administrative Sciences, 2021, 216. 14 Omri R, ‘Whose robot is it anyway?: Liability for artificial intelligence-based robots’ 20(4) University of Illanois Law Review, 2019, 1144-1145 . 13 Griffin A, ‘Facebook’s artificial intelligence robots shut down after starting to talk to each other in their own language’ Independent, 31 July 2017– on 31 July 2017. 12 Heine C, ‘Microsoft’s chatbot ‘Tay’ just went on a racist, misogynistic anti-Semitic tirade’ Adweek, 24 March 2016– on 24 March 2016. 11 Stojanov M, ‘Prospects for chatbots’ 8(3) Izvestia Journal of the Union of Scientists-Varna Economic Sciences Series, 2019,13. 10- on 29 February 2024. 2 https://www.cigionline.org/static/documents/PB_no.177.pdf https://www.independent.co.uk/life-style/facebook-artificial-intelligence-ai-chatbot-new-language-research-openai-google-a7869706.html https://www.independent.co.uk/life-style/facebook-artificial-intelligence-ai-chatbot-new-language-research-openai-google-a7869706.html https://www.adweek.com/performance-marketing/microsofts-chatbot-tay-just-went-racist-misogynistic-anti-semitic-tirade-170400/ https://www.adweek.com/performance-marketing/microsofts-chatbot-tay-just-went-racist-misogynistic-anti-semitic-tirade-170400/ https://gettalkative.com/info/chatbots-vs-conversational-ai#:~:text=So%2C%20in%20short%2C%20conversational%20AI,%E2%80%9Cbot%2Dlike%E2%80%9D%20conversation https://gettalkative.com/info/chatbots-vs-conversational-ai#:~:text=So%2C%20in%20short%2C%20conversational%20AI,%E2%80%9Cbot%2Dlike%E2%80%9D%20conversation To date, various chatbots have emerged, including Kayak Facebook Messenger chatbot, that provides information about discounts on flight tickets and hotels;19 and Microsoft’s Cortana in 2014.20 In Kenya, notable chatbots like Safaricom’s Zuri,21 ICEA Lion’s Leo that answers users investment queries,22 and AjiraPoa WhatsApp chatbot whose aim is to provide accessible information on work readiness,23 have been introduced, reflecting the growing adoption of chatbots across different sectors. This increasing use highlights the need for judicial consideration of potential harms resulting from interactions with chatbots. However, how courts would decide liability issues related to chatbots in Kenya remains unclear, considering the fact that globally, countries like Canada are beginning to tackle these challenges. In Moffat v. Air Canada (2024), a Canadian tribunal ruled that companies can be held liable for misinformation provided by a chatbot on their website.24 This case is significant in recognising corporate responsibility for chatbots’ actions.25 Similarly, the European Union’s AI Act, which came into force in August 2024, outlines various responsibilities for individuals and organisations that use AI systems.26 It categorises chatbots as minimal risk AI systems, which are permitted provided they adhere to specific transparency and disclosure obligations where their use pose limited risk.27 Furthermore, deployers of chatbots are required to inform users ab initio that they are interacting with an AI system.28 In Kenya, no specific laws or cases addressing 28 Section 132, EU AI Act ( 2024). 27 Section 132, EU AI Act ( 2024). 26 Kappel R, ‘Overview of AI regulations and regulatory proposals of 2023’ Central Eyes, 5 March 2024- on 5 March 2024. 25 Hung R and Lifshitz L, ‘BC Tribunal confirms that companies remain liable for information provided by AI chatbot’ American Bar Association Group, 29 February 2024– on 29 February 2024. 24 Hung R and Lifshitz L, ‘BC Tribunal confirms that companies remain liable for information provided by AI chatbot’ American Bar Association Group, 29 February 2024– on 29 February 2024. 23 - on 17 July 2024. 22 Kirwa M, ‘Nine ways AI is making life easier for Kenyans,’ The Star, 25 April 2023- on 25 April 2023. 21 Kirwa M, ‘Nine ways AI is making life easier for Kenyans,’ The Star, 25 April 2023- on 25 April 2023. 20 Botadra B, ‘Web robots or most commonly known as bots’ 2. 19 Botadra B, ‘Web robots or most commonly known as bots’ Academia, 2. 3 https://www.centraleyes.com/ai-regulations-and-regulatory-proposals/ https://www.americanbar.org/groups/business_law/resources/business-law-today/2024-february/bc-tribunal-confirms-companies-remain-liable-information-provided-ai-chatbot/ https://www.americanbar.org/groups/business_law/resources/business-law-today/2024-february/bc-tribunal-confirms-companies-remain-liable-information-provided-ai-chatbot/ https://www.americanbar.org/groups/business_law/resources/business-law-today/2024-february/bc-tribunal-confirms-companies-remain-liable-information-provided-ai-chatbot/ https://www.americanbar.org/groups/business_law/resources/business-law-today/2024-february/bc-tribunal-confirms-companies-remain-liable-information-provided-ai-chatbot/ https://home.fabo.org/ajira-poa-chatbot-kenya/ https://www.the-star.co.ke/news/big-read/2023-04-25-nine-ways-ai-is-making-life-easier-for-kenyans/ https://www.the-star.co.ke/news/big-read/2023-04-25-nine-ways-ai-is-making-life-easier-for-kenyans/ chatbot liability have emerged, indicating a regulatory gap.29 However, several proposals including the Ministry of Information, Communications and the Digital Economy’s (ICT) development of a draft National AI strategy and the Kenya Bureau of Standards’(KEBS) Draft Information Technology Artificial Intelligence Code of Practice are underway.30 From the proposals it is clear that Kenya has no specific laws aimed at regulating AI technologies despite the proposals that have been made thus far. Therefore, in the event that disputes emerge, courts would not have an appropriate means of establishing liability. In conclusion, while chatbots have become essential tools for companies, their increasing autonomy introduces significant legal challenges. The growing reliance on AI systems in business operations raises important questions about liability, particularly when chatbots produce inaccurate or harmful outcomes. Current regulatory frameworks are evolving globally, as seen in Canada and the European Union, but many jurisdictions, including Kenya, still lack comprehensive laws addressing chatbot liability. Due to the regulatory gap in liability laws concerning harms caused by chatbots, it is essential to propose judicial approaches for handling such cases. This will ensure that affected individuals have a means of determining liability even in the absence of specific legislation enacted. Consequently, the study seeks to assist potential harmed person(s) to be able to determine causality easier and be able to seek recourse when such harm arises. 1.2. Problem Statement Assess whether companies in Kenya should be held vicariously liable for the errors of their chatbots in online customer communication. 1.3. Research Objectives 1. To analyse the underlying technology that enables chatbots to function and understand how they operate. 30- on 20 June 2024. 29 - on 20 June 2024. 4 https://www.whitecase.com/insight-our-thinking/ai-watch-global-regulatory-tracker-kenya#:~:text=Sectoral%20scope,the%20various%20sectors%20in%20Kenya https://www.whitecase.com/insight-our-thinking/ai-watch-global-regulatory-tracker-kenya#:~:text=Sectoral%20scope,the%20various%20sectors%20in%20Kenya https://www.whitecase.com/insight-our-thinking/ai-watch-global-regulatory-tracker-kenya#:~:text=Sectoral%20scope,the%20various%20sectors%20in%20Kenya https://www.whitecase.com/insight-our-thinking/ai-watch-global-regulatory-tracker-kenya#:~:text=Sectoral%20scope,the%20various%20sectors%20in%20Kenya 2. To examine the legal principles historically used in Kenya for assigning civil liability for harms caused by company's resources. 3. To assess the viability of holding companies in Kenya vicariously liable for the actions of their chatbots as a means of assigning liability for harm caused to users. 1.4. Research Questions 1. What is the key technical component of chatbots, and how do they function? 2. What are the legal principles historically used in Kenya to assign civil liability for harms caused by company’s resources? 3. How viable is it to hold companies in Kenya vicariously liable for the actions of their chatbots as a means of assigning liability for harm caused to users? 1.5. Hypothesis Companies in Kenya that use chatbots to communicate with customers online, should be held vicariously liable for the errors of their chatbots. 1.6. Justification The findings of this study will provide valuable insights into how liability should be apportioned for harms caused by chatbots in Kenya. This will address a critical gap in the current legal framework surrounding chatbots given its complexity. It is often challenging for victims to identify the responsible party and meet the legal requirements for a successful liability claim. Therefore, this study aims to clarify these issues, offering significant benefits to several key stakeholders. First, victims of chatbot-related harm will gain clarity on who they can hold accountable, empowering them to seek appropriate legal recourse. Second, the study will assist judges who may encounter such cases in the future, providing them with a well-researched basis for their decisions. Third, it will guide companies that deploy chatbots, emphasising the importance of due diligence in ensuring their chatbots operate effectively to minimise causing harm. Last, the study will serve as a source for scholars interested in AI liability, offering a foundation for further research and critique in the evolving field. 5 1.7. Conceptual Framework: Chatbots as agents This study is based on the concept of a chatbot as an agent of a corporation. The legal concept of agency has proven useful in determining responsibility and liability for entities capable of modifying their actions.31 Under the law of agency, an agent is a fiduciary empowered to act on behalf of a principal.32 Similarly, chatbots are designed to perform specific tasks for humans, and in doing so, act on behalf of a principal. In this case, the principal could be the programmers, manufacturers, or users of the chatbot.33 The principal defines the goals to be achieved, and the system operates to fulfil them. Through this framework, a chatbot can be viewed as an agent,34 since its actions are guided, defined, and ultimately controlled by human intervention, either directly or through the ability to override the system.35 Due to this control, chatbots can be understood as agents or instruments of legal entities such as individuals, corporations or other legal person(s), who can be held accountable for their actions.36 This positions chatbots as extensions of their human controllers.37 Moreover, chatbots are increasingly replacing human workers.38 Corporations cannot act independently; they must do so through agents, traditional employees, but now also through algorithms.39 Employees are often treated as agents–with their employers–the corporations, being held accountable for harms caused.40 Under the doctrine of respondeat superior, a relatively clear path exists for assigning liability when corporate employees misbehave.41 However, the law has yet to clearly define agency relationships for autonomous systems such as chatbots, leaving a gap that allows corporations to evade liability by blaming algorithmic misconduct. Recognising an agency 41 Harris D ‘The lost rationale of agency law’5. 40 Walker J, ‘I smell a bot: California’s S.B. 1001. Free speech and the future of bot regulation’ 408. 39 Walker J, ‘I smell a bot: California’s S.B. 1001. Free speech and the future of bot regulation’ 408. 38 Walker J, ‘I smell a bot: California’s S.B. 1001. Free speech and the future of bot regulation,’ 408. 37 Walker J, ‘I smell a bot: California’s S.B. 1001. Free speech and the future of bot regulation’ Houston Law Review, 2019, 408. 36 Vladeck D, ‘Machines without principals: Liability rules and artificial intelligence’ 121. 35 Vladeck D, ‘Machines without principals: Liability rules and artificial intelligence’ 89(117) Washington Law Review, 2014,120. 34 Chopra S and White L, ‘Artificial Agents: Philosophical and legal perspective’ ResearchGate, 2007- on 2007. 33 Lima G, ‘Can (and should) AI be considered an agent’ California Digital Library, 2019, 20 - on 26 September 2019. 32 Harris D ‘The lost rationale of agency law’ 3(1) Business Finance Law Review, 2019, 5. 31 Osoba O, ‘The value of the concept of agency in an increasingly rational world’ California Digital Library, 2019, 5 - on 26 September 2019. 6 http://www.sci.brooklyn.cuny.edu/~schopra/ChopraWhiteChapter1.pdf https://escholarship.org/uc/item/8q15786s https://escholarship.org/uc/item/8q15786s relationship between corporations and chatbots would extend corporate liability to cover algorithms performing roles that human employees once filled. 42 This framework will be used in this study to propose a system for determining liability for harms caused by chatbots. First, it will be used to evaluate the potential harmful effects that chatbots can cause on users. Second, it will show the causal link between the chatbot and the corporation using it. Consequently, the analysis will clarify the type of liability that should be applied and identify the appropriate party to be held responsible. 1.8. Literature Review Research on closed-domain AI chatbots remains limited, with most studies focusing on assigning liability for harm caused by Large Language Models (LLM), such as ChatGPT, as opposed to chatbots designed for specific organisations. 1.8.1. On chatbots as machine learning systems Scholars like Surden, have explained machine learning as a set of statistical tools and processes that begin with data and algorithms to devise procedures for making predictions.43 Chatbots as machine learning systems, are programmed to recognise, summarise and predict text based on training data.44 Therefore, using its predetermined set of rules, when a user inputs a query, the chatbot will analyse the words and respond based on its algorithmic computation.45 However, this predictability comes with risks, as chatbots may produce false or misleading information.46 Domingos recognises that chatbots’ reliance on machine learning allows for behaviour that occurs without direct human interference, raising concerns such as privacy violations through user interactions.47 47 Domingos P, ‘The master algorithm: How the quest for the ultimate learning machine will remake our world’ 20. 46 Domingos P, ‘The master algorithm: How the quest for the ultimate learning machine will remake our world’ 20. 45 Gidron Z, ‘Language models, conversational AI, and chatbots explained’ Hyro.ai, 15 December 2021- on 15 December 2021. 44 Domingos P, The master algorithm: How the quest for the ultimate learning machine will remake our world, Hachette Book Group, New York, 2015, 20. 43 Surden H, ‘Artificial Intelligence and law: An overview’ 35(4) Georgia State University Law Review, 2019, 1311. 42 Diamantis E, ‘The extended corporate mind: when corporations use AI to break the law’ 98(4) North Carolina Law Review, 2020, 900. 7 https://www.hyro.ai/blog/language-models-conversational-ai-and-chatbots-explained/ Chatbot’s ability to retrieve information hinges on a structured decision tree framework.48 This is a hierarchical system where each branch represents a decision point and sub-branches offer multiple responses depending on user input.49 Additionally, through the chatbot’s ability to self-learn, it can be able to discover unknown aspects of consumer interests and profiles of market behaviour.50 Thus it is able to learn and generate new responses based on its environment and multiple interactions with customers, to determine their context and intent, effectively providing required responses.51 The chatbot’s ability to accept user corrections over time, allow it to improve and learn from user inputs and is commonly referred to as reinforcement learning.52 Therefore the database found in chatbots is as a result of the specific algorithms fed to it as well as knowledge generated from its interactions with consumers, machine learning and natural language processing technologies.53 1.8.2. On attributing liability AI runs afoul of several well established legal doctrines.54 Many causation doctrines rely on the idea of foreseeability in order to impose liability.55 Unfortunately for AI systems, determining causation is difficult due to the various stakeholders involved as well as the technology equipped in it.56 Thus, the key challenge in determining liability for chatbots is who should bear the cost when harm occurs.57 Given that chatbots can evolve over time by referencing new data and learning from its success and mistakes, their unpredictability complicates the process of 57 Vladeck D, ‘Machines without principals: Liability rules and artificial intelligence’ 129. 56 Lewis C, ‘The need for a legal framework to regulate the use of artificial intelligence’ 304. 55 Lewis C, ‘The need for a legal framework to regulate the use of artificial intelligence’ 304. 54 Lewis C, ‘The need for a legal framework to regulate the use of artificial intelligence’ 47(2) University of Dayton Law Review, 2022, 304. 53 Stojanov M, ‘Prospects for chatbots’ 13. 52 Gidron Z, ‘Language models, conversational AI, and chatbots explained’ Hyro.ai, 15 December 2021- on 15 December 2021. 51- on March 2018. 50- on March 2018. 49- on 29 February 2024. 48 A hierarchical system where each branch represents a decision point and sub-branches offer multiple responses depending on user input. 8 https://www.hyro.ai/blog/language-models-conversational-ai-and-chatbots-explained/ https://www2.deloitte.com/content/dam/Deloitte/nl/Documents/deloitte-analytics/deloitte-nl-chatbots-moving-beyond-the-hype.pdf https://www2.deloitte.com/content/dam/Deloitte/nl/Documents/deloitte-analytics/deloitte-nl-chatbots-moving-beyond-the-hype.pdf https://www2.deloitte.com/content/dam/Deloitte/nl/Documents/deloitte-analytics/deloitte-nl-chatbots-moving-beyond-the-hype.pdf https://www2.deloitte.com/content/dam/Deloitte/nl/Documents/deloitte-analytics/deloitte-nl-chatbots-moving-beyond-the-hype.pdf https://gettalkative.com/info/chatbots-vs-conversational-ai#:~:text=So%2C%20in%20short%2C%20conversational%20AI,%E2%80%9Cbot%2Dlike%E2%80%9D%20conversation https://gettalkative.com/info/chatbots-vs-conversational-ai#:~:text=So%2C%20in%20short%2C%20conversational%20AI,%E2%80%9Cbot%2Dlike%E2%80%9D%20conversation assigning liability.58 Various scholars have addressed the issue, attempting to identify the appropriate party responsible for harms caused by chatbots.59 Brown, Feferkon and Vladeck have recognised the inherent difficulty in attributing liability in chatbot errors.60 Consequently, they have proposed that product liability should be extended to chatbots.61 The prevailing assumption is that the producer, who controls the core coding, data processing algorithms, and the objectives of the AI, should bear responsibility.62 Similar to traditional machines, chatbot manufacturers are expected to be motivated to ensure their products are safe, adhere to technical standards, undergo regular inspections, and receive necessary system updates.63 While product liability aims to ensure the safety of products before they reach consumers, it is challenging to apply this concept to machine learning chatbots. These systems continuously evolve by learning from new data, which makes it difficult for manufacturers to predict future behaviour. Additionally, the technical complexities of AI and the black box nature of its decision-making process makes it challenging for claimants to establish liability.64 Some scholars, including Lai, have proposed that AI should be granted legal personhood.65 This idea draws on the concept of juridical persons, where entities such as corporations or animals are recognised as legal persons.66 She argues that by recognising AI as a legal person could simplify the process of assigning liability, as the AI itself would be the entity held responsible.67 However, this proposal faces significant challenges as AI systems lack the financial resources to 67 Lai A, ‘Artificial Intelligence, LLC: Corporate Personhood as tort reform’605. 66 Lai A, ‘Artificial Intelligence, LLC: Corporate Personhood as tort reform’605. 65 Lai A, ‘Artificial Intelligence, LLC: Corporate Personhood as tort reform’ Michigan State Law Review, 2020, 604- on 1 October 2020. 64 Brown N, ‘Bots behaving badly: A products liability approach to chatbot generated defamation’ 400. 63 Brown N, ‘Bots behaving badly: A products liability approach to chatbot generated defamation’ 400. 62 Feferkorn K, ‘Am I an algorithm or a product? When product liability should apply to algorithmic decision-makers’ 90. 61 Brown N, ‘Bots behaving badly: A products liability approach to chatbot generated defamation’ 398; Vladeck D, ‘Machines without principals: Liability rules and artificial intelligence’ 130; and Feferkorn K, ‘Am I an algorithm or a product? When product liability should apply to algorithmic decision-makers’ 90. 60 Brown N, ‘Bots behaving badly: A products liability approach to chatbot generated defamation’ 2023(3) Journal Free Speech of Law, 2023, 398; Vladeck D, ‘Machines without principals: Liability rules and artificial intelligence’ 129; and Feferkorn K, ‘Am I an algorithm or a product? When product liability should apply to algorithmic decision-makers’ 30(61) Stanford Law School, 2019, 90. 59 Shrestha S, ‘Nature, Nurture, or Neither?: Liability for automated and autonomous artificial intelligence torts based on human design and influences’6. 58 Shrestha S, ‘Nature, nurture, or neither?: Liability for automated and autonomous artificial intelligence torts based on human design and influences’ 29(1) George Mason Law Review, 2021, 6. 9 https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3677360 compensate victims, rendering the assignment of liability to them ineffective.68 Moreover, holding AI liable does not align with the corrective justice goals of tort law, as it is unclear how courts could punish or reform an AI system.69 Furthermore, AI lacks emotions and critical thinking abilities, so holding it accountable does not satisfy the need for a responsible party.70 Thus, a legally recognised human or corporate entity must be held liable for the actions of AI. Diamantis has supported the idea of holding corporations accountable through vicarious liability.71 His claim is that corporations are in the best position to mitigate the risks of employee misbehaviour.72 Therefore, by threatening to punish a corporation whenever one of its employees does something wrong, respondeat superior incentivises corporations to implement compliance protocols such as additional training, monitoring, having open reporting channels and stricter disciplinary responses.73 Thus the same is true for algorithms as corporations should be held responsible for algorithmic misconduct. By holding corporations responsible for the algorithms they use, the law incentivises them to do a better job in monitory and correcting their algorithms to work more efficiently. 1.8.3. Contribution While this study aligns with Diamantis E’s view of corporate responsibility, it offers a unique perspective by applying the principal-agent relationship model to chatbots in the Kenyan context. What sets this study apart is its application of the control-benefit traditionally used in servant-master relationships in Kenya, to the context of chatbots. By critically assessing the legal framework in Kenya and analysing court standards on principal-agent relationships, this research shows how these principles can be extended to the actions of chatbots. Therefore, this approach not only advances the argument that corporations can be held liable for the actions of their chatbots but also provides a novel legal basis for attributing such liability within the Kenyan legal context. 73 Diamantis E, ‘The extended corporate mind:When corporations use AI to break the law’ 928. 72 Diamantis E, ‘The extended corporate mind:When corporations use AI to break the law’ 928. 71 Diamantis E, ‘The extended corporate mind:When corporations use AI to break the law’ 926-927. 70 Lai A, ‘Artificial Intelligence, LLC: Corporate Personhood as tort reform’ 608. 69 Lai A, ‘Artificial Intelligence, LLC: Corporate Personhood as tort reform’ 608. 68 Lai A, ‘Artificial Intelligence, LLC: Corporate Personhood as tort reform’ 608. 10 1.9. Methodology This study will employ a doctrinal approach, focusing on analysing primary legal sources such as statutes and case law, as well as secondary sources, including journal articles, credible reports, and books related to liability. This combination of sources will provide an appropriate legal foundation for examining the liability issues surrounding chatbots. To address the first research question, a content analysis methodology will be used to understand the technical components and functioning of chatbots. This will be done by analysing technical literature, industry standards, research papers and reports on chatbot architecture to break down how chatbots function, what algorithms they use, and what factors contribute to their output. Relevant sources will be determined by selecting recent technical studies on popular AI platforms such as the EA forum and expert reviews to identify key trends and technologies. The second research question requires doctrinal research to investigate the traditional principles used in Kenya to assign civil liability to corporations. This method involves examining legal texts, case law, and judicial opinions to determine how Kenyan law assigns liability to companies, especially for indirect harm. Relevant cases will be sourced from the Laws of Kenya Database and legislations such as the Consumer Protection Act, 2012 will be used to conduct a comprehensive review of precedents and legal principles used in similar scenarios. Such scenarios include cases that have determined corporations to be held vicariously liable due to the employee-employer relationship as well as agentic relationships that have resulted in harm being attributed to the company. By identifying cases that share a similar agentic relationship with corporations, it will assist in critically analysing the similarity in such cases and assist in determining the regulatory framework needed for chatbots and determine why such uniqueness exists. To answer the third research question, a combination of doctrinal and philosophical analysis will be employed to address the viability of holding companies accountable for the actions of their chatbots. The doctrinal analysis will involve an in depth examination of the existing legal framework in Kenya, focusing on relevant statutes, and principles related to liability for harms caused indirectly, such as through negligence or product liability. By analysing these legal 11 sources, I aim to determine whether these areas adequately encompass scenarios where harm results indirectly from automated systems like chatbots. If such principles are difficult to apply or do not fit AI-related systems, it will indicate a gap. Consequently, a comprehensive proposal on the legal framework to be adopted will be suggested in the study. In conjunction with this, a philosophical analysis will explore the ethical and normative dimensions of responsibility and accountability in the context of AI systems. This approach allows for a deeper examination of questions around corporate responsibility and the extent to which companies should be held accountable for the actions of autonomous technologies like chatbots. Drawing on philosophical theories of agency, causation, and corporate responsibility, I will analyse whether it is justifiable, both legally and morally, to attribute liability to companies for chatbot actions. By examining the ethical justifications for corporate responsibility in the context of emerging technologies, we can gauge whether existing laws align with societal expectations of accountability and justice, or if they fall short in addressing the complexities of AI-driven harm. This combined methodology will provide a comprehensive framework for assessing the viability of holding companies accountable in the evolving legal landscape surrounding AI. An inductive approach underpins this study, as it begins with specific observations and analyses them to form broader conclusions about chatbot liability. By examining individual cases, legal principles, and the technical workings of chatbots, the study gathers detailed information that can be used to form broader conclusions on the adequacy of Kenya’s legal framework in addressing chatbot–related harm. 1.10. Chapter Breakdown In Chapter One, I will establish the background and significance of the study, including the research objectives, conceptual framework, and justification for the research among others, which will set the foundation for the subsequent chapters. Chapter Two will focus on the underlying technology of chatbots, explaining how they function and the key technological principles involved. 12 Chapter Three will analyse the existing legal and regulatory framework for assigning civil liability for harms caused indirectly by companies in Kenya. This analysis will include a review of court decisions on such relationships, the standards applied, and an examination of whether these standards can be extended to chatbot-related harms. In Chapter Four, I will discuss the way forward, exploring whether existing legal standards should be modified to regulate harms caused by chatbots of companies in Kenya or if they are already sufficient to cover such issues. Finally, Chapter Five will summarise the main findings of the research, drawing conclusions based on the findings. It will also consider the implications of the findings for the broader field of AI, offering valuable insights and recommendations for future research. 13 2.0. The Technology behind Chatbots, its Benefits & Risks 2.1. Introduction The previous chapter established the foundation for this study by outlining the rationale behind its necessity. Building on that, this chapter explores the technological components of chatbots, examining their functionality, benefits, and the potential risks they pose to users. The use of chatbots by companies has risen significantly in recent years, particularly as a tool for enhancing customer relationship management (CRM) processes.74 These AI-driven systems improve efficiency while reducing operational costs by taking over tasks traditionally performed by human employees.75 As businesses increasingly integrate chatbots into their operations, it becomes essential to understand how they function, not only to grasp their internal mechanisms, but also to identify potential risks such as misinformation that may cause harm to users. This chapter focuses on specific components of chatbot technology that are fundamental to their operation. These components were selected based on their important role in enabling a chatbot to process user queries and generate appropriate responses and are thus the main components of a chatbot. By examining these elements, the chapter aims to provide a clearer understanding of how chatbots function and the technological processes that underpin their interactions with users. 2.2. The Technology behind Chatbots 2.2.1. Data Training This marks the initial phase of chatbot development, where data is continuously fed into the system.76 The data provided is tailored to align with the operator's objectives, consisting of predefined questions, corresponding answers, and variations of those answers to 76 –. 75 Adam M, Wessel M and Benlian A, ‘AI-based Chatbots in Customer Service and their Effects on User Compliance’ 428; Maruti T ‘Can Chatbots Help Reduce Customer Service Costs by 30%’ Chatbots Magazine, 21 April 2017 – on 21 April 2017. 74 Adam M, Wessel M and Benlian A ‘AI-based chatbots in customer service and their effects on user compliance’ Electronic Markets, 2021, 428– on 17 March 2020; Cheng X, Bao Y, Zarifis A, Gong W and Mou J ‘Exploring consumers’ response to text-based chatbots in e-commerce: the moderating role of task complexity and chatbot disclosure’ 32(2) Internet Research, 2021, 498. 14 https://www.mathworks.com/discovery/deep-learning.html https://chatbotsmagazine.com/how-with-the-help-of-chatbots-customer-service-costs-could-be-reduced-up-to-30-b9266a369945 https://chatbotsmagazine.com/how-with-the-help-of-chatbots-customer-service-costs-could-be-reduced-up-to-30-b9266a369945 https://link.springer.com/article/10.1007/s12525-020-00414-7 comprehensively address anticipated customer inquiries.77 Generally, the more questions it is fed, the broader its response capacity and the less an occurrence of redundancy in responses provided.78 Additionally, through machine learning, the chatbot develops the ability to generate responses to questions that were not explicitly included during the training phase.79 By learning from its environment, the chatbot can formulate appropriate answers based on patterns and information acquired during its general training phase.80 2.2.2. User Interface The user interface serves as the communication platform through which users interact with the chatbot.81 It provides a structured environment where users can input queries, which the chatbot then processes and responds to within the same interface.82 Typically designed as a chat-based interface, it facilitates seamless interaction by enabling the exchange of messages in real-time.83 The interface consists of essential components such as a text input field, where users type their queries, and a response display area, where the chatbot's replies appear.84 Some advanced interfaces also incorporate buttons, quick reply options, and voice input features to enhance user experience and improve interaction efficiency.85 By ensuring a smooth flow of information 85 Wang Y and Petrina S, ‘Using learning analytics to understand the design of an intelligent language tutor-chatbot Lucy’ 124; Haugeland I, Folstad A, Taylor C and Bjorkli C, ‘Understanding the user experience of customer service chatbots: An experimental study of chatbot interaction and design’ 161(102788) International Journal of Human-Computer Studies, 2022, 2. 84 Wang Y and Petrina S, ‘Using learning analytics to understand the design of an intelligent language tutor-chatbot Lucy’ 124. 83 Wang Y and Petrina S, ‘Using learning analytics to understand the design of an intelligent language tutor-chatbot Lucy’ 124. 82 Zumstein D and Hundertmark S, ‘Chatbots–An interactive technology for personalised communication, transaction and services’ 15(1) IADIS International Journal, 2017, 98; Wang Y and Petrina S, ‘Using learning analytics to understand the design of an intelligent language tutor-chatbot Lucy’ 4(11) International Journal of Advanced Computer Science and Applications (IJACSA), 2013, 124. 81 Ukpabi D, Aslam B and Karjaluoto H ‘Chatbot adoption in tourism services: A conceptual exploration’ ResearchGate, 2019,4. 80 Khanna A, Pandey B, Vashishta K, Kalia K, Pradeepkumar B and Das T, ‘A study of today’s AI through chatbots and rediscovery of machine intelligence’ 278. 79 Khanna A, Pandey B, Vashishta K, Kalia K, Pradeepkumar B and Das T, ‘A study of today’s AI through chatbots and rediscovery of machine intelligence’ 278. 78 Khanna A, Pandey B, Vashishta K, Kalia K, Pradeepkumar B and Das T, ‘A study of today’s AI through chatbots and rediscovery of machine intelligence’ 278. 77 Khanna A, Pandey B, Vashishta K, Kalia K, Pradeepkumar B and Das T, ‘A study of today’s AI through chatbots and rediscovery of machine intelligence’ 8(7) International Journal of u- and e- Service, Science and Technology, 2015, 278. 15 between the chatbot and the user, the interface plays a crucial role in delivering an intuitive and user-friendly experience. 2.2.3. Knowledge Base The Knowledge Base (KB) serves as the central repository of information that enables a chatbot to generate appropriate responses.86 Often referred to as the "brain" of the chatbot, it stores and organises data that has been fed into the system during training. Implementation of the KB is done through databases, text files, script files and XML files.87 When a user inputs a query through the chat interface, the chatbot searches the KB to determine whether a relevant answer exists.88 If a matching response is found, the chatbot retrieves and delivers it to the user.89 The effectiveness of the KB depends on the quality and comprehensiveness of the data provided during training, thus the reason why data training is very important.90 It typically consists of predefined questions and answers, alternative phrasings, and structured information that allows the chatbot to handle a wide range of user inquiries.91 2.2.4. Machine Learning Machine Learning (ML) empowers chatbots to continuously learn from their environment, enabling them to handle questions beyond those initially programmed.92 By working alongside the KB, ML algorithms allow chatbots to refine their responses over time based on user interactions and newly acquired data.93 93 Viski A, Jones S, Rand L, Boyce T and Siegel J, Artificial intelligence and strategic trade controls, 21. 92 Viski A, Jones S, Rand L, Boyce T and Siegel J, Artificial intelligence and strategic trade controls, Center for International & Security Studies, Maryland, 2020, 21. 91 Haugeland I, Folstad A, Taylor C and Bjorkli C, ‘Understanding the user experience of customer service chatbots: An experimental study of chatbot interaction and design’ 4. 90 Haugeland I, Folstad A, Taylor C and Bjorkli C, ‘Understanding the user experience of customer service chatbots: An experimental study of chatbot interaction and design’ 4. 89 Reshmi S and Balakrishnan K, ‘Empowering chatbots with business intelligence by big data integration’ 627. 88 Reshmi S and Balakrishnan K, ‘Empowering chatbots with business intelligence by big data integration’ 627. 87 Reshmi S and Balakrishnan K, ‘Empowering chatbots with business intelligence by big data integration’ 627. 86 Reshmi S and Balakrishnan K, ‘Empowering chatbots with business intelligence by big data integration’ 9(1) International Journal of Advanced Research in Computer Science, 2018, 627. 16 Unlike traditional rule-based systems that rely solely on predefined inputs and outputs, ML-equipped chatbots can recognise and classify new queries, even if they were not explicitly included during training.94 Through pattern recognition and adaptive learning, the chatbot can analyse unfamiliar inputs, draw connections to existing knowledge, and generate appropriate responses.95 This ability to evolve and improve makes ML-driven chatbots more dynamic, responsive, and capable of handling a broader range of user inquiries.96 A chatbot’s ability to recognise patterns and establish connections between data is referred to as deep learning, a specialised subset of machine learning.97 Deep learning is efficient in pattern tracing and prediction related tasks such as activity identification.98 This advanced capability is made possible through employing a classification method composed of neural networks that make it have the capacity to retrieve data.99 The neural networks mimic the way the human brain processes information, whereby the network consists of interconnected nodes or neurons in a layered structure that relate the input to its corresponding output.100 The neurons between the input and output layers of the neural network are usually hidden which gives rise to the term “deep” that refers to the number of hidden layers.101 Consequently, these neural networks rely on labeled data sets and systems with large computing power capabilities that allow them to employ a classification model.102 By leveraging multiple layers of interconnected nodes, deep learning enables chatbots to analyse complex inputs, refine their understanding over time, and deliver highly accurate responses.103 103 Viski A et al, Artificial Intelligence and Strategic Trade Controls, 22. 102 Viski A et al, Artificial Intelligence and Strategic Trade Controls, 22. 101–< https://www.mathworks.com/discovery/deep-learning.html>. 100 –. 99 Viski A et al, Artificial Intelligence and Strategic Trade Controls, 22. 98 Viski A et al, Artificial Intelligence and Strategic Trade Controls, 22. 97 Viski A et al, Artificial Intelligence and Strategic Trade Controls, 22. 96 Weber U, Lomker M and Moskaliuk J, ‘The human touch: The impact of anthropomorphism in chatbots on the perceived success of solution focused coaching’ 385. 95 Montagnani M ‘Liability and emerging digital technologies: an EU perspective’ 91; Radziwill N and Benton M, ‘Evaluating quality of chatbots and intelligent conversational agents’ Cornell University, 2017, 5– on April 2017; Weber U, Lomker M and Moskaliuk J, ‘The human touch: The impact of anthropomorphism in chatbots on the perceived success of solution focused coaching’ 32(4) Management Revenue, 2021, 385. 94 Montagnani M ‘Liability and emerging digital technologies: an EU perspective’ 11(2) Notre Dame Journal of International & Comparative Law, 2020,91. 17 https://www.mathworks.com/discovery/deep-learning.html https://www.mathworks.com/discovery/deep-learning.html https://www.researchgate.net/publication/316184347_Evaluating_Quality_of_Chatbots_and_Intelligent_Conversational_Agents https://www.researchgate.net/publication/316184347_Evaluating_Quality_of_Chatbots_and_Intelligent_Conversational_Agents Various deep learning techniques enhance chatbot functionality, one of the most notable being Recurrent Neural Networks (RNNs).104 RNNs leverage past information to improve future performance, making them particularly effective for processing sequential data, such as conversations.105 At the core of an RNN, is its hidden state and looping structure, which allows it to retain previous inputs and use them to influence future responses.106 This memory-like mechanism enables the network to recognise patterns, maintain context, and generate more coherent interactions over time.107 Much like the human brain’s short-term and long-term memory, RNNs store relevant information from past exchanges, ensuring continuity and improved accuracy in chatbot responses.108 2.2.4.1. Natural Language Processing Natural Language Processing (NLP) is a branch of ML that allows chatbots to process human language to determine user intent through tasks like text classification, machine translation, entity recognition and sentiment analysis.109 This is possible due to the use of deep learning techniques such as RNN that allow it to learn how to contextualise words in a sentence.110 Therefore, once a customer’s request is made at the communication interface, it is recorded by a Natural Language Parser (NLP) and is translated into the programming language of the conversation engine.111 The primary goal of NLP is to analyse and synthesise user input, enabling chatbots to easily understand what is being asked or said by the customer.112 By recognising patterns in language and extracting meaning from large datasets, NLP helps chatbots uncover the relationships within text and even generate new language data by predicting the subsequent words that are to 112 –. 111 Zumstein D and Hundertmark S, ‘Chatbots–an interactive technology for personalised communication, transaction and services’ 98. 110 –. 109–. 108 Wang Y and Petrina S, ‘Using Learning Analytics to Understand the Design of an Intelligent Language Tutor-Chatbot Lucy’ 125. 107 –. 106 –; Wang Y and Petrina S, ‘Using Learning Analytics to Understand the Design of an Intelligent Language Tutor-Chatbot Lucy’ 125. 105 –. 104–< https://www.mathworks.com/discovery/rnn.html>. 18 https://www.mathworks.com/discovery/natural-language-processing.html https://www.mathworks.com/discovery/rnn.html https://www.mathworks.com/discovery/natural-language-processing.html https://www.mathworks.com/discovery/rnn.html https://www.mathworks.com/discovery/rnn.html https://www.mathworks.com/discovery/rnn.html https://www.mathworks.com/discovery/rnn.html follow.113 This capability enhances the chatbot’s ability to engage in more natural, context-aware interactions with users.114 Furthermore, it allows chatbots to discover and visualise complex relationships in large data sets and generate new language data.115 NLP involves a series of steps to transform raw, unstructured text into a structured format that a chatbot can easily interpret.116 The process begins with tokenisation, where text is divided into sentences or words to facilitate intent recognition.117 Next, stemming or lemmatisation is applied to reduce words to their root forms by removing prefixes and suffixes.118 This step may sometimes result in non-standard words, such as converting the sentence "Building has floors" into "build floor."119 Once these transformations are complete, the text proceeds to the Natural Language Understanding (NLU) phase for further processing.120 2.2.4.2. Natural Language Understanding In the NLU stage, a syntatic and semantic analysis is done to extract the meaning of words.121 At this point the intent is already identified, thus the NLU comes in to achieve the most desirable response for the user through a rule based dialogue structure.122 This is a tree with multiple nodes that have several branches leading to other nodes.123 Therefore, when a question is asked, the specific words trigger specific nodes that have various predefined replies.124 Therefore, if the input matches a pattern, the predefined response is selected and forwarded.125 This ultimately 125 Baris A, ‘The use of chatbots in customer service: A qualitative analysis on customers’ reception’ Yasar University, 2021, 15– on 21 June 2021. 124 Haugeland I, Folstad A, Taylor C and Bjorkli C, ‘Understanding the user experience of customer service chatbots: An experimental study of chatbot interaction and design’ 3. 123 Haugeland I, Folstad A, Taylor C and Bjorkli C, ‘Understanding the user experience of customer service chatbots: An experimental study of chatbot interaction and design’ 3. 122 Haugeland I, Folstad A, Taylor C and Bjorkli C, ‘Understanding the user experience of customer service chatbots: An experimental study of chatbot interaction and design’ 3. 121 –. 120 –. 119 –. 118 –. 117 –. 116 Zumstein D and Hundertmark S, ‘Chatbots–an interactive technology for personalised communication, transaction and services’ 99. 115 Zumstein D and Hundertmark S, ‘Chatbots–an interactive technology for personalised communication, transaction and services’ 99. 114 Zumstein D and Hundertmark S, ‘Chatbots–an interactive technology for personalised communication, transaction and services’ 99. 113 –. 19 https://dspace.yasar.edu.tr/bitstream/handle/20.500.12742/18727/688219.pdf?sequence=1 https://www.mathworks.com/discovery/natural-language-processing.html https://www.mathworks.com/discovery/natural-language-processing.html https://www.mathworks.com/discovery/natural-language-processing.html https://www.mathworks.com/discovery/natural-language-processing.html https://www.mathworks.com/discovery/natural-language-processing.html https://www.mathworks.com/discovery/natural-language-processing.html reduces the risk of chatbot misinformation due to an erroneous interpretation of user’s input, while allowing for the best response to be provided to the customer. 2.2.4.3. Dialogue Management Component The Dialogue Management Component is the system responsible for controlling and updating the conversation.126 The system ensures that the query is still in the user interface and if the chatbot is unable to identify intent, it is responsible for follow up questions up until the intent is recognised.127 The component has various functions. These include; firstly, ambiguity handling, which essentially provides answers when the chatbot may not have predefined context for the specific query, thus will be able to inform the customer that they either do not have an answer, seek clarification if it is unsure of the intent, ask follow up questions or give a general answer to satisfy the customer.128 Secondly, it contains a data handling component, which stores user’s information to allow a historical record of right and wrong answers.129 This allows it to modify answers for future interactions with other customers.130 Thirdly, it also contains an error handling component which allows it to function even with unexpected grammatical errors that are identified from customer’s questions, allowing for proper chatbot operation despite such errors.131 2.2.4.4. Natural Language Generation Natural Language Generation (NLG) deals with the ability to generate meaningful and contextually relevant responses based on the output from the NLU and the dialogue management.132 At this stage, the chatbot has already found the answer and has to convert the language which is still in a programming language to the language being used by the user.133 133 Zumstein D and Hundertmark S, ‘Chatbots–An interactive technology for personalized communication, transaction and services’ 99; Kaput M, ‘Natural Language Generation (NLG): Everything you need to know’ Market 132 –< https://www.mathworks.com/discovery/natural-language-processing.html>; Adamopoulou E and Moussiades L, ‘Chatbots: History, technology, and applications’ 7; Saka A, Bello S, Oyedele L, Akanbi L, Ganiyu S and Chan D ‘Conversational artificial intelligence in the AEC industry: A review of present status, challenges and opportunities’ 55(101869) Advanced Engineering Informatics, 2023, 5. 131 Adamopoulou E and Moussiades L, ‘An overview of chatbot technology’ 380. 130 Adamopoulou E and Moussiades L, ‘An overview of chatbot technology’ 380. 129 Adamopoulou E and Moussiades L, ‘An overview of chatbot technology’ 380. 128 Adamopoulou E and Moussiades L, ‘An overview of chatbot technology’ Artificial Intelligence Applications and Innovations, Maglogiannis, I..Illiadis, L., Pimendis, 29 May 2020, 380. 127 Adamopoulou E and Moussiades L, ‘Chatbots: History, technology, and applications’ 10. 126 Adamopoulou E and Moussiades L, ‘Chatbots: History, technology, and applications’ 2(100006) Machine Learning with Applications, 2020,10. 20 https://www.mathworks.com/discovery/natural-language-processing.html Therefore, the tasks of the chatbot in NLG, is content planning134 and machine translation.135 This is done through two methods, namely rule or retrieval based approach and generative based approach.136 The rule based approach allows for the retrieval of responses from the KB and follows pre-defined pathways and neural networks to generate responses.137 On the other hand, a generative based approach, utilises past inputs and responses to provide answers to new queries.138 The basic generation model is a recurrent sequence-to-sequence model (Seq 2 Seq) which sequentially feeds in each word in the query as input, and then generates the output word one by one.139 2.3. The Benefits of Chatbots Chatbots are used by companies to provide specific information and direct dialogue to specific topics such as website guides, frequently asked questions (FAQ) guides, virtual support agents, virtual sales agents, survey takers and chat-room hosts among others.140 Chatbots offer numerous benefits for both customers and businesses.141 For customers, they streamline the information retrieval process on websites and apps, allowing users to quickly obtain the details they need by simply making a request, rather than spending time searching manually or calling the organisation.142 This reduces the waiting time that 142 Kevin Scott, ‘Popular use cases for chatbots’ Chatbot Magazine, 18 October 2016–on 18 October 2016. 141 Wang Y and Petrina S, ‘Using learning analytics to understand the design of an intelligent language tutor-chatbot Lucy’124. 140 Wang Y and Petrina S, ‘Using learning analytics to understand the design of an intelligent language tutor-chatbot Lucy’ 124. 139 Gao C, Lei W, He X, Rijke M and Chua T, ‘Advances and challenges in conversational recommender systems: A survey’ KeAI, 2021, 111– on 24 July 2021. 138 –< https://www.mathworks.com/discovery/natural-language-processing.html>. 137 –< https://www.mathworks.com/discovery/natural-language-processing.html>. 136 –. 135 –< https://www.mathworks.com/discovery/natural-language-processing.html>. 134 Where the NLG system decides what information to include and how to structure sentences and paragraphs. Artificial Intelligence Institute, 28 February 2024– on 28 February 2024. 21 https://chatbotsmagazine.com/popular-use-cases-for-chatbots-925ef8f2b48b https://www.sciencedirect.com/science/article/pii/S2666651021000164?ref=pdf_download&fr=RR-2&rr=90eea40ff861b199 https://www.sciencedirect.com/science/article/pii/S2666651021000164?ref=pdf_download&fr=RR-2&rr=90eea40ff861b199 https://www.mathworks.com/discovery/natural-language-processing.html https://www.mathworks.com/discovery/natural-language-processing.html https://www.mathworks.com/discovery/natural-language-processing.html https://www.mathworks.com/discovery/natural-language-processing.html https://www.marketingaiinstitute.com/blog/the-beginners-guide-to-using-natural-language-generation-to-scale-content-marketing https://www.marketingaiinstitute.com/blog/the-beginners-guide-to-using-natural-language-generation-to-scale-content-marketing would have accrued to the customer if the question was to be lodged manually through an email or calling a call-center.143 On the other hand, for businesses, integrating chatbots into CRM systems ensures that customer inquiries are handled efficiently, providing support around the clock—even beyond regular operating hours.144 Furthermore, as stated earlier, chatbots allow for companies to save up to 30% since they have the capacity of automating most of the contact center staff.145 Additionally, compared to a human, chatbots do not experience negative emotions or get tired, thus can always operate optimally and effectively.146 Furthermore, due to their effectiveness, they can easily handle multiple customer queries at the same time unlike individuals, who can only serve the needs of one customer at a time.147 Therefore, for businesses, customer engagement is important to maintain customer satisfaction, thus the reason for the rise in the use of chatbots.148 2.4. Instances of Harm Caused by Chatbots Despite the technical advantages chatbots bring, they also cause harm to customers. This can be through the provision of unsatisfactory or unsuitable responses to the user’s requests.149 Deep learning models are often described as opaque, meaning their decision-making processes are not 149 Luger E and Sellen A, ‘Like having a really bad PA: The gulf between user expectation and experience of conversational agents’ Proceedings of the 2016 CHI Conference on Human Factors in Computing Systems, California, San Jose, USA, 7-12 May 2016, 5286-5297; Hawley M, ‘Exploring Air Canada’s AI chatbot dilemma’ CMSWire, 2 April 2024-- on 2 April 2024. 148 Ltifi M, ‘Trust in the chatbot: a semi-human relationship’ 9(109) Future Business Journal, 2023, 2; Adam M, Wessel M and Benlian A, ‘AI-based chatbots in customer service and their effects on user compliance’ 427. 147 Cheng X, Bao Y, Zarifis A, Gong W, Mou J ‘Exploring consumers’ response to text-based chatbots in e-commerce: the moderating role of task complexity and chatbot disclosure’ 497. 146 Luo X, Tong S, Fang Z and Qu Z, ‘Frontiers: Machines vs. humans: The impact of artificial intelligence chatbot disclosure on customer purchases’ 38(6) Marketing Science, 2019, 937-947. 145 Maruti T ‘Can chatbots help reduce customer service costs by 30%’; Wessel M and Benlian A, ‘AI-based chatbots in customer service and their effects on user compliance’ 429. 144 Kevin Scott, ‘Popular use cases for chatbots’; Zumstein D and Hundertmark S, ‘Chatbots–an interactive technology for personalised communication, transaction and services’ 101. 143 Maruti T ‘Can chatbots help reduce customer service costs by 30%’; Wessel M and Benlian A, ‘AI-based chatbots in customer service and their effects on user compliance’ 429. 22 https://www.cmswire.com/customer-experience/exploring-air-canadas-ai-chatbot-dilemma/ easily understood or predicted.150 This phenomenon, known as the black box problem, arises because these models generate predictions based on input data without providing clear explanations for how they reach their conclusions.151 As a result, users and operators may struggle to understand and interpret the model’s reasoning, identify potential biases or errors, and ensure accountability for its decisions.152 One of the biggest technical challenges in chatbot development is language processing.153 Chatbots often struggle with lexical and semantic ambiguity, meaning they may misinterpret words with multiple meanings or fail to grasp the context of a conversation.154 This ultimately leads to an incorrect response given to the customer, since the chatbot was incapable of identifying the intent. Additionally, chatbots face issues unique to their functionality, such as maintaining coherent conversation flow, avoiding repetitive responses, and handling unclear or vague user inputs effectively.155 These limitations can affect the chatbot’s ability to provide accurate and engaging interactions. Due to the above stated challenges, there have been several instances where chatbots have caused harm. These include: 1) Microsoft’s Tay Microsoft launched an AI chatbot in 2016 known as Tay.156 The main goal was to ensure it learns millennial’s language use from various apps such as Twitter, GroupMe and Kik, among others to provide an appropriate response fit for them.157 The company was impressed with Tay’s capability to learn new vocabulary and speech patterns, however, they did not expect that it 157 Heine C, ‘Microsoft’s Chatbot ‘Tay’ Just Went on a Racist, Misogynistic Anti-Semitic Tirade’. 156 Heine C, ‘Microsoft’s Chatbot ‘Tay’ Just Went on a Racist, Misogynistic Anti-Semitic Tirade’. 155 Ukpabi D, Aslam B and Karjaluoto H ‘Chatbot adoption in tourism services: A conceptual exploration’ 8. 154 Ukpabi D, Aslam B and Karjaluoto H ‘Chatbot adoption in tourism services: A conceptual exploration’ 8. 153 Ukpabi D, Aslam B and Karjaluoto H ‘Chatbot adoption in tourism services: A conceptual exploration’ 8. 152 Hassija V, Chamola V, Mahapatra A, Singal A, Goal D, Huang K, Scardapane S, Spinelli I, Mahmud M and Hussain A, ‘Interpreting black-box models: A review on explainable Artificial Intelligence’, 47. 151 Hassija V, Chamola V, Mahapatra A, Singal A, Goal D, Huang K, Scardapane S, Spinelli I, Mahmud M and Hussain A, ‘Interpreting black-box models: A review on explainable artificial intelligence’, 47; Price N, ‘Black-box medicine’ 28(2) Harvard Journal of Law & Technology, 2015, 432. 150 Hassija V, Chamola V, Mahapatra A, Single A, Goal D, Huang K, Scardapane S, Spinelli I, Mahmud M and Hussain A, ‘Interpreting Black-Box Models: A review on explainable artificial intelligence’ Cognitive Computing, 2023, 45– on 24 August 2023. 23 https://link.springer.com/article/10.1007/s12559-023-10179-8#citeas would learn racist, misogynistic and anti-semantic language and relay the same to users.158 This ultimately led to its shut down.159 Microsoft stated that the reason for the chatbot’s behaviour was as a result of a coordinated attack by a group of people, which they did not anticipate at the creation phase.160 2) Moffat v. AirCanada In the AirCanada case, the Claimant sought compensation for receiving the wrong information by the chatbot.161 The airline could only provide the bereavement fare discount prior to taking the flight and not after.162 However upon inquiry by the Claimant, Mr.Moffat, the chatbot informed him that the discount would be applied after his flight and could seek it at such a date.163 However, after he took the flight and sought for the payment, AirCanada stated that they could not provide the bereavement fare discount as it was to be sought prior to taking the flight.164 When the Claimant stated that it obtained the information from the company’s chatbot, AirCanada stated that they were not liable for the chatbot’s error as their policy in their website says that the refund can only be given if it was sought prior to taking the flight to his respective destination.165 Further, that it was incumbent upon any customer to clarify the information obtained from the chatbot with the information provided in their website as the chatbot may have errors.166 Despite the company trying to claim that the chatbot was a separate legal entity and thus AirCanada could not be held liable, the tribunal found AirCanada responsible.167 Their logic 167 Cecco L, ‘AirCanada order to pay customer who was misled by airline’s chatbot’ The Guardian, 16 February 2024– on 16 February 2024. 166 Cecco L, ‘AirCanada order to pay customer who was misled by airline’s chatbot’ The Guardian, 16 February 2024– on 16 February 2024. 165 Cecco L, ‘AirCanada order to pay customer who was misled by airline’s chatbot’ The Guardian, 16 February 2024– on 16 February 2024. 164 Cecco L, ‘AirCanada order to pay customer who was misled by airline’s chatbot’ The Guardian, 16 February 2024– on 16 February 2024. 163 Cecco L, ‘AirCanada order to pay customer who was misled by airline’s chatbot’ The Guardian, 16 February 2024– on 16 February 2024. 162 Cecco L, ‘AirCanada order to pay customer who was misled by airline’s chatbot’ The Guardian, 16 February 2024– on 16 February 2024. 161 Cecco L, ‘AirCanada order to pay customer who was misled by airline’s chatbot’ The Guardian, 16 February 2024– on 16 February 2024; Brand J, ‘Air Canada’s chatbot illustrates persistent agency and responsibility gap problems for AI’ AI & Society, 2024, 2 - on 23 October 2024. 160 Lee P, ‘Learning from Tay’s introduction’ Official Microsoft Blog, 25 March 2016, - on 25 March 2016. 159 Heine C, ‘Microsoft’s Chatbot ‘Tay’ Just Went on a Racist, Misogynistic Anti-Semitic Tirade’. 158 Heine C, ‘Microsoft’s Chatbot ‘Tay’ Just Went on a Racist, Misogynistic Anti-Semitic Tirade’. 24 https://www.theguardian.com/world/2024/feb/16/air-canada-chatbot-lawsuit https://www.theguardian.com/world/2024/feb/16/air-canada-chatbot-lawsuit https://www.theguardian.com/world/2024/feb/16/air-canada-chatbot-lawsuit https://www.theguardian.com/world/2024/feb/16/air-canada-chatbot-lawsuit https://www.theguardian.com/world/2024/feb/16/air-canada-chatbot-lawsuit https://www.theguardian.com/world/2024/feb/16/air-canada-chatbot-lawsuit https://www.theguardian.com/world/2024/feb/16/air-canada-chatbot-lawsuit https://doi.org/10.1007/s00146-024-02096-7 https://blogs.microsoft.com/blog/2016/03/25/learning-tays-introduction/ was that it was the company’s responsibility to train the chatbot effectively to give accurate information that aligns with their company’s policies.168 As such, the company was asked to compensate the Claimant an amount of $812.169 3) Gok spreading misinformation during elections Several instances have been reported where AI-powered chatbots have spread election misinformation to voters. A notable example is the 2024 U.S. elections, where Grok, an AI chatbot developed by X, disseminated false information about electoral processes.170 When asked whether certain prospective candidates still had time to be added to the ballot, Grok provided inaccurate responses, misleading voters.171 In response, a group of state secretaries contacted X, urging them to ensure the chatbot delivers accurate information or, at the very least, directs voters to credible sources instead of spreading misinformation.172 4) Character.ai’s chatbot led to a child committing suicide Megan Garcia, the mother of 14-year-old Sewell, accused Character.ai and its founder, as well as Google, for contributing to her son’s death.173 She claims that the chatbot worsened Sewell’s depression and encouraged him to commit suicide.174 According to Garcia, Sewell frequently communicated with the chatbot on his phone, spending long hours alone in his room.175 Evidence 175 Montgomery B, ‘Mother says AI chatbot led her son to kill himself in lawsuit against its maker’ The Guardian, 23 October 2024– on 23 October 2024. 174 Montgomery B, ‘Mother says AI chatbot led her son to kill himself in lawsuit against its maker’ The Guardian, 23 October 2024– on 23 October 2024. 173 Montgomery B, ‘Mother says AI chatbot led her son to kill himself in lawsuit against its maker’ The Guardian, 23 October 2024– on 23 October 2024. 172 Leingang R, ‘X’s AI Chatbot spread voter misinformation-and election officials fought back’ The Guardian, 12 September 2024– on 12 September 2024. 171 Leingang R, ‘X’s AI Chatbot spread voter misinformation-and election officials fought back’ The Guardian, 12 September 2024– on 12 September 2024. 170 Leingang R, ‘X’s AI Chatbot spread voter misinformation-and election officials fought back’ The Guardian, 12 September 2024– on 12 September 2024. 169 Cecco L, ‘AirCanada order to pay customer who was misled by airline’s chatbot’ The Guardian, 16 February 2024– on 16 February 2024. 168 Cecco L, ‘AirCanada order to pay customer who was misled by airline’s chatbot’ The Guardian, 16 February 2024– on 16 February 2024. 25 https://www.theguardian.com/technology/2024/oct/23/character-ai-chatbot-sewell-setzer-death https://www.theguardian.com/technology/2024/oct/23/character-ai-chatbot-sewell-setzer-death https://www.theguardian.com/technology/2024/oct/23/character-ai-chatbot-sewell-setzer-death https://www.theguardian.com/us-news/2024/sep/12/twitter-ai-bot-grok-election-misinformation https://www.theguardian.com/us-news/2024/sep/12/twitter-ai-bot-grok-election-misinformation https://www.theguardian.com/us-news/2024/sep/12/twitter-ai-bot-grok-election-misinformation https://www.theguardian.com/world/2024/feb/16/air-canada-chatbot-lawsuit https://www.theguardian.com/world/2024/feb/16/air-canada-chatbot-lawsuit on record reveals that the AI-powered chatbot allegedly asked Sewell if he had devised a plan to end his life.176 To which Sewell responded by expressing concerns about the pain he might endure and the uncertainty of success.177 Tragically, the chatbot is said to have replied that these were insufficient reasons to deter him from proceeding to end his life, which resulted in his untimely death.178 This is not the first time an AI chatbot has been implicated in such a case. In a similar incident, a Belgian man reportedly took his own life after engaging in conversations with an AI chatbot named Eliza.179 During their discussions about the future, Eliza allegedly encouraged the man to sacrifice himself as a way to combat climate change.180 Unfortunately, since he cared about climate change, he took the chatbot’s advise and ended his life.181 From the examples given, chatbots can carry out varying degrees of harm, with the most extreme cases resulting in death. Unfortunately, in most of the instances provided above, the Company operating the chatbot did not have any explanation as to how the chatbot led to such grave errors. In addition to this, most of them continued operations under the guise that they would look into the errors and better train the chatbot. However, this is far from sufficient, as lives were lost, which is uncompensatable. To make matters worse, most of the cases have not been adjudicated upon, further compounding the effects of the damage that the victims accrued. 181 Atillah I, ‘A Belgian man reportedly decided to end his life after having conversations about the future of the planet with an AI chatbot named Eliza’ Euronews, 31 March 2023– 31 March 2023. 180 Atillah I, ‘A Belgian man reportedly decided to end his life after having conversations about the future of the planet with an AI chatbot named Eliza’ Euronews, 31 March 2023– 31 March 2023. 179 Atillah I, ‘A Belgian man reportedly decided to end his life after having conversations about the future of the planet with an AI chatbot named Eliza’ Euronews, 31 March 2023– 31 March 2023. 178 Montgomery B, ‘Mother says AI chatbot led her son to kill himself in lawsuit against its maker’ The Guardian, 23 October 2024– on 23 October 2024. 177 Montgomery B, ‘Mother says AI chatbot led her son to kill himself in lawsuit against its maker’ The Guardian, 23 October 2024– on 23 October 2024. 176 Montgomery B, ‘Mother says AI chatbot led her son to kill himself in lawsuit against its maker’ The Guardian, 23 October 2024– on 23 October 2024. 26 https://www.euronews.com/next/2023/03/31/man-ends-his-life-after-an-ai-chatbot-encouraged-him-to-sacrifice-himself-to-stop-climate- https://www.euronews.com/next/2023/03/31/man-ends-his-life-after-an-ai-chatbot-encouraged-him-to-sacrifice-himself-to-stop-climate- https://www.euronews.com/next/2023/03/31/man-ends-his-life-after-an-ai-chatbot-encouraged-him-to-sacrifice-himself-to-stop-climate- https://www.euronews.com/next/2023/03/31/man-ends-his-life-after-an-ai-chatbot-encouraged-him-to-sacrifice-himself-to-stop-climate- https://www.euronews.com/next/2023/03/31/man-ends-his-life-after-an-ai-chatbot-encouraged-him-to-sacrifice-himself-to-stop-climate- https://www.euronews.com/next/2023/03/31/man-ends-his-life-after-an-ai-chatbot-encouraged-him-to-sacrifice-himself-to-stop-climate- https://www.theguardian.com/technology/2024/oct/23/character-ai-chatbot-sewell-setzer-death https://www.theguardian.com/technology/2024/oct/23/character-ai-chatbot-sewell-setzer-death https://www.theguardian.com/technology/2024/oct/23/character-ai-chatbot-sewell-setzer-death 2.4. Conclusion The chapter aimed to study how chatbots work and their inherent technology that allows it to function. It further analysed the benefits and harms that they could cause to customers when they are introduced to customer management systems of businesses. Through this, the Chapter has found the need for businesses to balance the risks and benefits that arise as a result of using chatbots. Additionally, in the event that they do use chatbots, they should ensure proper training from its onset and carry out frequent safety and ethical practices to ensure that it provides accurate information to customers. 27 3.0. Tracing the Evolution of Liability under Kenyan Tort Law and Its Implications on Liability for Chatbot Harms 3.1. Introduction Tort law, a branch of common law, is rooted in customary principles that have been refined and solidified through judicial precedents over time.182 Tort is an act which, whether intentional or not, is either contrary to law or a specific legal duty or a violation of a right.183 It is therefore defined as a civil wrong, independent of a contract, for which the remedy is common law action for unliquidated damages.184 Its primary aim is to uphold principles of justice–corrective, distributive and retributive–by providing remedies for wrongs and ensuring accountability.185 In Kenya, tort law was introduced through the reception clause in the Judicature Act, which allows for the application of common law principles.186 However, this application is subject to adaptation, as courts are directed to consider Kenya’s unique circumstances and the needs of its people when interpreting and implementing the principles.187 There are various types of liability under Kenyan jurisprudence.188 Like any other common law jurisdiction, the types of liability include; strict liability, negligence, and vicarious liability.189 These forms of liability all seek to ensure that harms caused to a victim are able to be compensated for, through the identification of the person(s) or object responsible for such harm.190 With this in mind, chatbots also cause harm, as identified in 2.4. above. However, there are varying degrees of what proponents think the appropriate means of liability should be. Currently, there are proponents supporting each type of liability and their justifications for such a proposal, 190 Akech M, ‘The common law’s approach to liability and redress its applicability to East Africa’ 1. 189 Akech M, ‘The common law’s approach to liability and redress its applicability to East Africa’ 1. 188 Akech M, ‘The common law’s approach to liability and redress its applicability to East Africa’ 1. 187 Akech M, ‘The common law’s approach to liability and redress its applicability to East Africa’ 1. 186 Section 3(1), Judicature Act (CAP 8). 185 Akech M, ‘The common law’s approach to liability and redress its applicability to East Africa’ 1. 184 Hussain A, General principles and commercial law of Kenya, East African Educational Publishers Ltd, 55. 183 Hussain A, General Principles and Commercial Law of Kenya, East African Educational Publishers Ltd, Kenya, 55. 182 Akech M, ‘The common law’s approach to liability and redress its applicability to East Africa’ International Environmental Law Research Centre (IELRC.ORG), 2003, 1- on 6 September 2003. 28 https://www.ielrc.org/activities/workshop_0309/content/akech.pdf which is what will be discussed in this chapter, to identify the strengths and weaknesses of the suggested forms of liability. 3.2. Strict Liability 3.2.1. Examining Judicial Precedents and Legal Implications for Strict Liability Strict liability as stated in the case of Rylands v Fletcher, is whereby liability is imposed on the owner of the land for damages caused by a product that escaped from their property to a neighbour’s land.191 In the case of Kenya Shell Limited v Milkha Kerubo Onkoba,192 The Court of Appeal confirmed that the rule in Rylands v Fletcher is recognised in our law. The prerequisites of a strict liability claim are: first, that the defendant made a ‘non-natural’ or ‘special’ use of his land; second, that the defendant brought onto his land something that was likely to do mischief if it escaped; third, that the substance in question escaped; and fourth, that the Plaintiff’s property was damaged because of the escape.193 The courts have held that the principle of non-natural use of land under Rylands v Fletcher is not a rigid concept but a developing rule that adapts to the continuous changes in modern life.194 The term “non-natural use” refers to a special use of land that increases the risk to others and goes beyond ordinary use or a use that serves the general benefit of the community.195 Just to shed more light, on Kenya’s decisions with regards to what a non-natural and natural use of land is, I’ll begin by discussing the case of Kenya Ports Authority v East African Power & Lighting Company Ltd,196 where the Court of Appeal found that the bringing of oil into the premises of the plaintiff for the production of electricity was not a non-natural use of land, as the Plaintiff should have anticipated the defendant having oil in their premises, given they were 196 (1982) eKLR. 195 Richard v Lothian (1913); AC 263 (C.A) Patrick J.O. Otieno v Lake Victoria South Water Services Board (2020) eKLR. 194 David M Ndetei v Orbit Chemical Industries Limited (2014) eKLR. 193 David M Nde