The use of virtual assistants (VAs) has seen rapid growth in both personal and professional spheres. These sophisticated applications excel at task execution, responding to inquiries, and proficiently communicating through synthesized voices. Users can utilize voice commands for a diverse array of tasks, spanning from answering questions to controlling household appliances and scheduling appointments. VAs offer various services,` such as information provision, guidance, decision support, and educational resources. Its central role is to provide real-time data, empowering operators and directors with easily accessible and curated information to bolster their decision-making processes.
Image credit: TippaPatt/Shutterstock
VAs and chatbots, known for their seamless natural language communication, also contribute significantly by disseminating static information, particularly in the context of educating about production processes. Solutions exist to answer queries using knowledge bases, with some employing virtual reality (VR) technology to provide information within immersive 3D environments. Furthermore, chatbots serve as valuable tools for training and knowledge transfer to new manufacturing workers, while adaptive frameworks aid in learning from a company's operational processes, utilizing graphical representations to guide and instruct new employees effectively.
Artificial intelligence (AI)-driven disembodied VAs, such as chatbots and voice assistants, harness natural language programming (NLP), machine learning, and deep learning to emulate human cognition. They excel at assisting customers throughout their shopping journeys, from acting as shopping aides to salespersons. These technologies hold the potential to revolutionize the customer experience in retail.
Difference between VAs and Chatbots
Chatbots and virtual assistants, both AI applications, serve various business purposes. Chatbots act as information interfaces, handle queries, schedule appointments, and interact with users. Meanwhile, VAs extend their reach into business support, reminder management, and note-taking, offering a broader scope of services. Both leverage AI technologies, with chatbots using information retrieval interfaces and script-bound, lacking learning and adaptability, while VAs employ AI, NLP, and artificial neural networks, making them more dynamic and productive.
Emergence of VAs
The history of VAs dates to the 1910s, with technological advancements and the application of AI significantly enhancing their capabilities. While VAs currently utilize narrow AI with limited options, the potential adoption of general AI in the future holds the promise of revolutionizing their service quality. The emergence of intelligent virtual assistants (IVAs) traces back to the 1990s, when companies such as IBM, Philips, Lemont, and Hauspie began integrating digital voice recognition into personal computers. In 1994, the IBM Simon, the first smartphone, set the stage for modern virtual assistants.
In 1997, Dragon introduced the Biologically Talking application, which could transcribe natural speech at a remarkable pace, and doctors still use it for medical records today. In 2001, Colloquies unveiled the Smarter Child, capable of interactive functions like gaming and weather checking. Siri, born in 2011 as an iPhone 4S feature, marked a significant leap in its capabilities, from texting and calling to recommendations, internet searches, and navigation. Amazon introduced Alexa and Echo in 2014, revolutionizing home automation. In 2017, Amazon opened the door for users to create conversational interfaces for any VA or interface.
Since 2017, VAs have evolved significantly, enhancing decision-making across industries. They interact through text, voice, and image processing, with AI techniques and machine learning driving continuous improvement. Voice-activated assistants use prompts such as "OK Google" or "Hey Siri" to initiate tasks but face growing legal concerns. IVAs assist with tasks ranging from setting alarms and weather updates to entertainment and government interactions. They also serve as valuable supplements and replacements for customer service, alleviating the burden on human-staffed call centers.
Applications of VAs
VAs used in several domains including education, industry, and finance.
Language Learning: VAs plays a significant role in the learning process, aiding in various educational contexts. They can serve as intelligent tutoring systems, enhance student participation, assist in repetitive tasks, and provide immediate information and guidance. In the context of foreign language learning, virtual assistants such as Google Home, Alexa, and Siri have the potential to revolutionize classrooms.
Industry: VAs acts as a software agent that performs actions or services based on commands or questions, activated through voice or text inputs. It acts as an abstraction layer on top of the system, employing technologies such as machine learning, speech recognition, and more. Chatbots, akin to VAs, interact with humans via text or voice and serve as fundamental components of VAs.
Finance: VAs offer genuine economic value for businesses, serving as knowledgeable, always-available aides. They can organize meetings, manage inventories, and verify data, making Internet of Things (IoT) adoption easier for small and medium-sized enterprises.
VAs Powered by Large Language Models
Integrating large language models (LLMs) opens doors to advanced VAs. NETA is a VA for active aging integrated into a robot proficient in Galician and Spanish and incorporating LLMs. Its primary objective is to enable elderly individuals to maintain their independence, prevent hospitalization, promote healthy habits, support medical treatments, enhance security, and foster social interactions through AI, including video calls and community participation.
GiDi is a collaborative project with Axyn Robotique, aiming to create a VA for Screening Protocols at Home (SPH). Like NETA, GiDi empowers elderly individuals to remain at home longer, avoiding hospitalizations, managing primary care and emergencies, and prioritizing the social dimension of patients.
The XIA project focuses on implementing conversational assistants for the Galician government's electronic administration procedures. These assistants operate in both Galician and Spanish, offering citizens a 24-hour channel for inquiries related to economic aid, official accreditations, and other administrative processes.
The conversational assistants in the initial project phases shared similarities with NETA and GiDi, employing modeled information and logic-driven responses. Subsequently, integration with LLMs was explored, enabling natural language responses based on contextual information. This integration significantly enhanced the capabilities of the conversational assistant.
Challenges and Ethical Implications of VAs
VAs have swiftly found applications across numerous sectors, including customer service, commerce, education, finance, healthcare, entertainment, and more. They offer real-time responses, catering to the needs of millennials and enhancing brand loyalty. However, it is crucial to recognize that VAs, which rely on human-built AI technology, introduce various ethical concerns.
Various countries have issued AI policy guidelines with a core focus on ethical principles. Ethical concerns arise when AI intentionally or unintentionally discriminates against groups, such as in biased loan application algorithms. Transparency plays a significant role in maintaining trust, with brands expected to be honest about virtual assistant interactions. Designers and developers need to disclose what VAs can do, how data is handled, and provide options for users to control their data.
Justice, fairness, and equity emphasize prioritizing consumer interests and providing impartial recommendations. This is particularly relevant in recommendation systems, where biases can affect suggestions. Non-maleficence involves ensuring consumer safety and preventing abuse, which can occur in interactions with VAs. Designers should create empathetic conversation flows and offer options for users to select their preferred virtual assistant persona.
Responsibility and accountability refer to defining roles and data ownership, particularly in terms of the data generated during conversations. Designers must clarify who bears responsibility, legal liability, and data ownership, especially as the data collected becomes increasingly personal. Privacy concerns revolve around data protection, sharing, and storage. Designers should outline clear privacy policies and address key privacy questions related to accessing, sharing, and storing conversation data.
LLM-powered virtual assistants can generate coherent but factually incorrect responses, raising concerns about misinformation. Detecting and mitigating hallucinations, where the LLM produces inaccurate information beyond the provided input, is essential for maintaining reliability and accuracy. LLMs trained on extensive datasets can inadvertently perpetuate biases present in those data sources. This can result in biased responses and discriminatory outcomes. Ensuring fairness and preventing discrimination are paramount, especially when developing virtual assistants for public services.
Understanding these ethical principles is paramount for designers, developers, and consumers, ensuring the responsible and ethical implementation of AI in VAs.
Despite the progress in AI and VAs, challenges hinder their widespread adoption. These include concerns about job displacement, a lack of understanding of AI's potential benefits, and the scarcity of relevant data. Many retailers employ rudimentary chatbots, while AI-powered VAs remain underutilized and underpromoted. Customer acceptance also poses a significant hurdle, with only a small fraction of digitized companies integrating AI VAs into their operations.
Considering the gap between the potential benefits of AI VAs and their actual utilization, exploring the obstacles hindering their wider adoption in the retail sector is imperative. Exploring internal and external challenges while considering customer-related and socio-cultural factors is imperative for designing and deploying effective virtual assistants in this industry.
References and Further Readings
Pereira, Rodrigo, Claudio Lima, Tiago Pinto, and Arsénio Reis. (2023). Virtual Assistants in Industry 4.0: A Systematic Literature Review, Electronics 12, no. 19: 4096. DOI: https://doi.org/10.3390/electronics12194096
Soofastaei, A. (Ed.). (2021). Virtual Assistant. IntechOpen. https://www.intechopen.com/books/10393
Arora, S., Athavale, V.A., Himanshu Maggu, Agarwal, A. (2021). Artificial Intelligence and Virtual Assistant—Working Model. In: Marriwala, N., Tripathi, C.C., Kumar, D., Jain, S. (eds) Mobile Radio Communications and 5G Networks. Lecture Notes in Networks and Systems, vol 140. Springer, Singapore. DOI: https://doi.org/10.1007/978-981-15-7130-5_12
Pereira, R., Reis, A., Barroso, J., Sousa, J., and Pinto, T. (2022). Virtual Assistants Applications in Education. In International Conference on Technology and Innovation in Learning, Teaching and Education (pp. 468-480). Cham: Springer Nature Switzerland. DOI: https://doi.org/10.1007/978-3-031-22918-3_38
Piñeiro-Martín, Andrés, Carmen García-Mateo, Laura Docío-Fernández, and María del Carmen López-Pérez. (2023). Ethical Challenges in the Development of Virtual Assistants Powered by Large Language Models, Electronics 12, no. 14: 3170. https://doi.org/10.3390/electronics12143170