[Evolving AI Assistants] From AI Assistants and Agents to Avatars as 'Another Me'

Beyond Secretary and Colleague, the Possibility of Creating 'Me' in the Digital World
SKT, Naver, and Other Companies Accelerate AI Agent Development
Must Increase User Trust and Eliminate Privacy Concerns

Reid Hoffman, co-founder of the global social platform LinkedIn and an early investor in OpenAI, conducted an interesting experiment with artificial intelligence (AI) technology last April. He trained an AI model on all the books, speeches, podcasts, and media interviews he had produced over 20 years to create his own avatar called ‘Reid AI.’

Reid Hoffman created an avatar called 'Reid AI' in April, which learned data about himself using various artificial intelligence (AI) services, and conducted an interview with it. Hoffman and Reid AI interacted by asking and answering questions with each other. [Source=YouTube]

Hoffman released a roughly 14-minute YouTube video showing a Q&A session between himself and Reid AI. In the video, Reid AI replicated Hoffman’s face and voice perfectly, with natural gestures. It summarized a book Hoffman wrote in one sentence and shared its views on government AI regulations. Around the middle of the video, Reid AI asked Hoffman, “What do you think about the issue of jobs disappearing due to AI?” demonstrating interactive communication. Viewers reacted with comments such as “I was amazed by Reid AI’s natural gestures,” “It’s so impressive it gave me chills,” and “It’s fantastic.”

Hoffman believes Reid AI can translate his thoughts into various languages and serve as a mirror to reflect on his ideas and personality. In other words, he is considering ways AI avatars can help connect and collaborate with more people.

In a post titled ‘Q&A with AI Avatar,’ Hoffman said, “A powerful technology has arrived,” and added, “We must use this technology to evolve in a more human way.” He emphasized, “A careful and cautious attitude is important as we shape the future.”

AI services are evolving step-by-step from simple chatbots to assistants, agents, and finally avatars. AI assistants, like Apple’s Siri, Amazon’s Alexa, and OpenAI’s ChatGPT, answer questions and help with tasks but require user commands and do not operate independently. In contrast, AI agents, which can perform tasks similar to humans, are characterized by autonomy?they perceive the environment, make decisions, and suggest what users need. Examples include OpenAI’s GPT-4o, reminiscent of the movie ‘Her,’ and Google’s futuristic AI agent ‘Project Astra.’ Amazon Web Services (AWS) describes AI agents as “autonomous intelligent systems that perform specific tasks without human intervention.” The next stage beyond AI agents is digital twins or AI avatars like ‘Reid AI,’ which perfectly replicate users.

The Ultimate Goal of Conversational AI: Autonomous and Creative Execution

According to global market research firm MarketandMarkets, the conversational AI market is expected to grow at an average annual rate of 24.9% through 2030. It is projected to expand from $13.2 billion (approximately 17.5 trillion KRW) this year to $49.9 billion (66 trillion KRW) by 2030.

Naver CEO Choi Soo-yeon is unveiling the large language model 'HyperCLOVA X' for artificial intelligence (AI) services. Photo by Kang Jin-hyung aymsdream@

South Korea’s leading conversational AI agent is NAVER’s ‘ClovaX.’ ClovaX recently upgraded to a multimodal AI capable of processing visual information. It can analyze tables and graphs, infer situations in photos, write code, and has a broader range of applications.

Recently, SK Telecom unveiled its ‘Personal AI Agent (PAA)’ strategy in collaboration with the U.S. AI unicorn company Perplexity. The development levels are divided into five stages, from Level 1 to Level 5. Currently, SKT’s app ‘A. Dot’ has been upgraded to Level 2, utilizing a multimodal large language model (LLM) with reasoning and search capabilities. At Level 2, the AI agent can analyze user preferences and assist with tasks in specific fields such as music, media, and securities.

The goal of Level 3 is for AI agents to independently and completely handle individual tasks. Level 4 involves analyzing complex tasks, planning, and executing them. At the final Level 5, AI agents will be capable of ‘creative task execution,’ autonomously proposing and designing tasks.

As AI technology advances, it could go beyond being an assistant or colleague to become ‘another me’ in the digital world. At this stage, it would not only enhance personal productivity but also expand connectivity by enabling communication and collaboration with people worldwide in the digital realm, regardless of language barriers. This vision aligns with Hoffman’s future outlook that AI technology can ‘extend’ rather than ‘replace’ humans.

Privacy ‘Hurdles’ Encountered in AI Evolution

However, nothing comes for free. The more an AI agent knows about a user, the more accurate and satisfying the results it can produce. In other words, using AI services requires providing companies with detailed personal data. Therefore, extensive personal data collection is inevitable. According to ClovaX’s terms of use, conversations between users and the AI are stored, and some conversations containing harmful keywords may be reviewed by humans.

When signing up for A. Dot, various information is collected, including the user’s name, phone number, social accounts, email address, voice and chat command languages, personal interests and preferences inferred from AI service conversations and Q&A, device location information, summarized call content from A. Dot’s phone service, images and video files uploaded by users, and media usage history. It even collects dozens of data points, from the time securities information was requested to photos uploaded for AI profile creation.

Jung Seok-geun, head of SKT’s Global and AI Tech Division, revealed a global expansion strategy using this data. He said, “We can provide new AI-based customer experiences,” and added, “We can expand our business model beyond telecommunications to services such as restaurants users like, vacation destinations, and news interests.”

‘Deep Voice’ Crime Risks if Personal Data Is Leaked

Kim Myung-joo, president of the International Association for Artificial Intelligence Ethics (IAAE), said, “If personal and private information accumulated by companies is leaked, it can lead to real damages such as smishing and voice phishing.” He added, “In the future, companies must seriously consider identity verification processes. If AI agents operate solely by recognizing users’ voices, deep voice technology could be used to imitate them for new types of crimes.” Kim recommended, “AI companies have an obligation to prevent user data leaks and collect only the information essential for their services.”

Global research firm Gartner stated in a July report titled ‘Challenges CTOs Face in Realizing AI Value’ that “extensive collection and storage of personal data can lead to unauthorized access or identity theft.” AWS also noted, “Developing and operating advanced AI agents requires acquiring, storing, and transferring large amounts of data,” and advised organizations to “recognize data privacy requirements and take necessary measures to strengthen data security posture.”

It is also expected that it will be necessary to clearly distinguish whether AI-recommended services reflect user preferences or are advertisements. Aravind Srinivas, CEO of Perplexity, said, “We will diversify revenue sources, including starting advertisements.” Perplexity’s AI search service displays follow-up questions recommended to users along with answers, which could expose advertisers’ companies. Similar to early portal sites where it was difficult to distinguish between pure related search results and advertising sites, a similar situation could occur in AI services.