Will Humans Fall in Love with AI? The Future Vision Microsoft is Painting with 'AI with Personality'
In recent years, generative AI and large language models have evolved rapidly, beginning to have a significant impact on our work and daily lives. Particularly in the field of chatbots, AI is emerging that goes beyond simple question-and-answer interactions to learn a user's 'context' and 'past interactions,' exhibiting behavior that resembles a personality. The utilization of such AI personalization features and 'memory' is expected to advance further, and IT companies are all competing to develop AI assistants that can 'converse more human-like and learn from the user's situation to provide support.'
In this article, based on an interview with Mustafa Suleyman, CEO of Microsoft AI and a co-founder of DeepMind, we will explain the future of next-generation AI assistants, personalization features, and the direction of AI development. He emphasizes that 'AI will evolve in a direction that understands the user's personality, past interactions, and goals to support the organization of thoughts and the completion of tasks,' but what exactly does that look like? Let's gently unravel these expert perspectives.
1. The Evolution of AI Assistants: Memory and Personalization
1-1. The Difference Between Traditional Chatbots and Next-Generation AI
Traditional chatbots were mainly Q&A-based, with a simple design that returned immediate answers to questions entered by the user. While this is certainly convenient enough on its own, Mustafa Suleyman points out, 'Until now, only IQ (intellectual elements) that summarizes facts briefly has been emphasized. However, from now on, performance that enables more human and comfortable communication with the user, also known as EQ (emotional elements), will become important.'
The key here is 'memory' and 'personalization.' By having the AI remember the user's basic information and what topics have been discussed in the past, interactions become smoother, leading to an experience where 'the AI understands without having to re-explain the situation every time.' Furthermore, as the AI learns the user's preferences, writing style, and even values, it will move closer to even more natural communication.
1-2. New Possibilities Brought by Memory Functions
According to Suleyman's explanation, a mechanism to 'continuously remember individual user information and learn styles and preferences' is scheduled to be added as a new feature of Microsoft's AI assistant, 'Copilot.' For example, it will be able to remember the pronunciation of your name that you taught it previously, your family structure, hobbies and preferences, and the tone of your writing (whether you like bullet points or conversational style, how much humor is appropriate, etc.), and use that to help in future conversations.
Suleyman states the following:
'Instead of discarding the history of conversations you've invested in, you build on that learning. That is the big point where users can feel the "magic."'
However, he also states that 'it cannot be called perfect yet,' and it is at a stage where challenges such as incorrect answers and forgetting remain in practical operation. On the other hand, the potential to dramatically change users' lives and work efficiency from these early stages is suggested.
1-3. Action Functions: Toward AI That Actually Operates
Furthermore, it is said that in the future, assistants will transition from mere conversation to a stage where they 'actually operate.' For example, features that 'remotely control the user's PC' will be gradually installed, such as Copilot performing browser operations on behalf of the user, booking flight tickets or restaurants, and changing application settings.
According to Suleyman, 'If this kind of automation is realized, for example, "troublesome parts" such as daily administrative work and various arrangements will be taken over by AI, and users will be able to secure more creative time.' However, he also states that caution is required regarding issues such as 'how much to allow AI to operate' and 'how to set boundaries for security.'
2. AI Assistants and 'Personality': Designing Boundaries
2-1. Avatars and Emotional Connections
In this interview, Suleyman also mentioned the possibility of AI assistants having avatars (visual characters). Some people may feel more attached to and interact more familiarly with an AI as a 'being with visuals and a voice' than a text-based one.
On the other hand, while introducing cases from other companies where 'multiple users have developed romantic feelings for AI and invited them to weddings,' he also clarified the policy that 'Microsoft will not design to that extent. We will set clear boundaries so that the AI firmly refuses interactions that deepen specific emotional exchanges too much.'
'It will be a polite and kind support presence, but we will draw a line in advance so that it does not step into a direction comparable to romantic feelings between humans. This is the "safe personality design" we are thinking of.'
2-2. Consideration for Privacy and Safety
An AI that remembers a user's past interactions and personal information will hold a vast amount of data. Since this comes with privacy and security risks in exchange for convenience, Mr. Suleyman emphasized that they will 'exercise the utmost caution.' In addition, he suggests that companies should take greater responsibility for the handling and design of AI to prevent unexpected trouble caused by the AI becoming too deeply involved in the trivial conversations of individuals and households.
3. Continuous Development of Large-Scale Models and Future Prospects
3-1. Is the Scale of Models Reaching Its Limit?
In recent years, large-scale models exceeding hundreds of billions to a trillion parameters have appeared one after another, but there is some debate that 'the pace of scaling up may be hitting a plateau.' However, Mr. Suleyman is optimistic, stating that 'there is still plenty of room for performance improvement.' For example, he notes that by enhancing 'Chain-of-Thought reasoning (logical reasoning ability),' the means for problem-solving will expand beyond mere knowledge volume.
However, research is also progressing toward achieving equal or better performance even with smaller scales through more efficient learning methods and innovations in inference, rather than just making models larger. There are teams within Microsoft developing proprietary models, and a dual strategy is being adopted: 'collaborating with OpenAI on cutting-edge models while pursuing efficient and user-friendly models in-house.'
3-2. Partnership with OpenAI and Expectations for AGI
Microsoft entered into a large-scale partnership with OpenAI in 2019, benefiting mutually by providing cloud infrastructure, Azure, while also gaining access to OpenAI's latest technology. Furthermore, as of 2023, additional investments in the billions of dollars have been made, and Mr. Suleyman states that 'this partnership will continue to yield fruitful results well beyond 2030.'
While it is a very difficult question to define when the arrival of 'AGI (Artificial General Intelligence)' will occur, Mr. Suleyman says, 'While I cannot rule out the possibility of it being realized within five years, my personal feeling is that it will likely take about ten years.'
4. Changes in Work and Learning in the AI Era
4-1. Impact on White-Collar Workers and Business Transformation
Mr. Suleyman clearly states that the progress of AI will 'force a transformation in white-collar work.' In particular, if AI can handle advanced reasoning and be entrusted with tasks such as writing, report creation, information analysis, and booking appointments, humans will inevitably shift to a role of 'giving instructions to AI and checking the results.' In that process, it is highly likely that some jobs will be replaced by AI or that the nature of work itself will be redefined.
However, Mr. Suleyman also emphasizes that 'people will be able to focus on areas involving creativity and interpersonal communication, allowing us to build a more productive society.' This is precisely why 'collaboration between humans and AI' becomes important.
4-2. A Message to the Youth Who Will Carry the Next Generation
Regarding what skills young people should acquire from now on, Mr. Suleyman recommends 'first, actively engaging with AI.' He says it is important to find your own ideas and ways of using it while experiencing the boundaries of what AI is good and bad at.
'It has as big an impact as when the internet was first connected. However, like other technologies, the value you get depends on how you use it. You should just keep trying.'
5. Brand Value and the Future of Advertising Models
While new-era AI assistants build deep 'one-on-one' relationships with users, balancing this with advertising, which is a source of corporate revenue, will be a major challenge. In particular, user backlash is expected if 'an AI assistant you talk to intimately on a daily basis suddenly presents an advertisement.'
Mr. Suleyman addresses this issue by mentioning that 'advertising is not necessarily bad; rather, it is important how naturally you can present higher-quality and more useful advertisements.' In short, the idea is that 'if an advertisement matches your needs and the timing is appropriate, it becomes useful information for the user.' Also, how brands can build 'trust and emotional connections' with users is considered a greater differentiator in the coming AI era.
The future vision of AI assistants presented by Mustafa Suleyman is not just about searching or answering questions, but as a 'personal companion' that handles necessary tasks in advance while building a continuous relationship with the user. There, 'memory,' 'personalization,' and the design of the entire user experience are essential, and advanced adjustments that delve into human psychology will likely be made.
Not only Microsoft and OpenAI, but other companies like Amazon and Google are also competing in the development of AI assistants, but Mr. Suleyman predicts that 'ultimately, people will choose the AI that suits their preferences.' Those 'preferences' may include not only functional superiority but also the AI's 'personality' and 'alignment of values.'
While we are currently at a stage where AI only partially assists with daily tasks and frees users from stress and boredom, in a few years, it may advance to a level where it further expands human potential. As personalization and memory functions mature, we can say that we are at a point where we must re-examine what it means to be human and how we will coexist with technology.
Related Articles