Disclaimer: Open AI’s GPT3 (davinci-003) was used for generating parts of this blog post. Text was generated, then edited and verified for factual mistakes by a human experienced in the domain in question.
In this post, we address three pertinent questions in AI personas, also known as AI-generated personas. We define them and discuss their creation and central challenges.
What is an AI-generated persona?
An AI-generated persona is a digital character generated by Artificial Intelligence (AI) technologies that imitate the behavior and qualities of a real human persona. AI-generated personas can be used in marketing, virtual conversation, conversational interfaces, search, and more. They are created as virtual avatars for a certain purpose or a set of tasks, for example to help companies engage with and understand their target customers.
How to create AI-generated personas?
1. Gather Data – When trying to create AI-generated personas, it’s important to first understand the customers that will interact with the business. Gather customer data such as emails, customer surveys, customer feedback, or customer interviews to begin gathering insight into customer needs, wants, attitudes, and preferences.
2. Analyze Patterns & Descriptors – Next, use AI and machine learning to analyze the qualitative data. AI can help identify patterns and gather descriptors that will allow the development of AI-generated personas. This is often referred to as ‘segmentation’.
3. Develop Personas – With the data gathered and analyzed, the AI can now begin to create personas. AI can develop personas based on different characteristics such as demographic, lifestyle, interests, behavior patterns, attitudes, and more. This is often referred to as ‘enrichment’.
4. Deploy Personas – With the AI-generated personas, businesses can use them to create campaigns, segment customers, understand user emotions and behavior, create personalized experiences, and analyze customer journeys.
What are notable research challenges in AI-generated personas?
1. Capturing User Preferences: Creating realistic AI-generated personas requires accurately capturing user preferences and behavior. This involves understanding the nuances of user behavior and preferences, and providing the AI system with a wide range of examples to learn from. There is no single correct way to create personas from the same data, but both algorithms and humans tend to interpret the data differently during different iteration cycles.
2. Modeling Language and Social Skills: AI-generated personas need to be able to converse naturally with users, which is an extremely difficult task. The AI system must be highly sensitive to the subtleties of human language to be able to accurately interpret user intent and respond appropriately. Generative AI models based on LLMs (large language models) seem to have addressed this limitation to a considerable degree.
3. Recognizing Context: AI-generated personas need to understand the context in which a conversation is taking place in order to generate appropriate and meaningful responses. This requires AI systems to be able to effectively reason about the present situation and also draw from past user interactions. This remains a challenge, as without understanding the context, the models can hallucinate, i.e., give false information. Because the outputs of the Gen AI models are usually written in perfect grammar and convincing tone, people might be persuaded to believe the information even when it is false.
4. Managing Complex Interactions: AI-generated personas should be able to handle complex interactions in a natural and efficient way. This involves AI systems being able to successfully track conversations that span long periods of time, identify relevant topics, and generate appropriate responses. Currently, the iteraction with AI personas is based on short, momentary events. Tracking the user history is typically not part of the AI personas’ operating logic, which limits their applicability.
Want to learn more about AI-generated personas? Try our systems: Automatic Persona Generation (https://persona.qcri.org) for web analytics data and Survey2Persona (https://s2p.qcri.org) for survey data.