Generative Pre-trained Transformer
There exists a break-out moment for A.I. unlike anything prior over the past half-century. A.I. is quickly moving out of R&D labs and into the foundational aspects of every organization – throughout their employee’s work and their customer’s touch-points. A.I. enables significant benefits to corporations by increasing efficiency, improving decision making, enhancing the customer experience, creating new products and services, as well as optimizing operations.
This break-out moment has been predominantly led by GPT, Generative Pre-trained Transformer, a state-of-the-art language processing model developed by OpenAI. It is capable of understanding and generating human-like text, making it useful for a wide range of natural language processing tasks, such as language translation, text summarization, and question answering. Most notably, GPT-3 (GTP’s third iteration) is able to perform tasks that were previously thought to require human intelligence, such as writing coherent and grammatically correct sentences, and even writing essays, articles or coding.
One of the key strengths of GPT-3 is its ability to generate high-quality text without any task-specific training. This makes it easy to use for a wide range of applications, from automated content generation to conversational interfaces. GPT-3 is also highly customizable, allowing developers to fine-tune it for specific tasks or domains. When GPT-3 was released in May of 2020, it launched with a deep-learning model for natural language processing, with 175 Billion parameters, 100x more than the previous version, GPT-2. For reference, GPT-1 was trained on just 117M parameters.
GPT-4 builds on the success of GPT-3 and is predicted to be significantly larger and more powerful than its predecessor, potentially having up to 170 trillion parameters; a 1,000x increase over anything the world has experienced prior. In the context of horizon-three innovation, this is not an “if”, but a “when”. This next evolution in artificial intelligence will be both transformative and disruptive to our core business over the next three to ten years; this is an inevitability.
Language Learning Models
Large language models such as GPT-3 are the latest state of the art models and have been trained with billions of words and have the ability to generate text that can be hard to distinguish from text written by a human.
A large language model is a type of artificial intelligence algorithm that has been trained to understand and generate human language. These models are typically trained on vast amounts of text data, such as books, articles, and websites, in order to learn the patterns and relationships between words, phrases, and sentences.
When a model such as GPT-3 is trained on top of additional parameters it enables an ability to unleash a level of proprietary contextualization to an organization, such as Northwestern Mutual. GPT-3 understands what “Whole Life Insurance” is in the context of a “Mutual”, but it does not understand what “Whole Life Insurance” means in the context of “The Northwestern Mutual Life Insurance Company”; however, it can be trained.
Imagine a future where a language learning model, such as GPT, could speak on all of our products at the same level as our most knowledgeable advisors, and do it in the style of any specific advisor while maintaining the tone of Northwestern Mutual and also being in total compliance with any and all regulations. With all of this communicated in a manner that is most receptive to the individual client in their specific situation.
Ironically, Northwestern Mutual is uniquely positioned to most take advantage of a more contextualized language learning model for A.I. because of its vast human enabled field force. Our field is specifically trained at “fact finding”, a compelling competitive advantage in the pursuit of training an A.I. to have the contextualization on the individual client level. It’s not just exclusive distribution, it’s also exclusive data input.
Future Applications of GPT-4
McKinsey estimates a potential annual value of up to $1.1 trillion if AI tech is fully applied to the Insurance industry. For Northwestern Mutual specifically, GPT could be used to analyze an individual's data and recommend life insurance products that are tailored to their specific needs and risk profile; other potential (high-level) use-cases include:
Policy contract generation: an ability to generate more personalized policy and contract documents more quickly, reducing the need for human input while also increasing accuracy.
Risk assessment: an ability to further analyze data from a much broader array of various sources, from medical records to social media, to help underwriters better assess an individual's risk profile.
Portfolio management: an ability to further analyze market trends and make investment recommendations based on a wider spectrum of inputs more quickly, helping manage portfolios more effectively.
Predictive modeling: an ability to create more predictive models that help anticipate future claims and adjust pricing accordingly based on more contextually relevant and more real-time data inputs.
Compliance: an ability to assist in compliance with regulations with employees and advisors on everything from product messaging to marketing communication.
Marketing and sales: an ability to analyze data and create individualized targeted marketing campaigns as well as personalized communications and engagement strategies for policyholders. Additionally, providing personalized customer service and answering customer questions, which could lead to improved customer satisfaction and retention.
Improved efficiency in underwriting: an ability to automate the process of analyzing and interpreting large amounts of data in order to assess an individual's risk profile and determine an appropriate premium.
Fraud detection and streamlined claims processing: an ability to analyze large amounts of data to identify patterns and anomalies that might indicate fraud in addition to analyzing claims data to assist with the claims process. All of which could help to reduce the time and cost associated with processing claims as well as help to prevent losses.
The Rabbit-hole: Recommend Readings
Meet GPT3. It Has Learned to Code (and Blog and Argue) https://nytimes.com/2020/11/24/science/artificial-intelligence-ai-gpt3.html
GPT3 and the Next Generation of AIPowered Services
https://datacamp.com/blog/gpt-3-and-the-next-generation-of-ai-powered-servicesA Complete Overview of GPT3 The Largest Neural Network Ever Created https://towardsdatascience.com/gpt-3-a-complete-overview-190232eb25fd
Why GPT3 is the best and worst of AI right now https://technologyreview.com/2021/02/24/1017797/gpt3-best-worst-ai-openai-natural-language/
What is GPT3? Everything You Need to Know https://techtarget.com/searchenterpriseai/definition/GPT3
Conclusion
The next generation of A.I., artificial intelligence, has the potential to become the most powerful business tool at our disposal from how our engineers write code, to how our actuaries underwrite, down to how our field markets and communicates to our policyholders. We no longer have to pick-two; you can do what you did yesterday tomorrow better, faster, cheaper. The future of A.I. is an augmentation to how we used to operate; one that will be as transformative, if not more significant, than what the introduction of the personal computer and the internet has been to our core business.
Marty Ringlein
General Partner
