Blogs
Attributed to Peter Lee, Distinguished Engineer & Chief Technology Officer; and Karen Ko, Partner, IBM Consulting, IBM Hong Kong A recent study of IBM Institute of Business Value indicated that...
Attributed to Peter Lee, Distinguished Engineer & Chief Technology Officer; and Karen Ko, Partner, IBM Consulting, IBM Hong Kong
A recent study of IBM Institute of Business Value indicated that 64% of the respondent CEOs faced pressure to accelerate the adoption of generative AI. In line with this global statistic, a global surveyi reported that finance executives in Hong Kong are moving faster than other major markets in adopting generative AI, with 38% saying their organization has already started rolling out the technology – the highest of all the markets surveyed and well above the global average of 26%.
In order to adopt generative AI for real business values, enterprises need to evaluate the best fit for their corporate objectives while considering the three modes of generative AI consumption: first is to leverage software with generative AI embedded; second is to query AI models through API calls; and third is to create (and then query) your own foundation models leveraging public and private data.
Another aspect to consider is that not all generative AI models are created equal. Enterprises need one that is tailored for their unique requirement and own data. The promise of foundation models is rooted in their ability to be tuned to an enterprise’s unique data and domain knowledge with governance and flexibility at their core, making AI deployment significantly more scalable, affordable, and efficient.
We announced IBM watsonx for the needs of enterprises. It consists IBM watsonx.ai that supports multiple foundation models and offers watsonx.ai studio to facilitate enterprises to develop, fine tune and deploy their AI applications with foundation models built on trustworthy dataset and AI governance.
Whether enterprises would like to fine-tune open-sourced models, creating their own, or deploying AI on-premises or in the cloud, IBM is committed to empowering a generation of businesses, spanning every industry – to embed AI into the core of their strategies. We provide open, trusted, targeted and value creating AI solutions for businesses. IBM watsonx offers a seamless, efficient, and responsible approach to AI deployment on Hybrid Cloud (on-premise or off-premise).
Harnessing the power of business-targeted models developed from sound data
Business leaders adopting generative AI need flexibility and choice of models. Recognizing that one size doesn’t fit all, IBM’s watsonx.ai offers families of foundation models with different sizes and architecture that could help clients to deliver performance, speed with cost efficiency.
The initial release of watsonx.ai included the Slate family of encoder-only models that are highly effective for non-generative AI use cases for enterprise natural language processing, such as sentiment analysis, entity extraction, relationship detection and classification. With the introduction of Granite, the IBM-developed generative foundation models, businesses can take advantage of a decoder-only architecture designed for generative tasks, such as summarization, content generation, retrieval-augmented generation (a framework for improving the quality of response by linking the model to external sources of knowledge), classification, and extracting insights.
Besides flexibility and choice of models, enterprises also need secured access to business-relevant data to accelerate time to value and insights. IBM provides an IP indemnity (contractual protection) for its foundation models including the Slate and the Granite familiesii. Now clients can develop AI applications using their own data along with the client protections, accuracy and trust afforded by IBM foundation models.
Developed by IBM Research, the Granite models — Granite. 13b.instruct and Granite.13b.chat — use a “decoder” architecture, which is what underpins the ability of today’s large language models to predict the next word in a sequence. Granite.13b.v1.chat is designed for conversations and QA, while granite.13b.v1.instruct is designed to provide concise responses based on short instructions. Granite.20b.code was recently released with the watsonx Code Assistant, a 20 billion-parameter code generation model that helps enterprise developers and IT operations generate program code faster and accurate with natural language prompts.
All Granite foundational models have been trained on enterprise-focused datasets and curated by IBM from five domains including internet, academic, code, legal and finance, with 10% of its training data sourced from legal and finance domains. This unique data composition enables the Granite models to deliver superior performance for financial tasks, including credit risk assessment, insurance QA, conversational financial QA, and summarization, or for clients with legal requirements.
Trained with high-quality finance data, Granite demonstrates exceptional performance in financial tasks while requiring fewer GPU resources with higher cost-effectiveness. IBM Research conducted a preliminary evaluation and testing, encompassing 11 different financial tasks, with results showing that by training Granite-13B models with high-quality finance data, they have the potential to achieve either similar or even better performance than much larger models, notably Llama 2-70B-chat, BLOOM-176B, and gpt-neox-20B, among others. The evaluated financial tasks include sentiment scoring for stock and earnings call transcripts, classifying news headlines, extracting credit risk assessments, summarizing financial long-form text, and answering financial or insurance-related questions.
Structured approach for maximum ROI and risk mitigation in the insurance industry
In Hong Kong where the insurance industry is booming, forward thinking insurers have a unique opportunity to win the competition with trusted generative AI to deliver unprecedented customer experience and efficiency.
IBM has been collaborating with insurance industry clients on their generative AI journeys and utilizing its diverse capabilities in various use cases.
IBM Consulting adopts a structured methodology that helps organizations identify the most promising starting point for maximum returns from generative AI implementation.
Leveraging IBM's Industry Model for Insurance, the team thoroughly analyzes the insurance company's workflow across different business domains, assessing the potential impact and readiness for AI and automation in each area. This analysis provides valuable insights into the areas that offer the highest ROI for the insurance organization – considering financial scale, efficiency gains, and other benefit drivers such as improved customer experience, enhanced decision-making, and increased sales opportunities.
The team also considers potential legal and reputational risks associated with generative AI implementation, devising strategies to mitigate these risks through control and governance measures. This comprehensive risk assessment ensures that the chosen starting point not only yields a favorable ROI but also minimizes potential drawbacks.
IBM believes generative AI can deliver the most value within insurance in three major aspects: customer care, operations, and application modernization.
- For customer care, generative AI can be used to realize high customer acceptance of personalized risk conversations and advice, virtual agents to assist customers while keeping agents and service reps in the loop, immediate and personalized recommendations to service reps for next best action on claims and policy admin.
- For operations, we can leverage generative AI for underwriting information gathering and synthesis across many data stores, summarization of claims case details to speed claims resolution with automated entry for jet resolution, and improvement of agent productivity with data fill and routine tasks, such as policy transfer and direct customer assistance with simple tasks.
- For application modernization, generative AI can be used to change business case for legacy core application modernization, retain key benefits of IBM Z as secure data repository and AI platform while enhancing performance, security, sustainability and reliability. It can also achieve AI-powered modernization of insurance rules and code bases with high reuse/acceptance.
Looking to the future
IBM watsonx will continue to evolve and make significant releases. We will be focused on expanding enterprise foundation model use cases beyond NLP and operationalizing 100B+ parameter models for bespoke, targeted use cases – opening the door to broader enterprise adoption. Moreover, IBM watsonx Granite is actively developing models with varying parameter sizes, offering a range of options to address the specific needs of different businesses.
We also look towards bringing to bear the strength of our AI governance capabilities with general availability of watsonx.governance – helping organizations to implement end-to-end lifecycle governance, mitigate risk and manage compliance to the growing AI and industry regulations. AI governance should never be an afterthought, so we encourage our customers begin governance of their ML models and foundation models at the outset.
IBM Consulting’s watsonx practice brings expertise in the generative AI technology stack as well as domain and industry experience that can help accelerate clients’ business transformations – addressing demands for AI to produce accurate and trustworthy results, the ability to scale across clouds, and to easily adapt to enterprise domains and use cases. Watsonx is designed to help them address those needs.
Let’s put AI to work and make the world work better — together.
For more details, please visit and explore www.ibm.com/watsonx
Statements regarding IBM’s future direction and intent are subject to change or withdrawal without notice and represent goals and objectives only.
i Source: https://www.finastra.com/financial-services-state-nation-survey-2023
ii Source: https://www.ibm.com/products/watsonx-ai/foundation-models