Making Data Simple

Unveiling the Power of AI and watsonx: In-Depth Conversations with Ruchir Puri, Chief Scientist of IBM Research

Wed Jul 19 2023

AIQuantum ComputingIBM WatsonGenerative AIModel GovernanceModel ParametersOpen Ecosystem

Description

Dr. Rishir Puri, chief scientist of IBM Research and IBM Fellow, discusses his extensive experience in AI and quantum computing, his work on IBM Watson, the current wave of AI, generative AI, IBM's approach to model governance and specialization, model parameters, and the power of an open ecosystem in AI.

Insights

AI Revolution Rests on Data, Compute, and Algorithm

The current revolution in AI is built upon three foundational pillars: data, compute, and algorithm. These pillars enable scalability, allowing AI models to learn from unstructured data without manual labeling.

Generative AI Solves Use Cases When Combined with Other Technologies

Generative AI is not a magic bullet, but it can solve a significant number of use cases when combined with other technologies and solutions. It opens doors for solving various tasks in customer care, software development, security, and more.

IBM's Watson X Offers Trusted Data and Curated Models

IBM's Watson X platform provides trusted data and curated models from open source and IBM-supported sources. It focuses on trust, responsibility, tuning, training technology, and specialized models to address pain points in the digitization of enterprises.

Specialized Models Provide Mastery in Specific Domains

Highly tuned models are better suited for specific domains and tasks, providing mastery in those areas. For example, a highly tuned model for translating Cobol to Java achieves 95.9% accuracy compared to an 18% accuracy of a chat GPT version.

The Power of an Open Ecosystem in AI

Enterprises have a choice between one model to rule all or an open community and the power of an open ecosystem. IBM's bet is on an open ecosystem, as demonstrated by their investment in Red Hat.

Chapters

Introduction
Passion for Algorithms and Optimization
IBM Watson and the Current Wave of AI
Generative AI and the Foundations of AI Revolution
The Power of Openness in AI
IBM's Approach to Model Governance and Specialization
Model Parameters and IBM's Watson X Platform
The Power of an Open Ecosystem in AI

Summary

Transcript

Introduction

00:03 - 07:41

Dr. Rishir Puri, chief scientist of IBM Research and IBM Fellow, is the guest on this episode.
Dr. Puri has extensive experience in AI and quantum computing.
He led IBM Watson as its CTO and Chief Architect from 2016 to 2019.
Dr. Puri has a background in software hardware automation methods, microprocessor design, and AI algorithms.
His passion is making things and he loves hiking and collecting baseball caps.
He started collecting baseball caps during hiking trips to national parks.
Dr. Puri always wears a baseball cap, even when visiting clients.
His brand is technology and he enjoys learning at every level of abstraction in the IT stack.
He has worked on radiating chips with alpha particles, designing circuits, building microprocessors, writing books and papers, and contributing to products like Watson.
Dr. Puri considers himself lucky to have experienced and contributed across the depth and breadth of the entire IT stack.

Passion for Algorithms and Optimization

07:18 - 15:12

The speaker's passion lies in algorithms and optimization techniques.
They have been applying these techniques to various applications, including designing chips.
Automation is becoming increasingly important in chip design due to the complexity involved.
The speaker emphasizes the importance of balancing technical expertise with leadership responsibilities.
They protect their time for staying connected with technology and defining what comes next.
Friday afternoons are dedicated to deep learning and research.
Experimentation and building apps are done throughout the week in smaller blocks of time.
The speaker advises technical leaders to protect their time for staying grounded and connected with technology.
The podcast mentions IBM Watson's success in beating Jeopardy contestants in 2011.

IBM Watson and the Current Wave of AI

14:43 - 22:47

IBM Watson has faced criticism for not delivering on some of its promises in the past.
IBM is launching a new AI platform called Watson X.
The criticism of Watson's original hype is partly fair, but there have also been successful use cases and broad adoption in customer care scenarios.
Pushing the limits of technology is necessary, even if it leads to challenges and valid criticism.
The current wave of AI is different from previous ones because it has reached billions of people and allows for seamless natural language interaction with machines.
Governments are starting to pay attention to AI as a societal topic.
The technology will go through cycles of hype and settling down, but each plateau will be significantly higher than before.
Machines are far from being able to reason like humans or pose existential threats.
Machines need to become multi-modal and develop reasoning abilities before they can reach the level where concerns about taking over the world arise.

Generative AI and the Foundations of AI Revolution

22:21 - 30:15

Existential threats posed by machines taking over the world are hyped up and unlikely to transpire soon due to regulations and government oversight.
Generative AI is not a magic bullet but can solve a significant number of use cases when combined with other technologies and solutions.
The current revolution in AI rests on three foundational pillars: data, compute, and algorithm.
Transformer architectures allow AI models to learn from unstructured data without the need for manual labeling, enabling scalability.
Large language models and generative AI open doors for solving various use cases in customer care, software development, security, etc.
IBM's solution differentiates itself by providing transparency and addressing concerns around data usage and potential risks to enterprises.

The Power of Openness in AI

30:01 - 37:36

Google is part of the triangle of AI players, but their approach is either you are here or you're nowhere else
IBM and Meta are influential AI players innovating in AI
Huggingface has become the largest marketplace of AI models with 200,000 models available
The collective power of smart graduate students drives innovation in AI
Open movement is catching up to proprietary models
Specialized models trained on a good enough base model are beneficial for specific use cases
Base models need to be good enough, domain data provides expertise and accuracy for tasks
Generative AI models can absorb unstructured domain knowledge without labeling data
IBM's Watson X offers trusted data and curated models from open source and IBM-supported sources

IBM's Approach to Model Governance and Specialization

37:07 - 45:01

IBM collects, curates, and governs data for their models in Watson X.
IBM has both open source models and specialized models that they stand behind.
They generate fact sheets for their models and data sheets for the data that went into those models.
IBM's differentiation comes from the trust of AI and the governance of data.
IBM has developed a model to generate code from natural language in the IT automation domain.
They are also working on translating legacy languages like Cobol into modern languages.
Another model they are building is for summarizing enterprise tasks in natural language.
Due diligence on the use case is important before applying AI solutions.
Building fine-tuned models on top of a base model requires ensuring the foundation model is governed and built with trusted data.
Enterprises worry about their data getting leaked to large AI models like OpenAI's GPT-3.
IBM focuses on trust, responsibility, tuning, training technology, and specialized models to address pain points in digitization of enterprises.
The number of parameters in a model does not necessarily determine its proficiency.

Model Parameters and IBM's Watson X Platform

44:40 - 52:44

The model with 175 billion parameters is very generalized and has ingested a vast amount of public data.
For specialized tasks like language translation, a smaller model with 350 million parameters can achieve higher accuracy compared to the larger model.
The highly tuned model for translating Cobol to Java achieves 95.9% accuracy, while the chat GPT version only achieves 18% accuracy.
Highly tuned models are better suited for specific domains and tasks, providing mastery in those areas.
IBM's approach includes guardrails and content grounding to reduce hallucinations in AI models.
Watson X is a platform consisting of Watson X.ai, Watson X.data, and Watson X.governance.
.data enables querying and governance of large amounts of data, while .ai provides curated models and fine-tuning capabilities.
.governance focuses on the trust and responsibility aspects of AI models, including tracking data sources and deployment details.

The Power of an Open Ecosystem in AI

52:14 - 57:00

Enterprises have a choice to make between one model to rule all or an open community and the power of an open ecosystem.
The power of openness and transparency is important in AI.
IBM's bet is on an open ecosystem, as demonstrated by their investment in Red Hat.
If you haven't started on the AI journey, it's important to start now as it will reshape every industry.
Strategic thinking and identifying use cases are crucial before implementing AI solutions.
Partner with someone who understands the technology and has experience in enterprise use cases.
Reach out to Rishire Pire on LinkedIn for more information.