Lucky analysts spent a day with IBM leadership last week, including the charismatic Director of Research, Dario Gil, the pragmatic experience of the SVP of software, Rob Thomas, and the leader of generative AI consulting, Matthew Candy, among other execs from IBM Research.
IBM’s goal was to align analysts’ view of IBM with the new reality of the AI business with a sneak peek at the Quantum realm before the IBM Quantum Summit in December. While you will have to wait until I attend the summit to hear those details, I can share a few things I learned on the AI side now.
The IBM AI and Cloud Strategy
As everyone knows, AI starts in the cloud. And since IBM’s enterprise client base is ramping up its AI efforts, its IBM Cloud, with a sizable GPU farm for training and new IBM AI inferencing processors, plays a critical role. IBM’s strategy is based on creating a library of tested and optimized Foundation Models, built with a watchful eye to optimize and annotate to filter hate, abuse, and profanity (HAP). IBM then takes these models, along with the governance and data preparation tools, directly to clients and makes them available via IBM Consulting clients.
As usual, IBM’s slideware is a bit busy, but as you will see, IBM has already done a ton of work to help clients along their journey to increased productivity, superior results, and lower costs through the careful selection of AI models.
The Big Three opportunities, with dozens more.
While IBM shared over two dozen use cases from their client engagements, the company says three stand out in traction and ROI: Digital Labor, Customer Care, and App modernization.
The specific use cases were startling. A year ago, most people barely knew what a foundation model was. Now, hundreds of IBM clients are fine-tuning models and beginning to deploy them with a full stack of hardware, data services, AI platforms, SDKs, and AI assistants. The three use cases outlined above are the low-hanging fruit nearly every enterprise can build and deploy. IBM has already benefited greatly, and almost everyone I spoke with at breaks and dinner talked glowingly about how much time they save using Ask HR, which automates many management tasks.
The watsonx platform has expanded dramatically, with new foundation models, AI tools, and assistant applications, with apps to discover, tune, govern, code, orchestrate, and improve data quality (the input) to yield significant business value.
Interestingly, the latter includes AI models that identify and filter hate, abuse and profanity (HAP) from input data. This data pre-processing step has been recently achieved using IBM’s proprietary inference accelerators, AIUs. These new accelerators are fast and ultra-power efficient and are an IBM Research prototype. While IBM has not publicly stated their intention of selling AIU’s, I saw a startup (NeuReality) with IBM AIUs in their booth at SuperComputing ‘23. (More on NeuReality coming next week!)
The IBM Garage method enables business transformation from concept to user adoption while tracking the value impact at every stage. It de-risks critical investments and enables long-term transformation through transparent measurement.
The circles shown illustrate the method by which Garage achieves this step by step. This year IBM embedded generative AI into these steps to further speed clients’ time to value.
1. Co-create (blue circles) – a combined team of diverse subject matter experts are immersed in intensive design thinking and research to expose the true nature and value of a client’s opportunity. This phase establishes alignment on a “big idea” and creates a vision for a minimum viable product (MVP) and its value.
2. Co-execute (middle red circles) shows the solution development cycle that uses DevOps and Lean practices to launch and test an MVP quickly. The goal is to validate and improve the MVP’s value in the marketplace through iterative testing, measuring and re-launching.
3. Co-execute (the green circle on the right) – hardens and scales the solution and the new culture of innovation across the enterprise. This phase is the time to broaden feature sets, stress test code, strengthen security and resilience, deploy solutions widely, and expand capabilities to continue to transform.
The results outlined on the right come from IBM Garage engagements. A 102% ROI is unheard of in the industry, all at 67% faster outcomes. Wow.
And IBM is just getting started. This month, IBM will GA the watsonx.ai for on-premises deployment to augment the IBM Cloud. Next month, IBM will release a new watsonx.orchestrate with over 1000 out-of-the-box skills, along with watsonx.governance and Tuning Studio. The two biggest inhibitors to enterprise adoption of AI are inadequate governance and the HAP mentioned above, which IBM is aggressively addressing.
Ok, I warned you earlier that IBM slideware can get a little verbose, right? See below. While this slide may take 20 minutes to read and comprehend, the lower third is compelling. This slide is only about recruiting and HR management; IBM has similar slides for other areas of AI impact. This is the tip of the iceberg. 85% first touch resolution? 66% reduction in human support for HR queries? 50% reduction in attrition? Seriously? Yeah. IBM’s got this and much, much more.
While the original Watson from IBM was not a success, the reincarnation of watson as a comprehensive AI platform that enables successful service engagements is beyond “impressive”. When they first announced it, I yawned. Now I am blown away by how watsonx is changing the IBM Company into an AI powerhouse and transforming IBM clients’ business.
The analyst community, in general, is a fairly skeptical crowd. But I can tell you everyone there walked away impressed. Yes, IBM still needs to work out how it will leverage technology like the AIU, and we all need to come to terms with how AI will impact employment. One is a nit, the other is beyond my ken.
Ok, I’ll settle down now, at least until the Quantum Summit on December 4.