IBM’s watsonx could be a generative AI game-changer

Disclosure: IBM is a client of the author.

IBM this week announced watsonx at Think, and it has the potential to be a generative AI standout. That’s important because generative AI has hit the tech industry like a Mack truck and appears to be advancing at an unbelievable rate. Just as quickly, well-founded concerns about the quality of the massive training set behind it have emerged. First, the technology is very new to most people and the risks surrounding it aren’t well known. And second, we still don’t know how to ensure that what’s produced by generative AI tools is accurate.

In addition, much as with some initial problems with Linux, intellectual property issues surround generational AI and are scaring creators half to death. Basic productivity tools such as those used for editing, formatting, and making presentations appear relatively safe. But when tools like ChatGPT are asked to create or to provide decision support — or even make decisions autonomously — those IP issues are more pronounced. And the speed of adoption could lead to future problems if the issues surrounding quality aren’t adequately addressed.

That’s where IBM comes in. It’s one company that now has decades of operational AI experience, one that identified years ago what the current concerns would be and worked to mitigate them long before we heard the term generative AI. IBM, under Thomas Watson, Jr. decades ago, established a policy of trustworthy products and services, and has more recently been outspoken that AI should be used to enhance, not replace, employees.

IBM’s unique strength

IBM’s capability is rare in this area, given its long history with Watson, the AI platform tied to healthcare and diagnostic support. (It’s a market that requires putting accuracy and data protection first.) In addition, IBM was arguably the most powerful champion for Linux, helping to assure the underlying intellectual property behind the open-source software. IBM  eventually bought Red Hat, the most powerful and capable provider of Linux, giving it unique legal access to massive training sets without violating privacy (healthcare) or third-party intellectual property rights (Linux). 

IBM’s big enterprise and government focus means it has taken reliability, availability, and security seriously well before I joined that company in the 1980s. Its defenses against malware are near legendary, its z series mainframe platform is the most reliable and secure platform in the market, and it’s an industry leader in hybrid computing. Its unique, secure IBM Cloud differentiates it from the pack and could be something of a model for generative AI as it evolves.

As AI advances, it will need to embrace a hybrid model because moving and updating the massive datasets that support it will be impractical, and moving at least some of the processing closer to the user will be necessary to avoid latency. For instance, if you were to use this technology on a manufacturing line to assure build quality, excessive latency could lead to higher failure rates and lower production volumes.

IBM’s one shortcoming

IBM does has one major shortcoming: it no longer has a desktop presence. It sold its PC company to Lenovo in the early 2000s. However, it still has relationships with Intel, AMD (Intel and AMD are clients), and, at one time, it had a partnership with Apple. (Apple appears to have been caught napping by the rapid emergence of generative AI, but could team up with IBM to provide this capability and close what will shortly be an embarrassing technology gap.) 

With neural processing units (NPUs), visual processing units (VPUs) coming to market next year, the need for a desktop solution will only rise. That will require IBM to leverage one of its many partnerships — and it needs to do so before these partners get too invested in a different technology. 

While generative AI could revolutionize technology, like any new product, it is having some teething issues when it comes to quality and trustworthiness. IBM’s watsonx could address these concerns because it’s based on decades of research, past trust-assuring practices, and mainframe standards that should make it the standard against which other efforts are measured. If IBM can figure what to do about its non-existent desktop business, its watsonx stands as the first, best hope for a trustworthy generative AI solution.