IBM Just Made AI Easier for All Companies

Face it: building an AI application is not for the faint-hearted.

Large enterprises may have the bandwidth and budget depth to fail fast, but smaller companies can’t afford to get it wrong—even once.

The problem is that doing AI right requires labeled data, deep pockets and time that many can ill afford.

IBM, a long-term proponent of AI through its IBM Watson mantra, offers an alternative: foundation models.

With its watsonx makeover, it aims at the biggest challenge facing AI projects: the lack of labeled data.

Making the AI journey less daunting

Speaking at a media briefing on the sidelines of IBM Think Singapore, Sriram Raghavan, vice president for IBM Research AI, noted that getting labeled data was never easy.

“Either you have it, or you must be willing to spend money to collect it. For enterprises, they need time to curate the data, and that limited the set of use cases," said Raghavan.

So, IBM took its previous Watson lessons to roll out watsonx. Through three major components—watsonx.ai, watsonx.data and watson.governance—the company is making it easier for any company to develop the right use cases without having labelled data.

“What's very exciting to us is the ability to build and retrain without labeled data,” said Raghavan.

“[Foundation models] open up the possibility that I can scale AI dramatically without having to pay the cost. So the ability to commoditize that cost across a base model is really very exciting,” Raghavan added.

There is another reason for companies to be excited about IBM's promises: workflow agility.

The world is waking up to the impact of AI, thanks to Open AI letting their ChatGPT loose onto the world. And it’s not just customers; regulators and governments are also starting to get involved.

Foundation models offer a way for companies not to return to the drawing board and do everything from scratch as new guardrails and regulations come online. IBM is allowing them to create models for specific industry use cases based on the enterprise data that many companies already have.

"What we are going to see for generative AI and foundation models is a workflow in which you often start not with base data but with a base model, potentially provided by a provider or third-party open source. And then you tune and adapt with enterprise data," said Raghavan.

Many to rule them all

IBM is going all in on foundation models.

Chetan Krishnamurthy, chief marketing officer and vice president for marketing and communications at IBM APAC, discussed how foundation models could help companies do large language models and classical AI (AI models based on a narrow focus).

The challenge is that many companies are already working on different models. So, IBM's answer is to lean on their industry partnerships and alliances to build a model ecosystem.

In August this year, IBM said it would offer Meta's Llama 2-chat 70 billion parameter model to select clients. It is also hosting StarCoder, a large language model for code, including over 80+ programming languages, Git commits, GitHub issues and Jupyter notebooks.

IBM is also creating and releasing new foundation models as we speak. Called Granite, these are multi-size foundation models that use generative AI for languages and coding.

Currently, Granite models use data from the internet, academia, unstructured code data sets, legal, and finance. “And by bringing these to enterprise, we get to both enable new generative AI use cases as well as accelerate the adoption of traditional AI use cases,” Raghavan explained.

Companies strapped for resources can then train these models on specialized data sets so that the outcomes are based on relevant industry knowledge. In addition, Granite models are trained to look out for hateful and profane (HAP) content, so you do not have to employ an army of content reviewers.

The future lies in model partnerships

What makes IBM's approach different is that it aims to work with all significant AI players and make the work available for all. And when they work with a major client, they then open-source it. Agnes Heftberger, general manager for IBM Australia, Southeast Asia, New Zealand, and Korea, highlighted the example of NASA.

For companies, this means that they can fast-track their AI journey and pivot or capture AI-driven opportunities quickly. And they can now do that with the data they already have or have access to.

But IBM is not stopping there. During the media briefing, Krishnamurthy talked about how regional companies can do application modernization and IT automation quickly without paying huge consultancy fees with IBM watsonx Code Assistant.

IBM watsonx Code Assistant offers pre-trained models based on specific programming languages. One model is IBM Z application modernization, which allows companies to translate COBOL code to, for instance, Java. Compared to previous modernization tools, it does not skimp on IBM Z's value proposition or produce Java code that is hard to maintain. Plus, IBM is backing it up with support with IBM Consulting helping companies with difficult modernization projects. 

Another beneficial use case is Red Hat Ansible Lightspeed. It's a generative AI service for Ansible content and resulted from Red Hat's Project Wisdom, a collaborative effort with IBM to add AI capabilities.

“So we want to bring the best of proprietary models, third party ones and open source ones to our clients and give them the ability to choose the best one,” concluded Raghavan.

Now, it’s up to companies to choose their next AI step.

Winston Thomas is the editor-in-chief of CDOTrends and DigitalWorkforceTrends. He’s a singularity believer, a blockchain enthusiast, and believes we already live in a metaverse. You can reach him at [email protected].

Image credit: iStockphoto/Mykyta Dolmatov