Five high notes from Think 2019
At IBM Think 2019, enterprises embarking upon their AI journey were squarely focused on how to get data ready for successful AI deployments.
I attended Think during my second week at IBM. Here are five notes from the event that, in combination with adopting a prescriptive approach to AI (modernize, collect, organize, analyze, infuse), will strike a winning tone for the AI era.
1. Watson Anywhere
Watson AI is coming to your cloud of preference: public, private, hybrid cloud. It will even be on Amazon Web Services (AWS), Google Cloud, and Microsoft Azure. Watson remains available on IBM Cloud, as well as on-premises, and can connect seamlessly across all those options. This should be music to the ears for Watson users: train applications on-premises, run them on a private cloud instance or any combination you desire.
Together, these improvements make Watson AI the most “open, scalable AI for business in the world.”
According to Rob Thomas, GM of Data and AI at IBM, “businesses have largely been limited to experimenting with AI siloes due to limitations caused by cloud lock-in of their data.” With Kubernetes under the hood powering the key IBM microservices supporting Watson Anywhere, data can be accessed wherever it is. It’s harmonizing those discordant data silos. With barriers removed, innovation soars.
2. No AI without IA
An important refrain at Think was the dependence of AI on information architecture (IA). Without it, AI adoption will not prove successful. When preparing for AI, the most time-consuming part is building the foundation to scale your cognitive platform and applications.
To do that, data must be high quality, accurate, and credible. It’s important to see the full picture of your data and integrate information assets across all the various sources and repositories in which it resides. Most critically, data assets must be properly managed, prioritizing security, as well as in compliance with data governance policies and best practices. Effective IA also requires alignment of culture and provisions for change management.
IBM Director of Offering Management Thomas Chu said that databases suitable for AI must also be smarter. What’s needed is an “AI-infused database,” such as IBM Db2.
Expect IA to continue as the most critical background player when it comes AI. Learn more about Db2, the IBM database “built for and powered by AI”.
3. Data ethics
Responsible stewardship, which includes the ability to properly govern an endlessly crescendoing volume of data, was also a chorus at IBM Think.
No matter the volume, every scrap of data that feeds machines, extracts insights through analytics, and teaches AI must be used responsibly. Use of customer data should trend towards several guiding principles, such as possessing a clear and traceable lifecycle and ample attention given to explainability. In equal measure, digital systems and services should be designed in a way that prioritizes transparency and consent.
Emphasis on fairness by keeping bias out of data collection practices, analysis, and algorithms is crucial as governance becomes less a check box to avoid fines or court dates and more of a legitimate source of competitive differentiation.
4. One-shot learning
Another important accompaniment to IBM message for the future of AI and data is using less data to train machines and power cognitive services. This approach, called one-shot learning, is important for a few reasons:
- Deep learning is presently one of the best ways to train AI, especially in lagging disciplines such as cognitive reasoning.
- Strategies employing deep learning tend to require massive amounts of data and piloted operations.
- Often, key domains don’t have enough data to train AI.
One-shot learning brings efficiency to the machine learning process and requires only a fraction of the necessary data and training cycles to prove effective. This cuts time for AI to learn, frees up human pilots that are running training protocols and gets nearer to unsupervised learning for AI machines. An AI that learns by itself or with minimal data aids could provide positive outcomes in industries with low-volume data sets, such as healthcare.
It’s still a developing field with exciting prospects, so expect more at IBM Think 2020.
5. AI for good
For a criminal justice department dropping local prison populations by 35 percent in two years, a car manufacturer that learned the art of conversation through chatbots, and a concert venue using AI and data to maintain respectful and sustainable use of the environment, AI is making the world more just, delightful, fun, equal, comforting, safe, interesting, inspiring and fair. It’s altogether a better place, even if only by a small margin.
It’s important to strive toward the net good AI can create in people’s lives and in the world.