Day 2 of KMWorld Connect 2020 opens with a focus on AI
KMWorld Connect 2020 began its second day with a slate of keynotes focused on how AI is changing the KM landscape.
David Weinberger, author, “Everything is Miscellaneous,” “Too Big to Know” and his latest, “Everyday Chaos: Technology, Complexity” discussed “Knowledge, Complexity & Possibilities: Not Knowing in the Age of AI”
Additional keynotes were presented by Paul Nelson, innovation lead, search and content analytics, Accenture; Robert Pashinsky, director, metadata solutions, Dow Jones; and Christophe Aubry, managing director, Expert.ai.
KMWorld Connect, November 16-19, and its co-located events, covers future-focused strategies, technologies, and tools to help organizations transform for positive outcomes
AI, the internet, and chaos
“We interpret ourselves in terms of our dominate technology,” Weinberger said.
The net has conditioned us to a particular type of chaos and AI is giving us a new model to interpret that chaos, he explained.
The idea under the internet itself is inter-operability, which is making resources available in the ways that the owners of the resources did not anticipate.
“It’s as if we spent 20 years making the world more unpredictable,” he said.
The internet is a complex dynamic system that is sensitive to initial conditions. On the internet, users see small events that can lead to big changes. For example, something that goes viral could wind up changing the world, like a butterfly effect, he said.
“We’ve been living in chaos and we’ve been enjoying it,” Weinberger said. “We think of it now as normal.”
Machine learning and AI has emerged to help us sort through the chaos.
Machine learning’s natural state is being a “black box.” However, this can inevitably reflect the bias in culture or businesses.
“This is a huge problem,” Weinberger said.
This has generated a moral panic because people don’t know how machine learning really works, he said. These systems can produce more accurate results than humans can.
“The world is really the black box,” Weinberger said. “Machine learning can help us figure this out. Our ability to comprehend is insufficient for what we want and need to do.”
He suggested 5 things knowledge managers can do:
- Strategies are overrated-strategies require a particular future. It’s already changing. A minimum viable strategy is where organizations are headed.
- Everyone should be a sensor-people should become more observant.
- It takes a network to make sense
- Make more knowledge possibilities-supporting existing standards, contributing to open source and access, learn in public, and interoperate.
- Reify knowledge-turn knowledge into a thing or making the abstract real.
Soon truth will reside in linked open data, knowledge graphs, and AI models, he predicted.
AI models learn by being used, which is new for a body of knowledge. It is important how people treat and use them.
But there are caveats. This method can reify bias, could centralize information, is probabilistic, can cause human “embarrassment.”
Reifying AI with KM
Intelligent document understanding is drastically changing the search and knowledge management landscape thanks to AI technologies.
As 80% of all enterprise data is unstructured, document understanding delivers tangible benefits across industries and business functions saving time, money and resources.
Reifying AI with KM, Nelson explained, can be seen in several examples. One is customer support where customers need answers to questions and the company can create a place where customers can get answers. Search can be used as the company adds document management and workflow. Customers can add comments and contribute to the discussion. Machine learning can identify hot topics and propose those to the tech writer. The tech writer could be replaced by a robot writer in the future.
Another example is research reports. Product managers need a materials test so company looks to create document management and approval workflow. This can create duplicates so a search engine can be created to cut down on those. A gatekeeper can come in and make sure things get done. This can be automated with AI. The gatekeeper can be replaced by a robot, which needs to be trained on science so it can find and produce documents that are needed.
The last example is proposal writing, he explained. How do you respond to RFPs and RFIs? Companies contact a variety of experts to get this done. This can be improved by creating a collaborative space so all pieces can be put together in one place, like Microsoft Teams. Then these proposals can get tagged with business data. A company can then use a neural network to understand the pieces.
“That just an idea of how we’ve used AI to help with knowledge management,” Nelson said. It is just more than classification and entity extraction.
Successful projects will improve the existing knowledge flow and include targeted AI, he noted.
AI solves problems in the real-world
Pashinsky and Aubry closed the keynote session with a talk on “Practical Semantic Hybrid Solutions.”
The artificial intelligence (AI) market is growing rapidly, gaining mindshare in the C-suite, and moving from early adopters to the mainstream as businesses begin to realize its value.
This extends to AI approaches to extract insight from language through natural language understanding and processing (NLU/NLP) to make better decisions and automate processes.
However, often the analysis and the decision process are still too focused on which technique to use rather than on pragmatically approach AI as a typical software selection process.
“We believe AI should be accessible and practical,” Aubry said.
Practical AI can lead to the seamless integration of multiple techniques in one platform to solve your business challenges.
Pashinsky presented a real-world example that included Dow Jones’ FACTIVA platform. A user can tap into the platform to access news and industry coverage in multiple languages.
The archive has some limitations, however, Pashinsky said. Users need to read and understand unstructured text. Language can be difficult to understand by data classification tools.
Using Dow Jones’ Intelligent Identifiers, the taxonomy evolves over time as current events unfold, Pashinsky said. Metadata enrichment is done by machines by using machine learning, NLP, and rules based engines. The metadata tools are supported by expert.ai and underpinned by knowledge graphs.
Replays of KMWorld Connect webinars will be made available for on-demand viewing on or about November 24.