Why AI Needs Good Semantics, And What to Do About It?
That’s Just Semantics
How often have you heard someone exasperatingly state, “That’s just semantics,” when claiming that one word is just a synonym for another?
That’s a reasonable statement in light conversation, but a misinformed one in today’s complex data management landscape.
Semantics Matter, A Lot!
Semantics matter a lot when it comes to data, particularly that which informs large language and other AI models!
That’s because similar entities can seem like synonyms when they aren’t. For example, customer and prospect, where one is an organization that has already purchased our product, and the other is an organization you hope to sell to. Your actions to maximize the value of each are distinct.
Further complicating IT semantics is that a single entity can be multiple similar things. For example, an existing customer who purchased product A may also be a prospect for product B. As such, this organization is both a customer and a prospect.
Another complication is that the “same” customer is often stored and managed differently in your
-
Billing and receivables systems
-
Sales force automation system
-
Marketing automation system
-
Customer service system
-
Other transactional and intelligence sources
This is a huge challenge, one that most organizations struggle with.
Relationships Matter Too!
Beyond understanding the semantics of your key entities (customers, citizens, products, suppliers, assets, sites, etc.), you must also understand how these entities relate both in the real world and throughout your many source systems and databases. For example, which suppliers deliver to which sites? Or which salespeople call on which customers. Thus, relationships magnify your semantics challenge exponentially.
And What About Unstructured Data Semantics?
Unfortunately SQL data source semantics and relationships are only half your challenge when informing and training your large language and other AI models.
Text, image, and video sources add valuable contextual data. In fact, according to the IDC white paper, Untapped Value: What Every Executive Needs to Know About Unstructured Data, unstructured data makes up 90% of your business data.
Further, natural language or text-based query support are common user interfaces in most AI applications today.
Many organizations that are SQL-centric in their skills stumble on this critical unstructured data dimension.
How Can You Overcome the AI Semantics Challenge?
Complex problems require complex solutions.
Because knowledge graphs capture complex structured and unstructured data relationships visually intuitively, many experts believe modern semantic layers powered by knowledge graphs are the best way to address your AI semantics challenges.
To help guide you in your AI semantics journey, we have assembled three experts:
-
Fellow Insight Jam Expert, Nicola Askham, Data Governance Coach at Nicola Askham Ltd,
-
Juan Sequeda, Principal Scientist & Head of AI Lab at data.world
-
Sumit Pal, Strategic Technology Director at Onotext,
for the next Insight Jam Session: Why AI Needs Good Semantics, And What to Do About It? on Friday, October 4, 2024, at Noon ET.
What You Will Learn
At that session, the panelists will address these critical questions:
-
Why are semantics extra challenging in the AI domain?
-
How can knowledge graphs help improve LLM and other AI model’s effectiveness, accuracy, performance, and more?
-
How do knowledge graphs enable and empower semantic layers for AI applications and beyond?
-
Which benefits are proving most valuable?
-
Where are the best places to start your AI semantics journey?
See you then and there.
