Five key factors blocking widespread AI implementation in organizations
47% of CIOs indicate that AI projects are not meeting expectations, according to a recent Gartner survey.
47%. Nearly half.
Why? Cost. Security. Hallucinations.
In our recent Service Management Unlocked (SMU) webinar, Xurrent’s VP of Artificial Intelligence & Integrations Engineering, Scott Hutchins, dug into the “why” with a specific focus on the security aspect.
But first, Scott broke down the three main buckets of AI.
Understanding the main AI learning models
There are three main types of AI learning models:
1. Supervised machine learning
2. Unsupervised machine learning
3. Generative AI
Supervised machine learning is a human-curated master dataset of truth. It uses an input-output model to train that data. Over time, the algorithm learns to map inputs to known outputs. This model is generally the most straightforward to train and evaluate.
Scott’s Hot Take: This approach can take time to explain to security teams due to the many different techniques. While typically pretty secure, various configurations and customization often mean longer reviews … which can become “too busy to look.”
Unsupervised machine learning uses unlabeled data to find “hidden” patterns/structures without predefined outputs. It predicts the output based on historical data.
Scott’s Hot Take: Most organizations can adopt this approach without concern because it behaves so much like search and not like traditional training models.
Generative AI creates new content based on training data. From an intellectual processing standpoint, this data is equivalent to someone with a high school education. The result is novel outputs rather than classification/prediction.
Scott’s Hot Take: While this is the most complex and sophisticated model, it is a standard interface used across most enterprises. Therefore, most organizations allow adoption within a predefined set of rules.
So why isn’t AI everywhere within today’s organizations?
The main reason organizations are not adopting AI
AI is revolutionizing our world! AI is pure magic in action! AI is changing everything we know!
It is difficult to go anywhere online or in person without hearing about AI. While organizations are certainly adopting AI, nearly half of CIOs indicate that AI projects are not meeting expectations.
Conventional wisdom suggests that more AI means more expenses. But not so fast. The price of Large Language Models (LLMs) tends to decrease as new models are launched.
So, if it’s not about price, what are the main barriers? The disconnect, we believe, is due to five main factors:
1. Technical security
2. Administrative compliance
3. Privacy
4. Trust
5. Usability
Scott dug into these factors during the webinar (watch now to hear them all), sharing that while large language models can be scary, the risk can certainly be mitigated.
The key: Every technology leader needs to understand what questions to ask to bound risk for each of the 3 types of AI.
Xurrent’s approach to AI
Xurrent leverages the generative AI models, all housed within AWS Bedrock — very consistent with how we deliver the rest of our technology suite.
As Scott shared, Xurrent AI does not introduce any new technical cross-section or vector for attack.
When it comes to AI, we do not …
- Store any new data
- Retain data for AI.
- Use new 3rd party subprocessors for AI.
Our privacy remains the same. AI respects all existing permissions. Humans cannot access data … unless submitted as AI feedback within the limited scope of feedback features.
At Xurrent, trust is earned. We deliver the AI transparently — clearly sharing which decisions are made by AI vs humans.
Equally important, usability is about understanding the customer’s problem rather than adding AI for the sake of having AI. We’ve built AI around the use cases in which it can deliver value. We’ve optimized for where AI can reduce the burden on humans, saving time and money.
But wait. There’s more.
SNEAK PEEK: Xurrent’s AI Virtual Agent
The webinar wrapped with a “sneak peek” into Xurrent’s AI Virtual Agent … some fascinating Q&A, and even a conversation about ring-fencing.
Ring what? Exactly.
Watch the full recording to learn more (about ring-fencing) and how to de-risk AI in your organization.
