As consumers, we understand that you must evaluate the risks and benefits of using AI in your business, including any implications that using AI tools may have on industry regulation requirements, data privacy, and information security. The speed with which some companies are developing and releasing new AI technologies, along with a lack of standardization, can make this even more challenging.
To create the most value for our customers, Omnissa is taking a systematic and responsible approach to developing AI capabilities within our products and services. As we have written previously, the Omnissa approach to responsible AI is based on maintaining your trust through governance, transparency, data privacy and security. In support of this approach, we considered the question:
What if there was a way to easily see how and where AI is being used in Omnissa products and services?
In response, the Omnissa AI Council developed a series of simple, readable AI nutrition labels, an emerging practice across the industry. Our nutrition labels cover the most frequently asked questions you, and your security teams, may have about our use of AI in the Omnissa Platform.
Much like a nutrition label on a cereal box, our nutrition labels cover the most important information for each AI feature we implement. You can think of it like the “what’s in the box” for each of our AI features.
The goal of an Omnissa AI nutrition label is to provide you with confidence in the tools you’re using and to help you make an informed decision on the benefits and risks of using each AI feature. This includes understanding how your data is used and by whom. Our hope is that these labels facilitate reviews and approvals of our AI features throughout your organization.
To better understand how to read Omnissa AI nutrition labels, let’s review each section of the Insights AI nutrition label below.
Feature name/Description/Products: In this section of the label, you will find basic information about the feature, including the name, a description of what it does, and where in our products and services it’s used.
Model type: This is the model type the AI feature uses – such as time series anomaly detection algorithm, a lift analysis calculation, generative AI (GenAI), and so on.
Model provider: Having full visibility into where your data is being processed and what company developed the algorithm or Large Language Model (LLM) goes a long way to in providing the confidence you need when determining whether to use an AI feature. The model provider section of the nutrition label describes whether the model is internally developed or uses a third-party generative LLM provider such as Azure OpenAI. In the case of Insights, Omnissa is using an internally developed AI algorithm.
Input data: We know you need visibility into exactly what data is being processed by an AI feature when you’re deciding whether to use it or not. That’s why we’ve included a section for input data. Input data is defined as the data that is used to generate a response from an AI feature. In this instance, Insights is using telemetry data from the Omnissa Intelligence data lake to alert you to interesting or potentially problematic data point changes. Depending on the AI feature, other types of input data could include user-submitted natural language queries.
Input data available for customer audit: Depending on the nature of the AI tool, you might want to see what your users are submitting to the feature. This section identifies whether the capability exists for you to audit the input data. In this instance, since Insights uses data from the Intelligence data lake, an option to audit is not available.
Adheres to data sovereignty: Omnissa understands that data sovereignty is a top priority for our customers. A “Yes” for data sovereignty means that data stored by the AI feature remains in your original storage region. For example, Insights data is stored in region with your Omnissa Intelligence data.
Trained on customer data: This section details whether the internal model or LLM is trained on customer input data. Note that, according to our internal AI governance policy, Omnissa does not train any models on customer data unless it has been anonymized.
Guardrails: This section details whether AI guardrails are in place. Guardrails are typically used with LLMs that use GenAI to help ensure the model is safe and its outputs are correct, non-biased, appropriate, and consistent. In this example, Insights uses an internally developed algorithm (not an LLM), so guardrails are not applicable to this feature.
Update frequency: This section details how often models are retrained. For example, Insights is retrained weekly, while Guided Root Cause Analysis (which is not trained on customer data) is updated as needed.
Data retention duration: In alignment with our privacy policies around data minimization, we limit data storage to the period that it’s needed. In this example, Insights data is stored with your Omnissa Intelligence data, so the data retention duration is the same.
Optionality: Every AI-based feature embedded within our products and services offerings has a way for the customers to opt-out of using the feature. In this example, Insights is part of the Experience Management solution within Omnissa Intelligence, so if Experience Management is enabled, then the Insights AI feature is enabled.
These labels can be found on Omnissa Docs in the section for each AI feature. This currently includes Insights and Guided Root Cause Analysis AI-enabled features in Workspace ONE® Experience Management:
As part of our commitment to responsible AI, we plan to publish an AI nutrition label for each feature powered by AI in our products. We hope this commitment to transparency allows you to make an educated decision on which features and products you choose to leverage–now and in the future.
此内容由惯性聚合(RSS阅读器)自动聚合整理，仅供阅读参考。 原文来自 — 版权归原作者所有。