Generative AI is in all places, and the alternatives for builders to create new
content material and new experiences are countless. With that, it is important that we construct
with ethics in thoughts and create secure experiences for our
customers. We intend to make AI useful for all, bettering the online ecosystem and
consumer experiences.
Construct accountable web sites, internet apps, and extensions
Google is dedicated to
accountable AI practices.
No matter choices you make, make sure to doc your practices for the longer term.
Hold a report of the outcomes and greatest practices so that you just and your colleagues
can proceed to take care of and enhance present and future functions.
Websites, apps, and extensions that generate content material utilizing AI ought to think about the
following:
- Rigorously take a look at AI instruments and fashions.
- Use pre-existing security instruments and settings.
- Give customers a method to report or flag offensive content material,
then handle these stories. - Promote your product responsibly.
Rigorously take a look at AI instruments and fashions
You or your organization are accountable for the expertise in your apps, so it is
crucial that you just perceive the underlying AI fashions and expertise. It is
vital to check quite a lot of use instances and safeguard in opposition to prompts that
might create dangerous or offensive content material.
Use pre-existing security mechanisms
Some APIs, corresponding to
Vertex AI,
have instruments and configuration choices for security filters. Security filters act as a
barrier to forestall dangerous output, however they do not immediately change the mannequin’s
conduct. We suggest you assessment and set these filters to a stage you deem
applicable.
Allow offensive content material stories
Even with testing and security filters, it is attainable that content material that is
thought of to be offensive content material might make its means via to customers. Monitor
and prioritize consumer suggestions round generative AI options, to make sure you’re
shaping the very best expertise. Set up a course of to handle stories of
offensive content material.
To mitigate potential offensive content material, you possibly can carry out
a number of of the next actions:
- Block related inputs that led to the offensive content material
- Redact the mannequin output
- Advantageous tune the mannequin
- Change the mannequin solely
Should you construct browser extensions, at all times observe the related insurance policies. Chrome
Extension builders who want to publish their extension on the
Chrome Net Retailer ought to observe the
Chrome Net Retailer insurance policies.
Should you publish content material generated with AI, purpose to create
useful, dependable, and people-first content material.
Promote responsibly
When promoting, it is vital to take action in a means that is secure and respectful of customers. Assessment your advertising supplies to make sure that your advertisements precisely signify your web site or app’s precise capabilities.
Seeking to the longer term
We’ll proceed to offer greatest practices and share sources, such because the
Folks + AI Guidebook, to assist
builders in constructing progressive and accountable apps.
Share your suggestions with us at @ChromiumDev on X or
Chrome for builders on LinkedIn,
and discover all of our AI content material.