AI Cartographer

Use our new AI tools to process layers and give them a cleaned up layer title, attribution, and cartographic settings that match your data.

When uploading new layers to SeaSketch, you have the option to process layers using our AI Cartographer system. This system can assign layers a well-formatted title, an attribution if metadata is available, an appropriate cartographic style, and interactivity settings (e.g. popups). This can save you a great deal of time, and is also a great way to learn how our cartography tools can be used to fit your visualization needs. As always, you can use the cartography tools to adjust these automated settings afterwards as needed.

In this example, the system identified a categorical color scheme driven by the Morphotype column as the most appropriate way to present a Seamounts layer.

Here, the system identified a raster as containing bathymetry data and chose a blue color scheme where deeper areas are a darker color. It reversed the default color palette order in order to show shallower areas in a lighter shade. Also note that the system cleaned up a complex filename to produce a clear layer title.

Enabling AI Processing

AI processing is a feature enabled or disabled for each individual SeaSketch account. Individuals have control over whether their uploaded data is processed.

With the launch of this new feature, you will be prompted to enable or decline the use of this system with your next upload automatically. If you want to change your settings later, you do so from the Overlay Layers section of your project's the admin interface.

Choose Settings -> AI layer processing... from the Overlay Layers admin to enable or disable these features.

Using with Sensitive Data

In order to provide AI services related to an uploaded layer, SeaSketch sends metadata about it to a 3rd-party data processor, OpenAI, to analyze it. The information sent to this service may include:

• The filename

• XML or textual metadata included with the layer

• The number of bands, min, max, and histograms of value distributions for raster data

• The geometry type, and number of features in vector data layers

• Column stats (min, max, avg, histograms) and sample values (e.g. Morphotype and SeamountId values from the example above).

SeaSketch does not send complete data layers to processors for inference, and for sample string values it will shuffle records to make reconstruction infeasible. SeaSketch also scans incoming data for personally identifiable information (PII) on local servers. Columns identified as likely to contain PII such as person names, emails, phone numbers, or government identifiers are redacted before sending to AI services. You can read more about how we redact PII here.

Details on how our data processor treats this data can be found on their website: OpenAI: Your data in the API. In summary, this information is not used for training OpenAI models and is retained only for logging and abuse prevention. Typically for less than 30 days. This is very similar to the policies related to other paid cloud computing services SeaSketch utilizes.

Even though we make every effort to prevent sharing of PII, SeaSketch cannot detect all sensitive information. Person names can be difficult to detect in global languages, and there are many other data values (like sensitive coordinates) which we cannot automatically redact. For that reason, for truly sensitive data we recommend disabling AI processing for your account when uploading those particular layers. This avoids metadata and sample values being sent to 3rd party data processors like OpenAI. You can always re-enable these features when uploading public data later.