---
title: What is InteractivAI?
description: InteractivAI
---
[Skip to content](https://go.bluware.com/adoption-module-0-what-is-interactivai#main-content)
English
Show submenu for translations
[Submit a ticket](https://go.bluware.com/kb-tickets/new?hsLang=en) [https://go.bluware.com/](https://go.bluware.com/tickets-view?hsLang=en)
[](https://go.bluware.com/?hsLang=en)
- Solutions
Show submenu for Solutions
- [VDS](https://bluware.com/data-solutions/vds/)
- [FAST](https://bluware.com/data-solutions/fast/)
- [InteractivAI](https://bluware.com/data-solutions/interactivai/)
- [Services](https://bluware.com/consulting-services/)
- Resources
Show submenu for Resources
- [Use Cases](https://bluware.com/resources/use-cases/)
- [Resources](https://bluware.com/resources/)
- [News](https://bluware.com/resources/)
- [Support](https://bluware.happyfox.com/home/)
- Company
Show submenu for Company
- [About](https://bluware.com/about/)
- [Partnerships](https://bluware.com/about/partnerships/)
- [Events](https://bluware.com/events/)
- [Careers](https://bluware.com/careers/)
- [Contact](https://bluware.com/contact/)
Open main navigation
Close main navigation
- Solutions
Show submenu for Solutions
- [VDS](https://bluware.com/data-solutions/vds/)
- [FAST](https://bluware.com/data-solutions/fast/)
- [InteractivAI](https://bluware.com/data-solutions/interactivai/)
- [Services](https://bluware.com/consulting-services/)
- Resources
Show submenu for Resources
- [Use Cases](https://bluware.com/resources/use-cases/)
- [Resources](https://bluware.com/resources/)
- [News](https://bluware.com/resources/)
- [Support](https://bluware.happyfox.com/home/)
- Company
Show submenu for Company
- [About](https://bluware.com/about/)
- [Partnerships](https://bluware.com/about/partnerships/)
- [Events](https://bluware.com/events/)
- [Careers](https://bluware.com/careers/)
- [Contact](https://bluware.com/contact/)
- English
Show submenu for translations
- [Submit a ticket](https://go.bluware.com/kb-tickets/new)
- [https://go.bluware.com/](https://go.bluware.com/tickets-view)
- [Contact us](https://go.bluware.com/submit-ticket)
[Contact us](https://go.bluware.com/submit-ticket?hsLang=en)
How can we help you?
- There are no suggestions because the search field is empty.
1. [Bluware](https://go.bluware.com/?hsLang=en)
2. [InteractivAI](https://go.bluware.com/interactivai?hsLang=en)
3. [Getting Started](https://go.bluware.com/interactivai?hsLang=en#getting-started)
# What is InteractivAI?
*Applies to: InteractivAI version 6.0 Adoption Module 0*
This is the introductory module to InteractivAI. In this module, we explain what InteractivAI is, how it works, and what it is built to do.
Modern seismic interpretation faces two recurring challenges:
- Some targets are difficult for humans to interpret consistently because of geometric complexity.
- Some targets are easy to interpret but require enormous amounts of repetitive manual effort.
Traditional interpretation tools struggle with the first problem. Automated AI tools often struggle with the second because they lack geological context and user control.
InteractivAI was designed specifically to bridge this gap through continuous collaboration between the interpreter and the neural network. To help understand how, it will be helpful to quickly review the two primary approaches used to deploy AI in seismic interpretation: **automation-led AI** and **augmentation-led AI**. InteractivAI is an example of augmentation-led AI. The table below compares the key characteristics of both approaches.
| Automated AI | InteractivAI |
| --- | --- |
| Starts with a pre-trained model | Starts with your interpretation |
| Designed for one feature type | Feature agnostic |
| User validates results | User actively guides results |
| Hidden training assumptions | Learns directly from your labels |
| Goal = replace manual work | Goal = amplify interpreter expertise |
#### InteractivAI – What is it?
InteractivAI is a dedicated user- and data-centric deep learning application for accelerating complex1 or highly redundant2 tasks in the seismic interpretation workflow. InteractivAI is not a traditional seismic interpretation platform, meaning interpretation objects such as horizons, faults, and geobody surfaces are not actively interpreted directly by users but rather extracted from probability cubes created by training neural networks. This approach is ideal in scenarios where the target feature being interpreted has complicated geometries that are difficult to resolve using traditional interpretation techniques, such as sand injectites, or the target feature requires a lot of manual effort to generate a robust yet accurate interpretation, such as polygonal faulting.
> **The InteractivAI Principle:** If you can see the feature of interest in the seismic data, you can label it; and if you can label it, you can train a network to predict its geometry and/or distribution.
There are two key mechanical distinctions between training a network interactively in InteractivAI and training a network passively using a pre-trained, automated AI solution:
1. **Clean-Slate Networks** – New networks created in InteractivAI are empty; e.g., there are no pre-trained networks available for use in the application. This design is intentional and allows the network to train using the actual data covering the interpretation area, free of biases inherited from unrelated or synthetic training datasets typically found in pre-trained network models. InteractivAI networks learn what feature to look for directly from user-supplied labels, which are either manually painted on the seismic intersections or represented by a label VDS. Label VDS’ are binary volumes representing where features are located (value = 1) and where they are not (value = 2).
2. **Live Feedback Loop** – Once a network begins training, visible network predictions, or inference, appears every 30 – 45 seconds. The inference indicates the network’s predictions at the end of the previous training cycle, or epoch. Interpreters then interact with the network predictions by placing labels on top of inference results that are supported by the data and serve as solutions to the interpretation’s objectives, providing instant positive reinforcement feedback to the network, which incorporates it into the next training epoch. In this way, the network predictions are influenced by the human interpreter’s interactions, which in turn are tied to the objectives of the interpretation (e.g., interpret all basement faults or only a single primary bounding fault, etc.).
*Figure: The Live Feedback Loop in InteractivAI consists of labeling and prediction review stages (read clockwise from top).*
1 Complex tasks = sand injectite, volcanic sill, re-worked facies interpretation, etc.
2 Redundant tasks = simple horizon & unconformity, complete fault set interpretation, etc.
#### **Where in my interpretation workflow do I use InteractivAI?**
InteractivAI’s network flexibility makes it ideal for use during any process where manual interpretation is difficult and automated AI tools fail by either over-producing results or generating non-ideal interpretation geometries. Whether accelerating fault interpretation or resolving complex 3D geometries, InteractivAI is intended to improve both interpretation speed and confidence.
InteractivAI is intended to be used whenever speed and high-precision pattern recognition will add value to the interpretation cycle. Because the same underlying network architecture is used regardless of feature type, InteractivAI adapts naturally to a broad range of interpretation problems. This flexibility is another reason why InteractivAI is a more organic, seamless solution to AI-enabled seismic interpretation. When an automated, pre-trained model approach is used, in addition to unknown bias you’re also committing the network to a specific feature – think of the different fault or salt models available in black-box AI solutions and it becomes readily apparent that if you want to train a network for a feature that doesn’t have an associated pre-trained network, well – you’re out of luck.
InteractivAI removes this barrier through its feature-agnostic network design. The same underlying framework can be adapted to many interpretation problems simply by changing the labels and training data. This unique flexibility lends itself to accelerating interpretation tasks that other AI interpretation systems simply aren’t designed to handle – rapid AVO/AVA classification, re-worked facies distribution mapping, and 4D transfer learning are just some examples of advanced interpretation workflows that extend well beyond traditional structural interpretation.
#### What outputs does InteractivAI generate, and how are they generated?
Output information is generated in the form of probability cubes (binary network) or probability-class cubes (multiclass networks), which are then exploited by the system for feature extraction. Rather than interpreting horizons, a multiclass stratigraphy cube can be produced, and horizons extracted at the interface between two classes. Similarly, instead of interpreting faults in a grid-like fashion (every nth inline and every nth crossline) or trying to use unwieldy multi-z horizon interpretation tools, binary probability cubes for fault or geobody networks are generated and the required interpretation objects extracted from the probability cubes.
**To summarize:** the main output InteractivAI generates is probability or probability-class cubes. Features like fault sticks & faults, horizons, and 3D geobody surfaces are extracted from the probability output to generate interpretation objects, which can then be transferred back to your primary interpretation platform for integration into the larger holistic interpretation or static geomodel.
#### What puts the “Interactive” in InteractivAI?
InteractivAI’s unique rapid reaction-response behavior is enabled by two things:
1. The light and responsive 2D CNN network architecture
2. The VDS data format
##### 2D CNN network architecture
2D CNN network architectures have been around for a long time, and larger, smarter, more comprehensive 3D model architectures now exist, but the 2D CNN plays a critical role in keeping InteractivAI nimble and responsive. Without the large resource draw, InteractivAI can be scaled across organizations at an enterprise level without significantly increasing hardware spending or cloud compute costs compared to their newer but heavier 3D counterparts. When you place a label in InteractivAI, the label is identifying a pixel, which contains both an image and a value (single-volume network) or unique data relationship (multi-volume network). As the network learns from the 2D labeled pixels, predictions are made in 3D – in this context, InteractivAI is a “2.5D” network model, where the input is 2D but the output prediction space is represented in 3D.
##### VDS data format
VDS (Volume Data Store) is a bricked data format that’s optimized for interactivity via its random-access nature. Other bricked seismic volume data types are accessed sequentially – in some cases applications allow the pre-fetch of subset volumes of data to cache, but the data access remains sequential. This is non-ideal for deep learning and ML in general because the learning algorithms rely on a certain degree of randomness to generate predictions.
In a traditional deep learning workflow using sequential-access data formats, the data must be pre-processed by line extraction, segmentation, randomization, and 1D stacking before it’s ready to be consumed by the learning engine (in this context PyTorch or TensorFlow). VDS allows us to skip these interim processes, as the data is already optimized for random access, so no pre-processing is necessary.
What value can I realistically expect InteractivAI to deliver?
Adding InteractivAI to your seismic interpretation toolkit will deliver notable value in the following areas:
- **Faster interpretation cycles** – The augmentation-led AI approach used by InteractivAI acts as a force-multiplier for interpreters, shifting the interpretation cycle time for each set of geologic features from weeks – months to hours – days. Rather than using an “every nth inline x every nth crossline” interpretation strategy, InteractivAI is able to effectively extrapolate predictions from labels on an average of 1%–2% of available seismic lines. The quality of labeling is more critical than the quantity of labels, meaning less manual time is spent drawing interpretations on seismic lines and more time is available for critical analysis of what the network output can tell us about the geological conditions in the interpretation area.
- **Improved interpretation consistency** – Interpretations extracted from InteractivAI probability cubes generated from trained network models ensure survey-wide consistency of geologic features by removing artifacts caused by irregular distribution of manually-picked interpretation hardpoints commonly seen in manual interpretations.
- **Better handling of complex geometries** – InteractivAI’s multi-volume capability allows networks to train by looking for unique data relationships defined by labeled pixels. When using input data volumes decoupled from amplitude bias, this behavior can handle features with very complicated geometries that are difficult to resolve with manual interpretation methods or traditional deep learning solutions using stacked amplitude data alone.
- **Ability to test multiple geological hypotheses** – Labels are the governing features that drive network training in InteractivAI. An infinite number of networks can be linked to the same label set, enabling testing of different combinations of input data volumes, network types, and loss reduction methods so the best combination of parameters for resolving feature geometry can be identified quickly. This same mechanism also provides capabilities that allow different angle-stack representations of the same dataset to be used to produce P10, 50, 90 volumetric boundary estimations, or different AVO attributes to be used for identifying amplitude-driven prospectivity in regions with known hydrocarbon amplitude signatures.
- **Reduced repetitive interpretation effort** – The rapid convergence of networks trained using the cooperative human-network feedback loop eliminates the most time-consuming processes of manual interpretation, such as correcting unintentional cycle-skipping in horizon interpretations or reassigning incorrectly assigned fault sticks. Removing these hurdles shortens the interpretation cycle time by up to 80%, depending on the interpretation goals, target feature, and data quality.
The remainder of these adoption modules will build on these concepts, beginning with creating your first InteractivAI session and progressing toward advanced multi-volume and transfer learning workflows.
#### What is InteractivAI NOT designed to do?
- InteractivAI is a dedicated augmentation-led AI application for seismic interpretation; as such, it does not contain manual interpretation tools, attribute calculators, or seismic modeling processes such as RGT modeling or spectral decomposition. It’s a flexible, adaptable deep learning tool specifically for structural interpretation and multi-attribute analysis using neural networks.
- InteractivAI is intended to be a powerful addition to your seismic interpretation toolbox, meant to replace automation-led black box AI utilities with an augmentation-led, human-in-the-loop AI system. It is not designed nor intended to be a wholesale replacement for general seismic interpretation platforms.
#### How do I generate VDS files for InteractivAI?
To facilitate creation of VDS volumes, InteractivAI is bundled with **VDS Converter™**, a bi-directional conversion tool that enables two-way conversion of SEGY and ZGY files to VDS format. The VDS connector uses a separate installer package than InteractivAI.
If you use the **Petrel Connector** to fetch base seismic data for InteractivAI sessions from a Petrel project, the use of the VDS converter is not required. Data conversion to VDS (or transcoding, if FAST is being used) is handled in the backend by the connector without the need for any user intervention.
***[Download the PDF version of this module](https://go.bluware.com/hubfs/Knowledge%20base/Adoption%20Series/PDFs/am0_what-is-interactivAI.pdf?hsLang=en)***
- [InteractivAI](https://go.bluware.com/interactivai?hsLang=en#main-content)
- [Getting Started](https://go.bluware.com/interactivai?hsLang=en#getting-started)
- [Installation and Set up](https://go.bluware.com/interactivai?hsLang=en#installation-and-set-up)
[](https://bluware.com)
[mailto:support@bluware.com](mailto:support@bluware.com)
Copyright © 2026, Bluware
```json
{
"@context" : "https://schema.org",
"@type" : "VideoObject",
"caption" : {
"@type" : "MediaObject",
"contentUrl" : "https://go.bluware.com/media-transcripts/411014412004/en.vtt",
"inLanguage" : "en",
"name" : "en Captions"
},
"contentUrl" : "https://6868771.fs1.hubspotusercontent-na2.net/hubfs/6868771/video_assets/411014412004/inherited/web_optimized.mp4",
"dateModified" : "2026-10-15T22:03:15.910Z",
"duration" : "PT8M",
"height" : 2160,
"name" : "am0_what-is-interactivAI_conditioned",
"thumbnailUrl" : "https://go.bluware.com/hubfs/am0_what-is-interactivAI_conditioned.mp4/medium.jpg",
"uploadDate" : "2026-10-15T22:02:51.173Z",
"width" : 3840
}
```