Oracle’s Artificial Intelligence Offerings: An Overview
As with many software vendors, Oracle is keen to promote and differentiate its AI offerings. There are a multitude of these from Oracle, so my post offers a review of the AI categories, services and machine learning options. Inevitably, with the promotion and adoption of various AI/machine learning services, the ‘marketplace’ for these will become more crowded.
As ever, the term ‘AI’ covers a number of different technologies; the specifics of which we won’t be delving into here!
Oracle’s offerings are categorised as follows:
· AI Services — a very broad term to group Oracle’s ‘pre-packaged’ / pre-trained services. These can then be customised as necessary.
· Machine Learning (ML) Services — focused on data scientists to build and deploy ML models which can perform specific tasks accurately using patterns.
· Generative AI — Creates content — along the lines of ChatGPT (text to text) or dall-e (text to image). This is a partnership with Cohere, using OCI for deployment and is currently in beta — not released at this time so I will blog about this when released.
· AI Infrastructure — Superclusters running large numbers of GPU in OCI for training generative AI; if you already have models and don’t need Oracle’s ML services, Oracle can still provide the infrastructure — based on GPU per hour.
Another way to look at the first two services is as follows. Both offer machine learning; AI Services are pre-defined and pre-built by Oracle and offer some customisation. However, if you need a fully customisable and ‘in-house’ built model using large amounts of disparate data sources in your enterprise, you would use Machine Learning services.
In case you don’t understand the relationship between the above services and their differences, ML ‘learns’, creates and improves algorithms for specific tasks. Generative AI uses ML techniques to create content based on patterns and knowledge derived from existing data. AI Services are pre-trained, ‘out of the box’ offerings to meet specific business needs with limited customisation.
In this blog, we will look broadly at Oracle’s offerings and also summarise the pricing model for these services. It’s worth noting that the services are all exclusively cloud-based and offered via Oracle Cloud Infrastructure (OCI); on-premises equivalents do not appear to be available at this time (and probably won’t ever be).
AI Services
This is where most enterprises will initially consider Oracle’s AI offerings and provide the most immediate business benefit.
Oracle’s AI services are split into obvious groupings (based on application):
· Digital Assistant
o The familiar ‘chatbot’ interface your bank or utility company probably has, which leads you through pre-set questions and responses before handing you to a human, if necessary.
o Oracle offers this based on a ‘Requests per Hour’ metric
o There is also a variant (Digital Assistant Platform for Oracle SaaS) which is offered on the Fusion SaaS price list and uses familiar metrics from SaaS products on that price list.
§ You will need to match metrics for Digital Assistant Platform for Oracle SaaS to your ‘pillar’ product; e.g. HCM will use the Employee metric.
· Document Understanding — extract content such as text and tables using either standard format documents or customised templates based on your documents (this will require ‘training’ the model). Can be customised.
· Vision — identify objects in images and return their ‘location’ in the image. Can be customised.
· Speech — speech to text.
· Language — identify language and entity types from documents. Also classes ‘sentiment’ from the content. Translate between languages. Customisable with extended data, e.g. SKUs, and codes specific to your organisation.
· Anomaly Detection — understand and process datasets, accounting for errors, to determine problems with high confidence. Can be customised.
· Forecasting (currently in beta) — forecast various scenarios based on previous data; e.g. utility consumption, server capacity planning, and sales forecasting.
Depending on whether an AI service can be customised (trained to your data/business-specific format/documentation) dictates the metric you can buy these services on.
For example, Language Pretrained Inferencing has the common metric of 1,000 transactions. For a custom model, you need to train it with Custom Training (based on Training hours (equivalent to training on an 8-core machine)). Once trained, you then need to run this using dedicated inferencing (where you have a dedicated endpoint running your custom model); you pay per hour used by an inferencing unit (which processes approx. 500 characters per second).
This ‘pre-trained’ and custom model applies wherever I’ve mentioned customisation in the descriptions above — you have different metrics for each service type; typically transaction-based for pre-trained and training hours for customising the mode.
Machine Learning
A more advanced offering or need to model ‘from basics’ rather than using pre-packaged models, with significant customisation based on your data
According to Oracle, this OCI-hosted service is ‘a fully-managed platform for teams of database scientists to build, train, deploy and manage ML models’.
What this means is a lot of training of models (data science) on very large datasets to customise/train the output for your business use case. The volumes involved require large clusters of compute (such as that in OCI or AWS) power, with very high-performance processing and network connectivity. Oracle offers such processing power as ‘OCI GPU’ (Graphic Processing Unit) which offers high parallelism processing of these large datasets and is offered in the ‘GPU per Hour’ metric (the more GPUs you put to work, the quicker it will be, the more it will cost).
GPUs will provide the computing power for AI model training (where training takes place against large datasets). This creates a model which can then be deployed for use in model inferencing (again, using GPUs).
· Data Science
o End-to-end development environment for building, training and deploying machine learning models. Metrics are based largely on GPU or OCPU per hour or Gb (storage or per hour).
· Virtual Machines for Data Science — pre-packaged VMs available from the Oracle marketplace, pre-loaded with the necessary libraries and tools for building models. These are charged based on the shape of the pre-configured virtual machines — so GPU per hour.
· Data labelling- the creation of labels for data showing the data ‘context’ which can be used to train machine learning and enable accurate ‘predictions’ and usage of the data. Paid for based on Annotated Data Records — i.e. a record which has label(s) applied to it for classification purposes.
· Machine Learning in Oracle Database- allowing natural language-based queries to Oracle databases. Improve collection of data for building models, spot patterns and relationships in the database. This is actually ‘just’ a rebranding of Advanced Analytics which was also known as Oracle Data Mining. There is no cost for this; it is included as part of the Database license fee.
About the author
Paul Bullen has worked for Version 1 since 2009 and works in a team of recognised independent experts who help customers understand their use of Oracle products and services, as well as providing benchmarking and price comparison services.
About Version 1
Version 1 offers Artificial Intelligence innovation and expertise across multiple industries such as finance, pharmaceuticals and utilities with applied expertise in real-world use cases. Combined with our wider enterprise application expertise, cost optimisation and multi-vendor approach, we offer best-in-class solutions for your business needs. Contact us (https://www.version1.com/contact#form) to discuss further.
