Cloud providers offer similar looking AI platforms. This blog post presents some ideas to help you make an informed choice for AI platform.
So you are about to build an application using AI. You have to use artificial intelligence to solve a business problem and now you need a platform. You may not have a large data science team at hand. How to proceed now? APIs from AI platforms might be the answer to your situation.
Consider these items when choosing the AI tools:
- Cloud platform choice
- Data ownership and privacy
- Language choices
- API availability
- Costs
What is an AI Platform?
AI Platform is a loose term which can mean different things. Major cloud platforms provide a set of APIs for artificial intelligence. These APIs often include image recognition, speech and language recognition.
The AI APIs provide tools to solve business problems for organizations without large data science teams. Today the scale and the width of the AI APIs available is excessive. When evaluating APIs for your AI needs, APIs from major cloud vendors are a good starting point.
All the well-known cloud players have their own AI platform:
- Amazon AWS by AWS
- Google AI by Google
- Azure AI by Microsoft
- IBM Watson by IBM
Moreover, there are surprising contenders, such as Apple’s SiriKit and SAP Leonardo platform. Siri offers a comprehensive language support also for smaller language groups like Finnish and Swedish. Some platforms are dedicated for data science, for example Databricks with its Apache Spark offering.
AI platform elements
At first sight all the AI platforms look very similar. Each platform includes high-level APIs which hide the actual machine learning algorithms.
Cloud providers have a specific set of APIs for AI. These are targeted to developers without advanced machine learning expertise. Yet understanding machine learning basics is recommended to make the best choices and to use the large set of AI APIs.
APIs for artificial intelligence provide high-level abstractions of complex machine learning algorithms:
- Vision – APIs to recognize objects from images and videos
- Speech – APIs for speech recognition
- Language – APis to identify entities from free format text and translate text.
Machine learning tooling is an important element when working with data science. Each platform provides a notebook environment for experimentation, often based on Jupyter Notebook. Notebook environments are often integrated with an easy access to cloud provider’s machine learning libraries and infrastructure.
Cloud platform choice
Your organization’s cloud strategy may influence the AI platform choices. You need to figure out whether to pick all the components from your organization’s preferred cloud platform. The ”best-of-breed” approach can also be applicable when choosing AI APIs.
You can mix and match the AI APIs from several cloud platforms. Cloud providers have different focuses on their AI APIs. The hybrid approach is often the way to go to achieve the best combination.
Data ownership and privacy
Data ownership and privacy are major factors when evaluating cloud platforms. GDPR privacy ruling in EU since 2018 has put more emphasis on validating the privacy requirements.
Machine learning uses training and test data sets. Training data from EU citizens must locate within EU borders. You need to read the small print from API licence agreement to identify the storage location.
Check also the ownership of the training data. You do not want your customer’s photos or text content to appear in other applications. The data breaches from Facebook and Cambridge Analytica are a constant reminder of data ownership.
Availability of pretrained models is a major advantage. Here you must also check the cloud provider’s licence agreement to verify the data compliance.
All languages are not equal in AI
If your business case requires understanding of speech or natural language, you do have to check the language support. Speech recognition and text analysis using NLP (Natural Language Processing) require a profound command of the language.
If your language is English, you are good to go with speech and language APIs. Large cloud providers provide outstanding support for large language groups, including English, Spanish. The rest of the world is not so well covered when it comes to natural language support.
Life cycle costs
Costs are important criteria when considering the application’s entire life cycle. You can be charged on various bases, including:
- subscription-based charging
- transaction-based charging
- CPU-based charging or
- some other charging basis.
The costs might vary depending on the region where the service is running. One needs to consider also the costs for development and test environments. Cost calculator can be an extra advantage for the cloud platform. This helps to avoid unnecessary surprises.
The key item is to calculate the life-cycle costs, for example costs over 5 years. Do not forget costs related to data migration and data transfers.
How to move forward
Cloud providers announce frequently new artificial intelligence APIs. Initially they are often for the preview usage only; and the official general availability will arrive later. It’s always good to verify that the API is available in the region where the services are running.
These are the key factors affecting your selection of APIs for AI:
- Decide whether to use hybrid or preferred cloud
- Who owns the data and where is it stored
- Remember the language (limitations)
- Check the API availability in your region
- Estimate the life-cycle cost.
The world of artificial intelligence and APIs is rapidly evolving, so keep looking for the great new APIs.
This blog post was originally published at apiscene.
