Artificial intelligence (AI) technologies open new horizons for organizations from different industries. With the help of AI, you can process huge sets of data at high speed, analyze visual, audio, and text content to extract specific features, and easily solve complex tasks that humans can’t.
With the rising popularity of AI, cloud service providers (CSPs) have started to offer services dedicated to specific tasks: detecting objects in video, recognizing faces of celebrities, turning speech into text. Some of these providers have even taken steps toward offering a more useful solution: an AI Platform as a Service (AI PaaS).
In this article, we explain where the term AI PaaS comes from, what it means, and what AI service vendors offer the richest set of AI services, tools, and solutions. This article will be helpful for developers and business owners who want to introduce new AI capabilities to their products.
Artificial intelligence technologies have gained wide recognition and are being implemented in healthcare, marketing, cybersecurity, and other areas. But since AI is a resource-hungry technology, deploying it may be costly. This is why many organizations use AI as a Service (AIaaS) offerings from cloud vendors to experiment with AI-based technologies.
Cloud vendors provide AIaaS for those who want to benefit from AI technologies without spending too much money, time, and effort on it. At first, CSPs focused on separate AI solutions meant for solving specific tasks:
- Processing and analyzing big data
- Processing specific types of data (image, video, language)
- Building prediction models
- And so on
Today, however, there’s an emerging trend among AI service providers: they’re moving from offering a range of separate services leveraging AI capabilities to providing full-scale AI platforms.
Saas, PaaS, and IaaS Explained
This is where the AI Platform as a Service (AI PaaS) model comes into play. For now, AI PaaS seems to be more of a trend than an actual offering, since the term has been around for only a couple of years and the market is still forming.
AI PaaS was first listed among AI trends that are “at the peak” of popularity in Gartner’s 2018 report, and it remains a trend even a year later. Gartner’s 2018 report underlines the efforts of four CSPs – Amazon (AWS), IBM (Watson), Microsoft (Azure), and Google – to provide comprehensive AI platforms as an alternative to separate AI services.
Following Gartner’s lead, we can define AI PaaS as a set of AI and machine learning (ML) platform services for building, training, and deploying AI-powered functionalities for applications.
In the next section, we compare AI PaaS with the general PaaS concept and take a look at the key elements an AI PaaS architecture may include.
Gartner sees AI PaaS as a set of separate AI services. However, we can also look at the concept of AI PaaS from the perspective of the classic Platform as a Service (PaaS) model.
In a PaaS, a cloud vendor provides an environment for building, deploying, and maintaining applications. Such an environment usually includes two main groups of components required for application development:
- Hardware infrastructure (computing power, data storage, networking infrastructure, virtual machines)
- Software solutions, tools, and services
The key hurdle to generalizing a similar scheme for the AI PaaS architecture is that there’s no general model for AI PaaS yet. The market is still forming, and different vendors offer completely different sets of services under the same umbrella term.
Yet there are a number of elements that are common to the majority of today’s AI PaaS and AI service platforms:
- Data storage
- Pre-trained ML models
- AI APIs
Let’s take a closer look at each of these components.
AI is all about processing enormous amounts of data, which, in turn, requires extensive computing power. This is why, similar to the standard PaaS model, many AI service providers offer infrastructure resources, computing, and virtualization capabilities.
All the big data required for training machine learning models and improving AI solutions needs to be stored somewhere. This is why data storage resources are a common element of AIaaS and AI PaaS products. In some cases, though, cloud storage is offered not as part of the platform but as an additional service.
Pre-trained machine learning models
Machine learning algorithms and models play the main part in implementing AI-based solutions: they process and analyze data, solve specific tasks, and provide the end result you expect. However, building and training a machine learning or deep learning model from scratch requires resources and expertise.
This is why many cloud vendors offer pre-trained models and algorithms that can solve specific tasks: extract features, make predictions, run complex calculations, and so on. Some pre-trained AI services can only perform a limited set of operations, while others can be customized to the needs of a particular project.
APIs make it even easier to implement AI functionalities in your application. Fortunately, some vendors provide ready-to-use APIs for various AI functionalities as part of their platforms or as a separate service.
The most common APIs offered by AI service providers include APIs for computer vision, natural language processing, and text to speech conversion.
Now, let’s see the benefits and drawbacks of third-party AI services.
CSPs make AI capabilities available for developers, business owners, and researchers, which is great. Let’s take a look at the most important pros and cons of using a third-party AI service in your project.
Benefits of using third-party AI services
There are three main benefits of implementing a third-party AI service instead of building your own AI solution from scratch:
- Reduced costs and time
- No need for a high level of expertise
- High scalability
Reduced costs and time. Implementing ready-to-use AI features and solutions can save you considerable time and money. With an AI service, you don’t need to hire experienced professionals for developing, implementing, and maintaining AI-based functionality for your application or business process. Most of these tasks can either be solved or simplified by the cloud vendor. Plus, the traditional pay-as-you-go model used by most CSPs allows you to effectively control your spending.
No need for a high level of expertise. Usually, AI service vendors take care of the most technically challenging part of AI feature development. Some of them also make it easier for their customers to deploy and maintain ready AI-based functionalities by offering additional technical support and consulting services.
High scalability. You can start small and scale your AI-based project as you need, without worrying about computational power. Scalability can be crucial when processing huge sets of data, deploying your solution across multiple platforms, and so on.
Now, let’s see the key drawbacks of using third-party AI services in your projects.
Drawbacks of using third-party AI services
The key disadvantages of using third-party AI tools and services are all connected to the way these services are delivered:
- Limited capabilities
- Data compliance concerns
Let’s look closer at each of these concerns.
Limited capabilities. One of the key issues regarding the use of third-party AI services is that you can’t control most processes. For instance, while many AI service providers offer pretrained models, the range of customization options is usually limited. So if your solution requires non-standard approaches, you may need to look for an AI vendor who allows full or at least extensive customization of pretrained models.
Data compliance concerns. In order to get highly accurate results, ML models require lots of data. However, you might have concerns regarding the security of the data processed by your AI solution. For instance, you need to be sure that all data is stored and transferred safely, especially if you store it in a public cloud.
Next, let’s look at the four providers who are closest to offering a full-scale AI PaaS so far.
According to Gartner, Amazon, IBM, Microsoft, and Google are the leading CSPs offering AI services. In general, these CSPs focus on three areas:
- Machine learning
- Computer vision
- Language processing
It’s noteworthy that while most providers offer similar AI services, particular services may differ in terms of capabilities, technical details, and limitations.
Below, we take a brief look at the key AI services offered by each of these four cloud vendors.
AWS AI services
Amazon Web Services (AWS) covers all three areas we mentioned. There are dedicated tools and features for working with ML models, analyzing natural language, and solving computer vision tasks.
AWS offers a rich set of AI services spread across nine categories, from text comprehension to speech recognition to personalized recommendations for end users.
Aside from AI services, there’s an ML service called SageMaker. It enables you to build, train, and deploy ML models easily and efficiently. The service allows you to use pretrained models as well as build and train complex models with reinforcement learning.
For those who want to speed up the process of building and training an ML model, AWS offers over 200 pretrained models and ML algorithms on the AWS Marketplace.
Google AI Platform
Google AI Platform is a code-based data science development environment whose focus is on machine learning. AI Platform allows teams to collaborate on ML projects from the dashboard within the Cloud Console.
Google AI Platform can facilitate five out of seven stages of an ML project:
- Training, evaluating, and tuning models
- Deploying trained models
- Getting predictions from models
- Monitoring ongoing predictions
- Managing ML models and versions
The only two stages a developer needs to take care of are the first two: preparing data and coding the model.
Google AI Platform provides developers with infrastructure and storage for preparing data, building and running ML models, managing models, and sharing AI data with others.
Here are some interesting ML services provided by Google AI Platform:
- Data Labeling Service – Provides you with a team of human labelers for accurately labeling the data you’ll later use for training a custom ML model
- Deep Learning VM Image – Provides a fully configured environment that supports all popular AI frameworks, including TensorFlow and PyTorch
- Kubeflow – A dedicated ML toolkit for Kubernetes
- AI Hub – A hosted repository where ML teams can easily discover and share their AI content
IBM Watson is a suite of both general and industry-specific AI services, applications, and tools. Similar to AWS, Watson covers all three areas of interest outlined by Gartner.
IBM Watson offers a large selection of tools and services for preparing data and building and training ML models. For each service, there are a range of deployment options. For instance, Watson Studio can be deployed on a public or private cloud as well as on a desktop. Watson Machine Learning, on the other hand, can be integrated with Watson Studio and deployed in a public cloud, private cloud, or multi-tenant distributed environment.
Watson also provides a set of ready-to-use APIs and allows you to generate your own APIs for integrating AI functionalities into your applications.
What differentiates IBM Watson from the other offerings in this article is its set of empathy tools for analyzing and understanding human emotions in text:
- Personality Insights predicts personality characteristics, needs, and values based on text written by a specific person
- Tone Analyzer understands human emotions in text
There’s also a tool called Watson OpenScale that you can use for analyzing the efficiency of AI technologies in your applications and automating the AI lifecycle.
Microsoft Azure AI platform
Just like similar services provided by Amazon and IBM, the Azure AI platform can be used for solving tasks related to ML, computer vision, and language processing.
At the same time, Azure splits its AI services into three groups:
- Machine learning – A Python-based service that provides ML capabilities for building, training, deploying, and automating different types of ML models
- Knowledge mining – A set of AI services for extracting insights from your content and turning forms into usable data
- AI apps and agents – A set of cognitive services and bot services for solving different kinds of AI tasks
Cognitive Services cover four groups of tasks: computer vision, speech processing, language processing, and decision assistance. Bot Service provides you with templates and an environment for easily and quickly creating chatbots.
As you can see, while these AI service vendors offer similar sets of AI capabilities, there are some nuances and limitations. Therefore, your choice of an AI PaaS vendor should depend solely on the needs of your products.
Here are some final thoughts on what to pay special attention to:
- Data quality – No matter what AI service provider you work with, the efficiency of AI features fully depends on the quality of the data processed. Put enough effort into the data preparation stage and use reputable databases with quality content.
- Compatibility – Pay special attention to the set of tools, services, frameworks, and programming languages supported by a particular AI PaaS. The more matching frameworks and tools, the easier it will be for your development team to work with this platform.
- APIs – Many AI service providers offer APIs for integrating AI capabilities in your application. Using APIs, you can introduce new AI functionalities into your application faster and with less effort.
AI PaaS is an emerging trend on the AI market and an umbrella term for a rich set of AI services provided by cloud vendors. Usually, AI PaaS covers three AI-related areas: machine learning, computer vision, and natural language processing.
There are four leading AI service providers who may be considered AI PaaS vendors: Amazon, IBM, Microsoft, and Google. While the selection of AI services offered by these vendors is quite similar, each has unique features and certain limitations.
At Apriorit, we have vast experience introducing AI-powered functionalities to applications and solving different AI tasks, including image classification, face recognition, and even object detection in video streams. Contact us to enhance your business with AI capabilities and bring your ambitious project to life.