Decomposing AI: The Many Flavours of Artificial Intelligence in a Composable World

Generative AI tools such as ChatGPT and Midjourney have sparked significant interest in the field of AI. There is more to AI than just content generation however, with countless tools to automate, predict, and explore, all of which require deep understanding and the right integration approach to create business value.?

Artificial Intelligence (AI) has as many definitions as there are authors writing about it, statisticians exploring it, and start-ups trying to sell it.

Within the field of statistics there are many techniques for analysing data, from modelling trend lines and predicting outcomes to identifying patterns and groups in data. Whilst some of these techniques could be applied manually with considerable time and effort, the power of computers has made using these statistical methods as simple as writing a few lines of code. Hence why what was once just “statistics” has now been dubbed Machine Learning (ML).

Some statistical methods could only ever be used with computers, and powerful computers at that. One such example is the Neural Network, a series of connected nodes not unlike the neurons of a brain. Various calculations are carried out between the nodes to train the network to accurately predict an output based on a set of inputs. The complex structure of the network can enable it to produce very accurate predictions, but at the cost of interoperability and processing time.

For many, myself included, AI is not the application of any ML technique, but the specific use of a Neural Network. Others, however, have played fast and lose in their definition, seeking to piggyback on the popularity of the term and the investment opportunities it offered. For a long time, anything statistical has been interchangeably referred to as AI or ML or AI/ML, downplaying the power true AI can bring to bear.

Now, with high-profile Generative AI (GenAI) examples from the likes of ChatGPT and Midjourney, the world is starting to see real AI in action and it’s raising questions on what else AI can deliver and how to integrate with it.

Generative AI

Generative AI tools take in prompts, usually in the form of text, to then use neural networks to generate text, audio, images and even video responses, creating this content in minutes if not faster. The tools have been trained on a wealth of existing material, from web pages to research articles to works of art and beyond.

There are countless vendors now working on GenAI tools, with a new one likely established by the time you finish reading this article. The next step for the field is where and how these tools are integrated into products and business ecosystems.

·??????Translation and dubbing. Youtube is testing integrations with Aloud, an AI-powered dubbing service, to automatically create multi-language audio tracks to videos, significantly reducing the time and cost for creators to reach a wider international audience.

·??????Search Engine Optimisation (SEO). Contentful is an example of a MACH vendor, a composable solution offering where a business can assemble the capabilities it needs in areas such as content and commerce from multiple vendors, instead of relying on a monolithic solution. Contentful is a Content Management System (CMS) and, as a MACH vendor, is API first, which increases the ease of integration with other tools such as GenAI. On the Contentful marketplace there are already multiple examples of AI content generators, automatically producing SEO keywords, image tags, image alt text and more.

AI Automation

Content generation is just one of the many use cases for AI. Through automating previously manual tasks, AI can create process efficiencies and support faster time-to-market.

·??????Data cleansing. Akeneo, a MACH Product Information Management (PIM) solution, recently announced new product capabilities including AI-driven data cleansing to support faster time-to-market by automating the cleansing, deduplication, and categorisation of product data.

·??????Software testing. Multiple vendors are now claiming to incorporate AI in automated software testing, detecting elements on screen, evaluating website fields and forms, applying test flows, and automatically identifying issues.

Predictive AI

At its core AI is a tool for prediction, looking at a set of input data to determine what’s the most likely outcome. Predicting the price a commodity may have the next day, or week, or month, so you can decide what to buy or sell today. Predicting how likely it is a person will buy the different items in a range of products, so you can then decide to show those with the highest likelihoods. Such Predictive AI algorithms are already in widespread use today.

·??????Film assessments. Warner Bros. has signed a deal to use Cinelytic’s AI-drive project management system to guide decision-making when greenlighting film and TV projects. The system uses comprehensive data and predictive analytics to assess the value of stars and predict how much a film is expected to make. Tobias Queisser, the founder of Cinelytics claims “the system can calculate in seconds what used to take days to assess by a human when it comes to general film package evaluation or a star’s worth”.

·??????Product recommendations. Commercetools, a MACH Commerce platform, has a wide range of AI integrations available in its marketplace, including tools such as Attraqt Fredhopper Discovery Platform, which uses AI to power search, merchandising, and personalization. Segment-based recommendations and AI driven “shop the look” features attempt to predict which other products a customer is most likely to buy based on historic purchases for that segment.

Even as I type this on Word, Microsoft is starting to predict the next word I’m likely to include, letting me add it in with the simple tap of the tab key.

Exploratory AI

Finally, we come to exploring data with AI. Rather than trying to predict an answer, or generate new content, here we are looking to gain insights about a set of data by detecting previously unseen patterns or groupings. This requires a different approach to the previous examples, where “correct” answers could be used to train an AI to calculate outputs for new data. With data exploration, there is no true, correct answer, only the different potential patterns we can surface.

·??????Customer segmentation. In many cases of data exploration, the approach taken is typically ML based, rather than AI with a neural network. For example, many businesses will use Customer Data Platforms (CDPs) with ML techniques to automatically detect customer segments based on their data. There are exceptions however, where different types of neural networks, such as Self Organizing Maps (SOM), are applied. Researchers are continuing to explore how to use SOMs to identify customer segments.

Composing suites of AI tools

There are hundreds, if not thousands, of AI tools available now, with more continually being developed. Each has the potential to add value to a business by itself, but by combining them the benefit delivered could be exponential. Imagine having an AI automatically discover new high-value customer segments in your database, then a different AI dynamically mapping new customers to these segments, with AI automation managing personalization campaigns where GenAI tools are creating new, targeted content to engage these new customer segments. All from a single press of a button.

This utopia is closer than you might think. To take advantage of it companies need to have the right foundations in place. A composable technical architecture gives businesses the modularity and support needed to seamlessly integrate not just a single AI tool, but multiple tools, orchestrated together to identify, plan, create, and publish compelling customer experiences.