AEM and Generative AI Unpacked Through a Critical Lens

As 2023 began, there appeared to be an unprecedented surge in enthusiasm for AI, specifically large language models (LMMs) and other types of generative AI technologies. Since then, LLMs have penetrated individual life and a range of industries based on web and visual communication, including design, marketing, journalism, writing, consulting, customer service, and a whole lot more.

Adobe has wasted no time in employing AI as an invaluable element of its offerings while showcasing AI’s transformative power at the Adobe Summit 2023. It seems as though GenAI is the biggest ‘revolution’ in the digital space since the advent of social networks. However, taking things with a healthy pinch of salt can actually help you balance excitement with actual business outcomes given the intensity of the current mixed emotions and insufficient knowledge about Generative AI (GenAI). Let’s take a closer look at AEM and Generative AI to help us gain a more poised stance on all this.

AEM and Generative AI: Two Sides Of The Coin

Surprisingly, the massive wave of ‘ChatGPTization’ across the IT world is not yet losing momentum. However, the ‘wow’ effect of innovation shouldn’t inspire businesses to throw all their resources at AI; so it’s time to take a sober, critical look.

Security Measures in Generative AI

Security is essential to implementing AI solutions. Generative AI is demanding partially due to the necessity to manage and protect sensitive data such as personal customer details, financial records, and other confidential information. By and large, the biggest GenAI concern is when you send sensitive data to free third-party providers or services, to train the networks, this puts the information in a potentially vulnerable position – breaching confidentiality can happen unadvertedly in the process. This has become a real problem for free services like ChatGPT and claude.ai. Yet, the OpenAI team is working toward a secure enterprise version of ChatGPT.

Adopting a multi-layered security approach with precise access control protocols is crucial to safe operations within the AI system. Access should be given to authorized staff, and permissions should be carefully distributed as to who can input data into the system. Additionally, companies can encrypt data and install firewalls to protect against cyber attacks or data leaks and theft.

Continually updating systems with the latest security patches and version upgrades are essential in order to maintain security and compliance. Implementing continuous monitoring tools that detect threats or breaches even before they occur can raise the level of protection significantly, safeguarding both stakeholders and end users of AI programs.

Far-Fetched ROI Value

You’d probably want to see immediate results upon AI implementation within your organization. ROI in AI is usually a long-term expectation, requiring time to eventually pay off. However, to see if your game is worth the candle, you can opt for an AI PoC (proof of concept) in order to project possible returns on your investment and increased value of your company as a whole.

AI Initiative Support

When companies choose to manage their own models, they typically encounter challenges at the infrastructure level. Generative AI models require significant computational and memory resources to run and demand regular maintenance and updates. In addition, these models may create new cybersecurity risks. It is important that you continually attempt to predict future patterns and make projections for the scaling capacity and storage space you may need down the road, even before building and rolling out your AI environment.

Time and Dataset Limitations

Large language models by themselves are incapable of generating fundamentally new meanings. Their capacity is limited to the content of the dataset that was fed to them and the timeframe of data relevance.

There is, therefore, a high risk of a network ‘hallucinating’ and presenting randomly generated information as ‘facts’. A language model can convincingly combine different facts that it knows; and since it is equally successful in embedding both ‘real facts’ and completely fictional ones in one output, we sometimes have to double-check the output with the subject matter expert, editor, or another reliable source. These issues can present hurdles that can be challenging, especially when dealing with sensitive information where accuracy is of uppermost importance.

Exclusive Data Ownership

Since generative AI is currently available under the Software as a Service (SaaS) model, and, as we touched upon earlier, the data necessary for generating content may be owned by, or provided to a third party. Many businesses and government agencies have privacy concerns as to how this sensitive information is being handled. Therefore, the best way to use AI boils down to creating AI tools that allow you to own the data you use to train your models, as well as the environment with access controls.

Difficulty Formulating Prompts

Integration of software products and systems with ChatGPT or similar models is becoming commonplace. The LLM integration may not fundamentally change the quality and functionality of products if done frivolously. As it turns out, the quality and thoroughness of compiling a prompt — a request to a neural network — is decisive to the success of working with complex LLMs. And that’s where many users struggle.

Regional Laws and Regulations

The use of large language models raises both ethical objections and sometimes open protests, having governments needing to step in to solve disputes – take, for example, Hollywood actors protesting against the potential rise of AI ‘actors’ taking the limelight.

In such cases state regulators must come into play and are ready to impose restrictions on the circulation of content generated by a neural network. For example, the Chinese government has recently demanded AI content be labeled in a specific way so that no one confuses it with the work of living people. It appears that similar measures will soon be enforced worldwide.

Ethical Behavior Shortcomings

A large language model, when used as the foundation of a chatbot, will act as a friendly and helpful interlocutor. As of writing this article, there are no guarantees, however, that GenAI will remain highly ethical at all times, no matter what it is being asked. Reliable mechanisms for ethical control are still being explored.

There is no question that Generative AI is here to stay, regardless of the challenges. Ignoring AI completely without exploring what it can do for you as a business owner would be to take a step back in the game. It’s time to explore where Generative AI can actually make a difference for your organization, and such a powerful platform as AEM.

When is the Right Time to Combine AEM and Generative AI in Your Project?

Generative AI certainly got the world talking and arguably boiled the ocean somewhat. Despite the hype, you need to focus and prioritize the benefits and value for your company over simply being mainstream or tapping into Generative AI just because your competitor is using it. Here are a few distinguished signs that can help you while considering integrating AEM and Generative AI into your business.

• You can pinpoint repetitive tasks with clear execution results
• You have large amounts of content, such as imagery, text, and videos
• You need deep machine learning algorithms to build predictive AI, AI analytics system, or AI recommendation models
• You want to give your users access to your products or services 24/7 through an AI-powered chatbot
• It takes you long hours to process data manually

Being on the safe side while still giving GenAI a shot is a balance you’d want to strike, especially when it comes to a multifaceted environment in AEM.

Generative AI Interwoven Into AEM Environments

Adobe Experience Manager is a complex content management system that helps you accomplish a number of goals simultaneously: building pages and distributing vast amounts of content to relevant audiences while managing marketing campaigns and analytics. Since AEM is tailored for large companies, there is much room for AEM and Generative AI to streamline workflows for publishers, developers, and marketers.

Digital Assets

You can integrate GenAI into creating and editing images, text, videos, or 3D models. Generative AI allows you to go far and beyond with images, allowing you to change vectors, edit images, and add many elements in just a few clicks.

Content creators can produce content with prompts and have it translated into a target language in a few seconds. You can also create and edit videos at will.

Additionally, Generative AI can feed on your assets to learn and eventually produce whatever it is you need: text, images, videos, or product models in 2D or 3D.

Finally, marketers produce content and distribute it across platforms and campaigns, applying Generative AI tools such as rewriting content for a certain audience with a peculiar tone of voice.

Creating Code

Generative AI can assist developers in building code or at least point them to tools such as widgets, libraries, or frameworks that can be used for a certain task or functionality. Such AI-powered assistants allow developers to perform twice as fast as working on a solution alone.

SEO Support

You can use GenAI to generate SEO metadata. For example, you feed the AI assistant a URL and ask it to derive meta keywords after reading the contents of the link. Or you can ask your AI assistant to produce a meta description for your AEM page.

AI-Enabled Algorithms

Knowing that AEM is a complex high-level system that can handle data and interactions both from administrators and customers, it must be powered by powerful software. In this case, AI is the technology that can elevate experiences for any type of user since AI algorithms help a system learn, analyze data, and make decisions based on past collective knowledge. These algorithms possess an intelligence that can recognize patterns, understand human language, and assist in problem-solving.

AI Chatbots

Your AEM site can be ‘open’ 24/7 for your customers provided you have an AI-powered chatbot trained to sell, resolve issues, or provide customer support. When there’s a necessity and the issue at hand is out of the bot’s scope, it can be escalated to a human assistant. Nonetheless, the more repetitive a certain interaction on your site is, the more likely it can be reimagined through Generative AI. And this means that you can direct your human resources to other tasks.

By the way, creating the content for a separate web page does not end here. Large language models have a powerful potential for targeting text to different demographics and translating text into different languages. This makes it possible to quickly create multiple variations of pages from one blueprint.

Once you decide to go with AEM and Generative AI, the key objective is to match your site with the right set of AI tools. At Exadel, we carefully evaluate available AI tools, APIs, and libraries that are compatible with your AEM configuration and can amplify the effort for multiple teams. Besides the Adobe GenAI offerings, we can also help you evaluate the likelihood and risks of using third-party resources specific to your unique needs. And that’s where an AI PoC (Proof of Concept) comes in handy.

How Does AI PoC Help You?

The essential idea of Exadel’s three-day AI PoC is to estimate the probability of matching AEM and Generative AI for just a fraction of the usual cost. With AI PoC, you test-drive the solution first and then invest in it and go live with your initiative. In the PoC, you’ll get the chance to cover all the common use cases, challenges, and solutions in less than a week and decide whether to give AEM and Generative AI a future:

In a nutshell, the three-day AI PoC lets you dip your toes in the water instead of jumping right in with undesirable consequences and unnecessary expenditures once you choose to combine AEM and Generative AI.

First-Hand Experience with LLMs

AEM is one of the world’s most advanced and sought-after platforms for web development and business over the internet. It provides multiple integrations with third-party services, including Generative AI.

In the spring of 2023, the Exadel Digital Experience Team created a prototype system based on AI tools to assist in content creation on an AEM platform. Consequently, Exadel Authoring Assistant for AEM was the end product with several LLMs at its core. For instance, the platform administrator can connect the OpenAI/ ChatGPT service or another open-source service such as Writesonic. Additionally, they can use an LLM deployed on their own hardware, such as LLAMA2 or StabilityAI.

The Exadel Authoring Assistant for AEM boasts multiple features for streamlining, authoring and publishing efforts by:

• Completing page layouts, individual text blocks, or images within a page
• Improving text readability, correcting errors, developing or shortening text in accordance with the conditions of the web page layout template
• Reformulating text for the language profile of a specific audience such as youth, business people, senior people, etc.
• Translating text into a foreign language when necessary
• Creating graphic illustrations for the finished text
• Automatically selecting tags (keywords) for text and images, creating summaries and SEO metadata for a page
• Enhancing site visitors’ experience with semantic search (going beyond thecontent that search word context and “understanding” the visitor’s request, providing one concrete answer supplemented by links to existing content)
• Creating an AI site bot assistant for users

Will Humans Still Be Building AEM Websites?

In short: yes, humans will still be creating and authoring. Such complex content platforms as AEM will continue to be major vehicles for website development. A website is a medium in the human world delivering a message from one human being to another, be it breaking news, a sales offer, or an opinion. Each AEM webpage, by nature, is a message originally born from a human. Imagine if you took the human out of the equation, you’d soon get an information hub full of ‘noise’ sent from no person to no person. The human will always be there.

You may have seen a complex website architecture and content structure that stands behind an AEM site. While AI is evidently good at fulfilling tasks, who is there to formulate them but a human? At the end of the day, a human decides how the website looks and feels. On top of that, a human makes choices as to what should and what should not be done; which features to implement and which to walk away from. Every good website manifests values and beliefs. It is indisputable that brand values, ethical standards, and emotions are essential to the human narrative and communication. Human nature is something that AI most definitely lacks. Intelligence deprived of human nature is cold, unfeeling, and unable to discern between ‘gray areas’ and abide by morale.

Additionally, the structure, the content, and the emotional vibe of a website must fit with the culture of the audience. A respected website without developers, trained editors, and publishers can go sideways if only governed by AI, leading to potentially inappropriate messages, inaccurate facts, or full-on hallucinations. It’s difficult to imagine an AEM website lifecycle without well-grounded and regularly reviewed business objectives and promotional campaigns, decided upon and led by human inspiration.

Being in Control Over Change and Innovation

Handing your entire business to AI, however, is futuristic and highly utopical, because the human touch and dedicated teams are what holds any AEM site together. At Exadel, expertise and sound, considered reasoning are inextricably linked. Instead of putting the cart before the horse, we help you map your steps, test each solution, and draw conclusions as to whether Generative AI will perform to the best of its ability in your next AEM and Generative AI project endeavor. Get on the call with us or leave your contact details. We’d be delighted to help you incorporate GenAI in AEM effectively and securely!