Why enterprises need an AI Architecture Practice

Current State :

There is a lot of push from business executives to inject or introduce AI, GenAI and other digital & emerging technologies into business applications, that can then be sold to clients. Given the perceived urgency, every single business leader, and their respective teams are solving this problem in a way that makes sense to them, which is a very natural way to react.

There is also the all-pervasive question of budget; today, business executives are open to invest in AI and GenAI initiatives, so technology leaders are keen on leveraging this opportunity to broaden their portfolio, hire more folks, build new products, garner credentials, and so on.

The first one to go to market with an AI enabled product might be declared the winner here, so speed and agility are being perceived as more important than stability, reliability, and robustness.

But what is happening on the ground :

Small POC(s), demos to clients are being built quickly, and the notion of ‘perceived value’ of an AI/GenAI infused asset is being ‘sold’ to the client.

But, once the deal is closed, and it comes to the actual question of building scalable, enterprise-grade AI applications, reusable AI assets, harnessing well-designed AI architectural patterns, and other such topics, many AI initiatives are stalling due to a lack of preparation, and absence of skilled architects who can help in designing well-rounded AI applications, and so on.

As a result of this, the overall value proposition of the AI revolution is not being realized, resulting in trust issues on the technology stack, IT capabilities and even the overall AI strategy of the organization.

Gartner estimates that “50% of IT leaders will struggle to move their AI projects past the proof of concept ( POC ) stage into production.”

What can we do :

Just like any other important stream like Data, Cloud, LowCode, and so on, an enterprise can invest in a dedicated AI Architecture practice, with specialized AI architects, which could significantly improve the chances of being successful in scaling the enterprise AI strategy across diverse business teams by helping define the architectural patterns, creating specific AI-enabled workflows, identify AI focused tooling, scaling AI operations, and so on.

Who are AI Architects :

‘AI architects are the curators and owners of the AI architecture strategy. They are the glue between data scientists, data engineers, developers, operations (DevOps, DataOps, MLOps) and business unit leaders to govern and scale the AI initiatives” - Arun Chandrasekaran, Distinguished VP Analyst at Gartner.

What attributes should an AI Architect possess :

How are AI architects different from normal solution architects :

There is significant overlap between the work that any solution architect does, and what an AI architect is typically expected to do. Most likely, an AI architect has spent a large percentage of their work experience as solution/infra-architects, being involved in designing and implementing technical solutions, solution architecture diagrams, and so on.

However, an AI architect will typically be more focused on particular fields of architecture, as in :

• AI Architects will have a more well-defined focus on scope like understanding AI algorithms, ML techniques, designing AI pipelines on top of data assets, whereas, solution architects will have a broader scope across different IT domains like ERP, Cloud Infra, Enterprise Service bus, etc.
• AI Architects will have more in-depth knowledge on AI algorithms, ML topics, NLP, deep learning and so on, with idea about designing, deploying, maintaining, updating, and executing AI algorithms, ML models. But, same level of expertise cannot be expected from a solution architect, who might not have an in-depth knowledge of all AI topics.
• Typically, an AI architect will have some experience in creating, managing, and maintaining AI infrastructure, that is a pre-requisite for AI systems, including selecting hardware/GPU(s), configuring distributed AI computing systems, and integrating with specific AI frameworks. However, solution architects are more general purpose and might not have such in-depth infra-awareness for AI focused systems.
• AI Architects are expected to address ethical and responsible AI concerns, such as bias, transparency, privacy, etc. It is their responsibility to ensure that the AI systems that they design are fair, unbiased, and respect individual privacy, human rights. Traditionally, solution architects do not have to consider such aspects in most of their day-to-day designs.

?What will the AI architecture practice do ?

The AI Architecture practice of an enterprise could help in any of the following ways :

• Rationalize all AI efforts/initiatives across the ecosystem.
• Work with the enterprise architecture practice, to layout AI Architecture principles, guidelines and standards that are applicable across the entire ecosystem.
• Evaluate possible AI service vendors for collaboration in critical aspects where the enterprise does not have sufficient expertise.
• Establish an audit framework/governance model to be able to ensure that all AI work – be it modelling, data collection, etc. is aligned with the best practices set at the enterprise level.
• Work with the bigger solution architecture team when they have specific requirements related to AI systems, algorithms, models, and so on.
• Lay out AI principles, and collaborate with Risk, Data Privacy and Protection teams to ensure that any AI related asset, artifact, or output generated by the enterprise is aligned with the AI principles, especially from an ethical bias, and data privacy perspectives.
• Work closely with the InfoSec teams to adopt frameworks, or processes to secure AI models, training data, put in place remediation methodologies for AI model theft, and so on.
• Work with enterprise DevOps & Operations team to guide AI operations, ML Ops, establish consistent support models, AI redressal mechanisms, and so on.

Conclusion :

In summary, it is safe to say, that the rapid push from business executives to adopt AI technologies has indirectly led to an urgent need for specialized AI architects, because the demands from an AI focused product/project is significantly different from what a traditional solution architect is used to.

Just like, there was an emergence of specialized Data Architects, and the specific practice of Data Architecture, with the advent of Big Data, Data Lakes; similarly with the advent of GenAI tooling like GPT, Copilot, it is apparent that specialized AI architects are needed to scale AI adoption across an enterprise.

These AI architects, who make up the AI architecture practice will be crucial for designing effective, ethical, and scalable AI systems. Their expertise can help organizations leverage the potential of AI technologies to solve complex problems and drive business innovation.

AI architects will be fundamental to the vision of building an enterprise-wide architecture practice, dedicated to AI, who will collaborate with data scientists, data engineers, security, legal, operations, developers, and so on, to help set up guidelines, processes, and frameworks that can help in rapid, yet governed AI adoption across the enterprise.