AI and Digital Transformation

You’ve been thinking about digital transformation for years. Maybe you’re even somewhere down the path of transforming your organization. And now there is this AI thing that looks real, even if you were skeptical at first. Does this mean you need two transformations? And then another one when quantum computing comes around? Do you need to shift the course of your current transformation? Or is it all just so overwhelming that you’re tempted to wait it out and let the next crop of executives in your company figure out what to do?

These questions would be easier to answer if we agreed on what digital transformation is and perhaps on what AI’s business implications will be. There’s no consensus, but let me give some definitions we can work with.

Digital transformation is the process of bringing digital technologies, practices, and ways of thinking into an organization. A digitally adept organization provides a superior customer experience through digital channels and harnesses digital-age practices and ways of thinking to become nimble, speedy, and cost-efficient.

AI is a group of technologies that act in ways we associate with human intelligence rather than mechanical computation. There are many approaches to AI, including symbolic reasoning and problem-solving models, causal models, and generative models—and more will soon be developed. AI is a frontier. It is the area of IT where innovation is rapid. New ideas and products emerge rapidly as the underlying science evolves. Costs have not yet stabilized. Capacity is limited today because of hardware constraints but will increase quickly. And money is pouring into the field, with the economics yet to be determined.

Common Elements

That leads to our first observation on the relationship between AI and digital transformation. Digital transformation teaches us how to manage rapid change and uncertainty by maintaining agility and fostering continuous learning. Through digital transformation, organizations build technical and cultural agility. They deliver IT product quickly and incrementally, with short feedback cycles and continuous improvement.

An organization that builds agility through its digital transformation is in a good position to handle the uncertainty of AI and other emerging technologies, such as quantum computing. You might say that digital transformation is a prerequisite for successfully adopting AI.

A second area where digital and AI transformation overlap is in the centrality of data. Digitally transforming organizations aim to become data-driven. They reconsider the health of their data and its availability across business silos while governing it to ensure privacy and protect intellectual property.

AI (to state the obvious) is also data-centric. Data is used to train and fine-tune machine learning models; it is also provided to models through RAG and used to establish domain-specific knowledge graphs. AI is part of an organization’s data pipeline. So digital transformation, through its attention to data, is simply part of AI transformation.

Digital transformation introduces new ways of thinking and working, such as incremental, iterative delivery with fast feedback cycles. These approaches are even more important with AI, which has unpredictable behaviors that require a “try it and see” mentality. A culture comfortable with incrementalism and experimentation benefits both digital and AI transformation.

New Capabilities

AI does change some things. For example, it substantially broadens the set of business problems amenable to an IT solution. At the risk of oversimplification, IT can now meet a huge set of business needs it couldn’t before by providing approximate, nondeterministic solutions. It can “solve” business problems that are imperfectly defined yet nevertheless understandable to humans.

The scope of those problems is open to your imagination. You may find that some AI initiatives now outrank earlier priorities in your digital transformation roadmap. Your goals may not change, but AI may offer better ways of accomplishing them. You may even be able to leapfrog steps by moving directly to AI solutions.

Obscurity and Nondeterminism

Just as human intelligence is a mystery in many ways, so is machine intelligence. A generative AI model’s intelligence is contained in hundreds of billions of parameters—far too many for us to fully grasp how it makes decisions. For that reason (and because it makes some of those decisions randomly within statistical parameters), we have to consider generative AI as nondeterministic—we can’t always predict what it will do. It is very different from traditional IT capabilities.

Nondeterminism based on statistical parameters will be a factor in quantum computing as well. It also applies in chaos engineering, a good practice in digital transformation. Chaos engineering is built on the premise that complex systems can fail in ways we can’t foresee—the only way to know about and deal with potential failure modes is to experiment on the system. Sometimes the only way to know how an LLM will behave is to experiment on it. We may even find that it has emergent abilities (i.e., abilities we didn’t expect).

Nondeterminism requires new practices and ways of thinking. For example, common digital approaches like CI/CD make assumptions of predictability. They rely on an automated test suite to validate behavior and on regression testing to confirm that behavior remains stable when changes are made. They assume that the only possible cause of a behavior change is whatever (small) incremental changes were made to the code; otherwise, the behavior should not “regress.” This reduces risk and supports frequent deployments. AI applications can be tested to make sure they stay within guardrails, but not always tested against known specific behaviors.

Changing the Basis of Trust

A parallel change in thinking about deployment is also necessary. An organization is willing to deploy software when it believes risk has been adequately mitigated. It reaches that point by thoroughly testing the software. But if the software’s behavior is nondeterministic, knowing when you have reached that point is more difficult. And with natural language interfaces, you have less control over what users may type or how they will use the software, so it’s harder to know what cases must be tested. If you are using a foundation model, upgrading or changing the model also introduces risks—regression risks, the scope of which is hard to determine.

There are ways to mitigate these risks—setting up guardrails by using Amazon Bedrock Guardrails, for example. Automated reasoning can be used to validate outputs. But you can’t make the risks disappear.

This isn’t really new. When you hire a human, you don’t know for sure how they will behave. You can interview them, monitor them, and give them feedback. But there is always a risk of poor decisions and inappropriate acts. Organizations accept this risk partly because they can reassign or terminate employees. But by then it might be too late.

You can’t deploy capabilities, whether provided by a human or software, into production until you trust them. The basis of trust for AI is different. You will need to oversee a cultural shift that allows for trust in autonomously acting AIs and consensus on reasonable criteria for that trust.

Again, this is not so different from digital transformation, where leaders must learn to trust autonomous, empowered teams. A large part of cultural change involves new patterns of trust in an organization. Deciding whether to trust AI has a lot of parallels with deciding whether to trust human employees.

AI Transformation is Digital Transformation Plus

This is certainly not a comprehensive list of how AI changes the course of digital transformation, but it shows that many of the considerations are the same or similar. Both involve building new patterns of trust. Both center on increasing agility and speed, mainly through fast feedback cycles and incremental delivery. And both require attention to data’s movement and use across the enterprise.

Yes, in modernizing the company’s technical stack for AI, a new set of tools must be considered. If you’re using the cloud, those new tools are readily at hand—and using them is like using other cloud services.

The goals of an AI transformation align neatly with those of our digital transformations. The big difference is learning to work with nondeterministic systems whose internals are somewhat obscure. Whether or not we consider AI “human-like,” we already manage similar uncertainty with human employees. Successful AI transformation depends on building governance and trust models designed for this nondeterminism in addition to the changes introduced in digital transformation.