From family dinners to weekend afternoons, I’ve spent a lot of time over the past six months playing with generative AI tools and thinking about how they will change “everything”. I’m increasingly confident they’ll have an impact, but it won’t be as massive or as fast as some might think, especially in the enterprise.
Let me start with all the reasons why generative AI takes some time to really scale into enterprise business processes and have a measurable impact on productivity. Primarily, enterprises achieve scale by implementing process controls and then automating systems. From inventory management to hiring, the key to scaling business systems is the ability to move people’s work efforts from individual transactions or activities to managing end-to-end processes.
Paul Brody is EY’s global blockchain leader and CoinDesk columnist.
Take something as simple as stocking a supermarket with food. Enterprise systems and retail point-of-sale (POS) systems have been carefully integrated over the years to automatically reorder out-of-stock items and, more importantly, to systematically predict and plan to avoid they are ever out of stock.
Generative AI systems, on the other hand, are not good at rigorously and consistently performing the same task over and over again with high precision. Ask a generative AI system similar but not identical questions and you might get vastly different answers. This kind of variance breaks business processes that are based on input consistency.
Generative AI systems are great at coming up with new ideas, and doing so at a tremendous speed, but business transformation is largely about change management – both people and systems. Enterprise ecosystems tend to transform at about the same rate as the slowest components in the ecosystem, not the fastest.
A good example of this comes from the early era of web commerce. It was soon possible to build web-based storefronts and accept credit card payments. However, shipping and packaging are built and optimized for a world of pallet-sized retail deliveries. To the extent that companies even had digital catalogs, they didn’t have photos of products. No grocery store supervisor needs to know what a can of soup looks like. They already know. They’re in the store every day. As a result, e-commerce took off much slower than analysts expected, hampered not by the Internet, but by warehouses and logistics systems.
Like e-commerce, generative AI systems will infiltrate enterprise systems alongside blockchain technology and they will eventually work very well together, but progress will be driven by careful design and integration, not rapid, large-scale adoption. While consumers are often able to adopt new technologies broadly in about a decade, it typically takes about 25 years for enterprises, and we should probably expect the same with generative AI and its integration with blockchain technology.
Now that I’ve got the bad news out of the way, I want to focus on the areas where we’ll see the most dramatic impact of how these two technologies will work together. I’ve identified four that could come sooner rather than later.
Enterprise business processes run on software and generative AI systems are exceptionally good at software development. It’s one of the few areas where we have strong, documented evidence that generative AI systems significantly improve productivity. Since integrating blockchains into business processes is a matter of both process and software integration, the likely impact will be significant and will be felt soonest.
Blockchains do a great job of improving data quality. When you think about products, services and systems moving from one company to another, data quality is one of the biggest victims of cross-company work. In a world of silos, data is re-entered into every business ecosystem. On a blockchain, tokens and hashes represent assets and data and can maintain their integrity as they move through an ecosystem. With better quality data, you can expect generative AI systems to perform even better analytics.
It also works the other way around: generative AI systems are great at matching and interpreting patterns. They will become the basis of blockchain analysis in a very short time, allowing trends to be identified and individual transactions to be classified.
One of the biggest emerging problems for AI systems is finding reliable source data. We’re in the early stages of an exa-tidal wave of AI-generated content. Much of it will be banal, generic and mediocre. How do we know what constitutes an authoritative, expert opinion on a topic or machine-generated pattern based on other machine-generated patterns? By verifying the authenticity and provenance of source data using blockchain hashes.
Italy’s ANSA news agency already legalizes nearly 1 million articles a year using EY’s OpsChain system. This was intended to combat fake news, but in the future, tools like this could be critical for verifying the sources of AI training data.
Just as generative AI systems are good at writing code, they are also good at interpreting error messages, problems and proposing solutions. Blockchain usage is still too complex, and conversational interfaces that can accept error messages, find and format suggestions, and act as a “co-pilot” in a process are likely to be immensely helpful to users.
In the early days when new technologies evolve and interact, the results tend to be both boring and predictable, as I described above. We saw this with GPS and Web commerce and mobile phones. In the beginning, we had an e-commerce experience that was little more than a paper catalog on a screen. Finally, we got push ads coming to us in a ride-sharing vehicle proposing to have food delivered to our destination.
And so it will be here as blockchain and AI begin to evolve and converge. We’re in the boring phase, but just wait until things get weird and wildly unpredictable. Because they will.