The cloud offers several advantages for implementing generative AI models, and we’ve discussed that to death here. In short, the cloud provides scalable computing power, flexibility, and accessibility, enabling enterprises to find the full potential of generative AI.
Cloud infrastructure allows seamless access to vast training data. Although it can be pricey, it also facilitates model development and refining. Furthermore, it enables faster and more efficient model training and inference, making generative AI more accessible to a broader range of users.
Slower adoption than expected
Based on what we’re seeing in the press, you would think there is a vast generative AI party out there. However, the reality of adoption is a bit different. Despite the clear benefits of generative AI in the cloud, I’m not seeing a massive move anytime soon at the volume many believe is occurring. And there are a few good reasons:
The skills gap is a major issue. Implementing generative AI models in the cloud requires machine learning, cloud computing, and data engineering expertise that does not exist at the level needed to be successful with this technology.
Enterprises need more skilled professionals who possess both a deep understanding of generative AI tech and how it can return value to the business. Thus, most enterprises are discussing generative AI but doing nothing yet.
Generative AI, and AI in general, is not something you can absorb in a weekend. It takes months of understanding the data, model implementation and tuning, and knowing when the darn thing is working correctly. I applaud those who have delayed implementation until they get the skills in-house; we learned from cloud deployments that a lack of qualified architects and developers usually causes projects to fail.
That said, a few enterprises are pushing ahead without the needed skills. We’ll hear about those failures in a year, as the inevitable generative AI hangover comes. I’ll point that out here.
Data isn’t ready yet. Generative AI models require high-quality data to learn and generate meaningful outcomes, and most enterprises don’t have a handle on that yet. Acquiring, cleaning, and preprocessing data is a significant challenge, especially when combined with heterogeneous data sources, privacy concerns, and data management regulations.
Organizations must invest time and resources to ensure data availability and quality before generative AI in the cloud can be a helpful resource. That takes more time and money than most enterprises understand. Pressing forward without dealing with the data is another surefire way to fail, and it’s good to delay the implementation of generative AI in the cloud until that problem is solved.
Setting policies is hard and politically charged. How do you protect against bias that will get you sued? Are you creating data regulation issues by taking unregulated data, using generative AI, and having regulated data pop out? What is the policy on people getting displaced by this technology?
Leveraging generative AI in the cloud is cost-intensive, particularly if not adequately optimized. Organizations must carefully evaluate the cloud resources required for model training and inference to strike a balance between cost and performance. Most will want to turn on the cloud computing tap, resulting in substantial cost overruns and little value returned to the business. We’ve made these mistakes with most cloud innovations in production, including serverless computing and container orchestration; it’s a surefire bet that we’ll do the same here, if not careful.
What to expect
If we’re going to be slow-rolling generative AI in the cloud, when will it show up at a level that moves the needle? For most, it will be much longer than expected.
I suspect we’ll see many proofs of concept next year, showcasing this technology’s capabilities. However, POCs only go so far as to bring value back to the business. For that, you need production systems that do high-value things, such as providing a better customer experience, intelligently automating a supply chain, finding the actual risk of insuring a driver, or diagnosing a patient with a more significant amount of digital expertise. You know, stuff that makes money.
I suspect we won’t see the larger value from this stuff for three or four years—something that’s not mentioned in the tech press because we have ADD in the technology marketplace. We’re not interested in stuff that far away.
However, generative AI is a major shift in how we deliver systems. I would rather wait and do it right than rush something out and fail, or worse, cause damage to the business. Most IT executives may feel justified to move aggressively, given the hype. They will likely be looking for jobs in a few years. Don’t be those people.