Six keys to generative AI success
As someone who has spent my entire adult life studying materials science, this is a very exciting time.
Generative artificial intelligence is upending our previous assumptions about how — and how quickly — new materials can be designed, with algorithms already helping companies and researchers to automate performance forecasting, improve existing materials, and even generate arrays of potential designs for new materials that meet specific criteria.
Understandably, companies in this space are racing each other to adopt and implement AI solutions that will give them a competitive edge. This is, doubtlessly, the right move, but organizations need to be thoughtful about how they build out their AI ecosystems — cutting through the hype and creating an environment that will result in real value.
Here are six important considerations.
Data Sources — Carefully think through what data will be used to train your generative AI systems. Publicly available data is easily accessible and available at a low cost (or even for free). It is also often expansive, covering more breadth than most companies’ internal data. For these reasons, it is typically an excellent starting point for AI training. However, competitors will also have access to this data, potentially limiting its upside value. Additionally, there can be a risk of inconsistencies or errors if the data isn’t well-curated, and it may lack the specificity necessary for niche applications. Proprietary data, by contrast, can be expensive to produce and may be limited in volume, but it will typically be a better fit to help solve an organization’s specific problems.
Intellectual Property — This is still an evolving area, but it’s an important one. If a company creates a new material using a combination of AI and publicly available data, it’s not entirely clear that the company will own the rights to the design. And in many cases, it won’t even be obvious when AI programs have used existing, patented ideas in the creation of a new design, since these tools typically lack transparency. Getting a grasp on this issue may require IT and business leaders to work closely with legal departments — and to keep an eye on this area of the law as it continues to mature.
Infrastructure — Advanced AI models rely heavily on the robust computational power provided by graphics processing units (GPUs). But as the demand for AI systems grows, we’ve seen a shortage of GPUs, with even the largest AI players fiercely competing with one another to get their hands on these essential resources. And even when organizations are able to procure the technology they need, the new infrastructure can create additional management and maintenance burdens. Business and IT leaders should carefully consider whether it makes the most sense, long-term, to buy and maintain their own infrastructure, or to partner with public cloud providers or specialized AI infrastructure vendors.
Talent — It’s difficult enough right now to find specialists with experience in AI. The demand is enormous, and since the field is relatively young, the supply of trained professionals is limited. Combine this with the challenge of finding people who not only possess AI skills but are also trained in materials science, and you’re going to be fishing in a very shallow talent pool. Rather than seeking out external expertise, organizations might consider training up their existing employees — particularly engineers with the sort of domain knowledge that will be necessary in setting up AI systems to answer the burning questions in the field.
Business Model — We don’t yet know what AI’s capabilities will look like five or 10 years from now. But just looking at the progress that’s been made over the past 12 to 18 months, it’s a safe bet that tomorrow’s applications will make today’s look pedestrian by comparison. This may mean that innovation essentially becomes commoditized. That’s a startling and, if we’re being honest, likely scary thing for people to hear. But that’s why it’s so important for business leaders to keep their companies at the forefront of technology, and to be bold enough to proactively adjust their business models rather than being disrupted by startups.
Tangible versus Hype — At the same time, businesses shouldn’t be placing all-in bets on AI applications that aren’t fully validated. Let me be clear: Generative AI is not a fad, and it’s not unsubstantiated hype. But there will be false starts and dead ends, and some seemingly promising applications are certain to end up in the graveyard of failed technologies. By exercising care along with their quickness, organizations can ensure that a single unsuccessful AI project won’t lead to a disaster — and can instead set themselves up to reap the rewards that surely await those who safely navigate the next few years.