Should your business invest in generative AI? Well, while there are many reasons to be excited about the future of the technology, that doesn’t necessarily make it a sound financial investment – at least for now.
The promise of generative AI is that of artificial creation. Drawing on vast quantities of training data and inputs from a user, generative AI models can produce content to fit a specific length, format, or topic. Ask an image-generation AI model to produce an image of a business logo with a red tree and you’ll be shown a number of images that fit the brief.
But while there are many reasons to be excited about the future of AI, the technology also comes with enormous risk for those investing early. Not only is it unclear exactly where the market is heading, we also don't have a clear idea as to how impactful the technology will be for productivity across the enterprise, or how quickly a business will see a return on investment.
Can your business even afford to invest in generative AI?
It could certainly be argued that generative AI has enormous economic potential, but the large up-front investment of physical, digital, and human capital required is proving prohibitive.
According to Goldman Sachs, the vast majority of investments in generative AI, which it believes will be around $200 billion globally by 2025, will likely happen before widespread adoption of the technology and before we see any gains to productivity.
It’s because of this that generative AI is still seen as a fledgling technology that represents a larger risk for any businesses looking to invest heavily. If a business does plan to invest, a comprehensive adoption strategy is required.
Among IT decision makers at medium to enterprise-sized organizations in the US, UK, Germany, and France, 76% think generative AI will have a significant, if not transformative, impact on their organization, according to Dell’s Generative AI Pulse Survey [PDF].
While 20% of those surveyed have rolled out generative AI tools and training for staff, only 9% have established core use cases for generative AI. Meanwhile, 7% have no formal strategy for the technology whatsoever, while a further 5% have ruled out the use of generative AI for the moment.
Vendors have spent the past year coming up with a range of platforms, architectures, and payment models for generative AI such as private AI training or off-the-shelf models through a firm’s current cloud provider. As analysts from CCS Insight warn smaller firms face “prohibitive” generative AI costs in 2024, those who haven’t yet bought into generative AI could hold back to assess the benefits and drawbacks of committing to the technology at this early stage.
Informed decisions on whether to invest in generative AI
Like all investment decisions, whether or not to implement generative AI is a balancing act between initial outlay and later benefits. Having the confidence that the later benefits outweigh the up-front cost might be more challenging than in some cases.
For example, while the return on investment for something as traditional as a financial management system is relatively easily modeled, leaders may struggle to quantify the precise benefits that generative AI brings to their business.
For example, switching one’s accounting or financial management systems involves moving from one system to a new, tried and tested alternative, and the return on investment is relatively easy to model.
One way forward is to use a series of gates or hurdles when it comes to testing generative AI. Modeling and piloting are good examples of gates, but far from the only options. George Lynch, head of technology advisory at global technology consulting firm NashTech suggests businesses “should consider first trialing generative AI internally and second, selecting a use case that relates to current processes which require humans to provide standard answers to standard queries”.
Another option is trying free public implementations to do relatively low-level tasks. While the potential power of the results will not approach what you might get from a more significant implementation, this might be a helpful step on the learning curve and it requires minimal outlay.
Generative AI has a hallucination problem
Whatever approaches to gates are taken, it is important to recognize that the outputs from generative AI models are only as good as the inputs users provide. Image-generation models require extensive training and context along with carefully-crafted prompts. For example, in the earlier example of the red tree image the user would have to specify each element of the image to produce the desired result.
“A prompt such as ‘meeting of famous computer scientists’ for use as cover art in a magazine may produce an image predominantly featuring male scientists in its results due to historical data biases,” Tojin Eapen, senior fellow at thinktank The Conference Board tells ITPro.
This exposes the limitations of generative AI prompts and provides an example for the kind of unwanted output that can arise from a poorly-crafted input. Even with the best inputs or training data, generative AI models are also prone to hallucinations, in which a model makes false claims on a given topic or produces content that does diverges noticeably from inputs.
A business could be in severe trouble if a generative AI produces misleading, biased, or plain wrong information that makes its way into the public domain or is used as a basis for internal decision-making.
Business leaders must understand this concept and the damage AI models could cause. It would be a bad day indeed for a chief technology officer if a hallucination damaged their business’ reputation or relationship with its customers.
Generative AI comes with its own cyber security threats, such as a new generation of social engineering operations and tailored text for phishing campaigns. These generative AI cyber security threats have to be taken into account when leaders examine the business case for generative AI within their own organizations.
Investing in generative AI could be a one-way ticket
Generative AI should be considered just another tool in your box aimed at achieving business outcomes. “If it does not improve business efficiency or is not favorably viewed by end customers, its use should be reviewed carefully and quickly,” says Eapen.
As just another tool in the box, pulling the plug should always be an option. However, both making and implementing that decision might be challenging. If the organization is using a public or open source implementation of generative AI for relatively low-level tasks, then stopping should be straightforward.
But larger businesses using services from their cloud computing service to manipulate existing data and store what is created in the cloud may find exiting a challenge, just as they might find challenges with cloud repatriation more generally.
The stakes are high with generative AI. The rewards could be grand, but despite the hype and buzz, it is far from truly embedded in businesses today. Many are taking a steady and careful approach. Asking questions, piloting, learning as you go, and keeping an eye on a graceful retreat may be a sound approach.
More on generative AI
