Should your business invest in generative AI? There are many reasons to be excited about the future of this technology, but it doesn't necessarily mean it's a sound financial investment, at least for now.
The promise of generative AI is artificial creation. Generative AI models can leverage vast amounts of training data and input from users to generate content that fits a specific length, format, or topic. If you ask an image generation AI model to generate an image of a business logo with a red tree, it will come up with 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 significant risks for early investors. Not only is it unclear exactly where the market is heading, but also how much of an impact this technology will have on overall company productivity, or how quickly companies will see a return on investment. I don't have a clear idea about that either.
Can your company even afford to invest in generative AI?
While it can certainly be argued that generative AI has enormous economic potential, the large upfront investments in physical, digital, and human capital required are proving prohibitive. .
according to goldman sachsthe bulk of the investment in generative AI, which the company believes will be around $200 billion globally by 2025, will likely occur before the technology is widely adopted and before productivity gains are seen. expensive.
Generative AI is therefore considered a nascent technology and a major risk for companies looking to make significant investments. If your company plans to invest, you need a comprehensive implementation strategy.
According to Dell's Generative AI Pulse Survey, 76% of IT decision makers in medium to large enterprise-sized organizations in the US, UK, Germany, and France believe that generative AI will have a significant, if not transformational, impact on their organization. I think it will have an impact. [PDF].
While 20% of those surveyed have deployed generative AI tools and staff training, only 9% have established core use cases for generative AI. Meanwhile, 7% have no formal strategy for this technology at all, and a further 5% have ruled out using generative AI at this time.
Vendors have spent the past year coming up with a variety of platforms, architectures, and payment models for generative AI, including private AI training and off-the-shelf models through companies' current cloud providers. Analysts at CCS Insight have warned that small businesses will face the “prohibitive” costs of generative AI in 2024, so companies that have not yet adopted generative AI should take advantage of the technology at this early stage. They may hesitate to evaluate the merits and demerits of tackling this problem.
Make informed decisions about whether to invest in generative AI
As with all investment decisions, the decision to implement generative AI must balance the upfront costs against the subsequent benefits. Being confident that the later benefits outweigh the initial costs can sometimes be even more difficult.
For example, while return on investment for something traditional like a financial management system is relatively easy to model, leaders may struggle to quantify the exact benefits that generative AI brings to their business.
For example, switching accounting or financial management systems requires moving from one system to a new, proven alternative, and the return on investment can be modeled relatively easily.
One way forward is to use a series of gates or hurdles when testing generative AI. Modeling and piloting are good examples of gates, but they are not the only options. George Lynch, head of technology advisory at global technology consulting firm NashTech, said companies should “first try generative AI in-house, then have humans provide standard answers to standard queries. “You should consider choosing use cases that are relevant to the current processes you need.”
Another option is to try free public implementations to perform relatively low-level tasks. Although the potential of the results does not match that obtained from more serious implementations, this can be a useful step in the learning curve and is minimally costly.
Generative AI has an illusion problem
No matter what approach you take to gates, it's important to realize that the output from your generative AI model is only as good as the input you provide. Image generation models require thorough training and context, as well as carefully crafted prompts. For example, in the red tree image example above, the user must specify each element of the image to get the desired result.
“Displaying a prompt such as 'Meeting of Famous Computer Scientists' as a magazine cover art may produce an image that primarily features male scientists in the results due to historical data bias.” said Tauzin Yipen, a senior researcher at think tank The Conference Board. IT professional.
This highlights the limitations of generative AI prompts and provides examples of undesirable output that can result from improperly crafted input. Even with the best input or training data, generative AI models are prone to the following issues: hallucinationThis is when a model makes false claims about a particular topic or produces content that clearly deviates from its input.
Companies could be in serious trouble if generative AI produces misleading, biased, or patently incorrect information that enters the public domain or is used as the basis for internal decision-making. You may fall into it.
Business leaders need to understand this concept and the damage that AI models can cause. It would be a really bad day for a chief technology officer if a business's reputation and customer relationships were damaged by an illusion.
Generative AI comes with its own set of cybersecurity threats, including a new generation of social engineering operations and text tailored to phishing campaigns. Leaders should consider these generative AI cybersecurity threats when considering their business case for generative AI within their organizations.
Investing in generative AI could be a one-way ticket
Generative AI should be thought of as just another tool aimed at achieving business outcomes. “If it does not improve operational efficiency or is not viewed favorably by end customers, then its use should be carefully and quickly considered,” Eapen says.
As just another tool in a box, unplugging should always be an option. However, that decision can be difficult to make and to implement. If your organization uses public or open source implementations of generative AI for relatively low-level tasks, stopping it is easy.
However, large enterprises that rely on the services of cloud computing services to manipulate existing data and store what is created in the cloud face challenges, similar to the challenges seen with cloud repatriation more generally. may be faced with a solution.
The risks of generative AI are high. The rewards can be huge, but despite the hype and buzz, they are not truly embedded in today's businesses. Many are taking a steady and cautious approach. Asking questions, maneuvering, learning as you go, and watching for a graceful retreat may be a healthy approach.
