Generative AI in Real-Time Decision Making: Applications in Dynamic Environments

Artificial intelligence had a banner year in 2023, capturing headlines for generative tools that went mainstream (ChatGPT, DALL-E, and Midjourney leading the way) and the reactive efforts by world governments to assess and start reigning in what AI can and cannot do. 

While these hands-on applications are likely to be the first (and in some cases only) examples most people think of when they hear “AI,” it is the application of advanced machine learning in dynamic environments that will have the most substantial long-term impact on our lives. And yet, many of the most impactful developments in this space remain years away. McKinsey predicts that Level 4 driving automation is still more than a decade away from mass market adoption. The EU AI Act is assessing where AI might be used in real-time environments as well, identifying several “high risk” categories, such as management of critical infrastructure, law enforcement, and border security, that will be more highly regulated (though not banned). 

One thing we do know, however, is that AI’s growth curve is not linear, and while projections might put generative and powerful AI systems years away, they are rapidly becoming more important in day-to-day operations, from analytics to emergency response. 

Generative AI’s Current and Growing Capabilities

Going into 2024, one thing we know is that generative AI is set to have a sweeping impact on almost all aspects of the economy. In some ways, it already has. SAS Chief Technology Officer noted that many companies will shift toward an integrative approach, pairing GAI with existing AI strategies based on individual industries. This manifests in banking as the GAI simulation for analysis and stress testing, the generation and communication of individualized care plans in healthcare that “combat the realities of being human (including fatigue and inattention); and the optimization of production in manufacturing set to generate $1 trillion in value by 2030.  

It is in analysis, in particular, that generative AI is best positioned to have an immediate impact on our systems. Generative AI systems are able to better create or generate new content in the form of text, images, or search results and communicate the results to non-technical users. For example, fraud has become a massive liability in the financial industry. A recent report showed that financial crimes increased by 43% in 2023, with total losses rising by 65%. Total losses are expected to pass $40B by 2027, but AI offers a resource to more efficiently identify fraud risks and communicate more clearly and effectively with consumers about those risks. 

Another burgeoning application lies in the increasing capabilities of robotics. While the hardware is not new–robotics have been supporting efficiency in manufacturing since the 1960s — the software running these robots is becoming increasingly intelligent. This has become an urgent issue in countries with rapidly aging populations like Japan and Germany, where they are losing millions of available workers each year. The integration of generative AI with robotics allows machines to start taking on real-time jobs that previously had to be completed by humans. As self-driving cars become smarter, humans will be able to hire a car service without the need for a driver, for example, and many organizations are starting to experiment with broader automation in place of low-wage employees. McDonald’s, for example, launched its first fully automated restaurant in January 2023 and, late last year, announced the broad integration of generative AI in its kiosks and ordering apps to improve the self-serve experience. 

A recent Valoir study found that AI could potentially automate up to 40% of the average workday. We’ve already seen the spikes in productivity offered by tools like ChatGPT that can gather and organize materials, support research, set up and process spreadsheets, and even write code. Many companies are already leveraging AI to automate, with an average of 20% of previously manual tasks being automated in the last two years. 

Concerns and Challenges in Real-Time AI Implementation

While the use of AI to supplement fast food workers and streamline office productivity is well underway, there are questions about how much we can and should rely on GAI in real-time scenarios, especially in dynamic and unpredictable environments. 

This balance of cutting-edge capability with technical and ethical considerations is perhaps best seen in the US Military, which established the Chief Digitial and Artificial Intelligence Office (CDAO) in the Department of Defense in mid-2022. The implementation of AI to support and enhance electronic warfare is a crucial part of the DOD’s strategy moving forward, and investments are being made by dozens of research branches within the armed services. AI is being used to support autonomous operations of drones and aircraft, threat assessment and response from digital attackers, and more. The big concern, of course, is how to balance the autonomy provided by these systems with acceptable risk and vulnerability from systems that are not touched by human operators. 

Another major concern around broader AI and ML integration is the cost of standing up new pilots to test applications. Business and IT decision-makers must be more engaged than ever to ensure AI is being used in a way that supports broader performance without creating more risk vectors for the organization. There is also concern over AI’s environmental impacts. Stanford’s AI Index Report, for example, reports that a BLOOM training run emitted 25 times more carbon than a single air passenger flying cross country, and Scientific American reports that training OpenAI’s GPT-3 “produced the equivalent of 500 tons of carbon dioxide.” With an ever-increasing focus on sustainability in technology, AI needs to find a balance between the energy savings it can help realize and the energy it produces to do so. 

Where We Go Next

Like electricity at the tail end of the 19th century and the launch of the commercial internet at the end of the 20th century, generative AI is poised to have a seismic impact not only on the economy but on how we interact with and engage with each other and the broader world. To do so safely will require careful attention to the balance between the key benefits of these systems and potential risks, as well as continued focus on sustainable, ethical growth. With the right approach, however, AI will have a monumental impact on real-time environments across industries.