With ChatGPT, we have undoubtedly reached the next level of capabilities in artificial intelligence. The vast amount of data processed and the hundreds of millions of users contributing to the model’s improvement through feedback loops make it significantly more powerful than any language model ever before. Nevertheless, recognizing patterns from history can help predict how the next 12-18 months might unfold, enabling Chief Data and Analytics Officers (CDAOs) to better prepare and manage expectations. History doesn’t repeat itself, but it can teach valuable lessons.

The societal impact of major technological breakthroughs, such as personal computing and the Internet, can be immense but often takes time to fully materialize. Technological innovations typically emerge within specific domains, addressing unique challenges or advancing specialized knowledge. While not universally applicable, these innovations can ignite excitement and generate technology hype. Enthusiasm can arise from overestimating a technology’s immediate applicability across industries or underestimating the time and effort needed for adaptation to broader contexts.

When I joined IBM in 2013 as part of their global advanced analytics competence center, the Watson hype was in full swing. Watson had famously defeated Jeopardy! champions Ken Jennings and Brad Rutter in 2011. In the following years, companies would lose millions of dollars on proof-of-concept projects that often fell short of expectations. The Watson hype gradually transformed from high hopes to widespread disappointment about AI. To better understand the trajectory of Generative AI, CDAOs can learn from this recent history.

IBM Watson’s 2011-2015 hype offers valuable insight into the current excitement around ChatGPT. Watson’s Jeopardy! win and advanced NLP capabilities set high expectations for its applicability across industries. It used machine learning and knowledge representation techniques to analyze data, learn, and improve. Its comprehensive knowledge base enabled it to understand various subjects and answer complex questions.

However, the hype led to unrealistic expectations that Watson struggled to meet in real-world scenarios. Its complex architecture required significant customization, making implementation costly and time-consuming. Rapid AI advancements and more specialized solutions outperformed Watson, reducing its competitive advantage. Concerns about interpretability and industry-specific adaptation hindered its ability to provide accurate insights, particularly in industries like healthcare and finance.

Recognizing that breakthroughs may need further development, customization, or integration before effectively applying them in other domains is crucial, as not every innovation leads to immediate widespread adoption. By maintaining a realistic understanding of a technology’s limitations and potential, stakeholders can better navigate the hype and make more informed decisions about implementation and use.

Generative AI has undoubtedly made significant strides in recent months. The models are more sophisticated, capable of consuming vast amounts of data from the internet, and continuously improving based on user feedback. As ChatGPT is now used by 100 million people, its development will continue to accelerate, leading to substantial enhancements. However, as a natural language professor recently told me, "GPT-4 is bigger and more powerful, but it is essentially just another model."

As Generative AI and tools like ChatGPT continue to impact various industries, they inevitably reshape the roles of Data and AI leaders. While these technologies present tremendous opportunities for businesses, they also pose inherent risks and challenges:

• Overestimating the commercial potential across domains
• Distraction from building a strong internal data foundation for the AI era
• Insufficient controls to separate correct answers from plausible but incorrect or inconsistent responses
• Potential generation of offensive, biased, or inappropriate content
• Lack of data privacy controls hindering adoption in commercial use cases
• Security concerns

The most pressing question is whether the current generation of Generative AI can create a significant business impact in commercial applications within the next 12 months, particularly when integrating proprietary data.

AI has proven commercially and strategically valuable when used with proprietary data for applications such as predictions, recommendations, pricing, forecasting, and fraud detection. Top executives see the potential of Generative AI to automate jobs in areas like marketing and customer service. Numerous high-cost proof-of-concept projects are currently underway at large organizations.

However, many of these projects may deliver less value than anticipated, as was the case with the Watson hype. If a majority of commercial projects fail in 2023 due to unrealistic expectations, lack of patience, compliance concerns, and the technologies’ unreadiness for such use cases, it could trigger another round of AI disillusionment in 2024.

Simultaneously, CDOs will need to cope with technologies that are still in beta. CDAOs must champion ethical AI practices within their organizations to minimize the risk of biased or harmful AI-generated content. They should work closely with cross-functional teams to establish guidelines and processes that promote responsible AI use. To protect data privacy and maintain data security, CDAOs should develop comprehensive data governance frameworks and enforce strict policies.

They must also collaborate with other stakeholders, such as legal and compliance teams, to ensure AI-generated content adheres to relevant regulations. Successfully navigating the AI revolution requires CDAOs to strike a balance between seizing the opportunities presented by Generative AI and addressing the challenges and risks associated with their implementation.

My advice to management boards is to avoid losing a lot of money on being early adopters of Generative AI as long as skills are rare and there are no established practices and enterprise-ready software tools yet. It is important to experiment with the technologies and observe until things become more stable and good practices and solutions have been established.

I would recommend doubling down on the investments made into your proprietary data and capabilities for data management at scale leveraging new paradigms like data mesh and data fabric. Your unique data will continue to give your company a competitive advantage when AI comes increasingly out of the box or is programmed for you by a machine.

About the Author

Alex Borek is a data executive, bestselling author, highly reviewed keynoter, content creator, data masterclass trainer, and advisor. He helped more than 50 organizations in Europe to advance their data journeys. He is the organizer of the Data Masterclass Europe 2023 in Berlin, which connects 70 CDOs across Europe to learn from each other on best practices.

As Director of Data Analytics, Borek drives the data mesh transformation across the Zalando Group, Europe’s largest online fashion retailer with more than 50 million customers. He co-leads the Zalando Data Foundation, the central data tech team of 130 people covering platform, data, analytics, and AI.

He holds a doctoral degree in Engineering from the University of Cambridge in data quality risk management.