In the past 18 months, generative AI has shifted from a curiosity to a business imperative. The barriers to entry have been largely dismantled, allowing individuals and organizations to experiment and innovate with these powerful tools. But here's the thing: the enterprise sector is at a crossroads. While we are seeing significant advancements in the consumer industry, the enterprise sector demands a more measured and bespoke approach.
AI is like an iceberg
Just like an iceberg, AI can be divided into two distinct parts: consumer AI and enterprise AI. While consumer AI is the flashy, visible tip of the iceberg, enterprise AI is the much larger, yet lesser-known, foundation that powers the majority of AI applications.
Consumer AI refers to the AI technologies that are designed to serve individual consumers. By now, we all know them well – virtual assistants like Siri, personalized shopping recommendations when you are scrolling, and countless others. I see these applications as the "face" of AI for the average consumer. However, they represent only a small fraction of the total AI landscape.
Enterprise AI, on the other hand, refers to the AI technologies that are designed to support business operations and decision-making – think supply chain optimization, customer service chatbots, digital assistants, and beyond. Enterprise AI is often invisible to the end-user, yet it has a profound impact on the efficiency, productivity, and competitiveness of organizations. It is the "engine" that powers the majority of AI applications and is what enables businesses to make data-driven decisions and stay ahead of the curve.
The opportunity for generative AI in enterprise is substantial, with potential market sizes reaching into the trillions of dollars. But it is not one size fits all.
What businesses need are productivity enhancements at scale. To achieve this requires a more nuanced approach, one that acknowledges the diversity of needs and the importance of domain-specific solutions.
With that in mind, large models are not always the most suitable choice. These models are often expensive to run and may not provide the level of customization that businesses require. Enterprises are asking for smaller, fit-for-purpose models that can deliver high-quality results at a fraction of the cost, with a focus on a specific use case. For example, a large banking client may require a model that can perform risk assessment that understands regulatory requirements.
The productivity paradox
Technology is advancing faster than ever, but productivity gains are not. Financial success is dependent on productivity, and AI is the answer to the productivity problem. Developers using generative AI are experiencing benefits like increased accuracy and freed-up time. Yet most enterprises have not moved beyond experimentation.
One barrier to deployment is the effective use of enterprise data in generative AI. Most organizations are not leveraging their own data to its full potential. With tools and techniques for fine-tuning models, enterprises can unlock the value of their data and address specific business needs. The biggest opportunity here is open-source AI, which allows companies to train and build models using their own proprietary data, along with prioritizing trust, transparency, and governance.
Domain-specific models: The future of AI
At IBM, we are building a diverse range of language models, including those for major Western languages, Arabic, Japanese, Spanish and others. Additionally, we are developing models for coding, time series, cyber, numerical, climate change and geospatial applications. Our focus is on creating domain-specific models that can deliver high-quality results for specific use cases, without sacrificing quality or scalability. With this sustained evolution, we are starting to see a continuum and shift from automated to autonomous. AI is starting to go from single-step processes, information retrieval and prescriptive tasks to multi-step processes, autonomous action-taking and self-correcting systems.
As we unlock the full potential of generative AI, it is essential to address the risks associated with its misuse. By promoting transparency, accountability, and responsible innovation, we can ensure that these powerful tools are used for the betterment of society. This involves governance, guardrails and the right humans – with the right skills and training – included in the process. But let's be real: the current governance frameworks are not equipped to handle the complexity of AI. We need to rethink the way we approach AI governance and create new frameworks that prioritize transparency, accountability and human values.
Activating AI within the enterprise requires a nuanced understanding of the opportunities and challenges that arise in the enterprise sector. By delivering tailored solutions, promoting transparency and accountability, and leveraging the power of domain-specific models, we can unlock the full potential of generative AI and drive business value at scale. But it is not going to be easy. We need to be willing to challenge the status quo, think differently and prioritize human values above all else.
Jonathan Adashek is senior vice president, marketing and communications for IBM, adding to his previous responsibilities as chief communications officer. He is responsible for overseeing the company’s global marketing, communications and corporate social responsibility organization, which includes full funnel marketing, corporate affairs, and ESG. In addition, he is responsible for federal client business development. Jonathan aligned IBM’s marketing and communications efforts under a single brand platform, Let’s create, in support of IBM’s mission to become the leading hybrid cloud and AI company. He is a member of the USC Center for PR board of advisers.