AI in Organizations

Ajay Agrawal, Joshua Gans, Avi Goldfarb: Power and Prediction (2022b, pp. 3-4)

 ''The reality is that we are already living in the early days of the AI Age, and we need to make some very important decisions about what that actually means.[…] Many people in organizations will play a role in shaping what AI means – but to make those choices matter, serious discussions need to start in many places, and soon. We can’t wait for decisions to be made for us, and the world is advancing too fast to remain passive.''

Ethan Mollick: Co-Intelligence (2024, pp. 32, 210)

• …act humanly, for example through natural language processing, knowledge representation, reasoning, computer vision, or robotics
• …think humanly, by representing of cognitive functions
• …think rationally, building on logics and processes beyond human capabilities
• …act rationally, the most commonly used definition, perceiving the environment, adapting to change, and pursuing goals to achieve the best possible outcome

The first distinction is between systems that rely on static, pre-determined rules, and those that are capable of adaptation and improvement over time.

• Rule-Based AI: Early systems, operating on pre-defined logics. Useful and efficient in handling well-defined tasks, but limited in complex and dynamic scenarios
• Machine Learning: Modern systems with self-improving algorithms, improving results with new training data. Includes various types of algorithms such as deep learning neural networks

A second distinction is by way of versatility in changing domains and the respective performance

Narrow AI: Designed for specific tasks with clear instructions and given training data. Uses ML algorithms to perform tasks efficiently and improve over time. Implemented via supervised, unsupervised, and reinforcement learning, using various algorithm types from decision trees to neural networks

• Generative AI (genAI): The latest frontier and most prominent trend in AI, focused on creating new content based on large input datasets. GenAI tools scan the internet, generate text, images, or code, opening new possibilities for content creation and problem-solving
• Artificial General Intelligence (AGI): The ultimate goal of AI development, not yet achievable. AGI could address complex, universal problems, matching or surpassing human intelligence across a wide range of tasks and domains.

At its core, AI is a prediction technology. Using input and training data, AI predicts the future and recommends outcomes or actions. For organizations, this predictive power can be applied in countless ways, such as strategic decision-making, expert analysis, or creating new content with sequences of words or numbers.

For some organizations, AI may upend their industry or significantly shift the product and services they are providing. For these organizations, AI affects the organizational core; the source of their competitive advantage

For these “shift towards AI” organizations, AI is a catalyst for fundamental transformation, it creates new business models and promises to upend industries. Creating AI solutions tailored to their specific requirements is essential and will become the new core of their business.

For other organizations, their competitive advantage may not be as directly affected by AI. For them, AI might be relevant for the organizational periphery, outside the source of their competitive advantage, as a tool promising efficiency and optimization, streamlining processes and enhancing effectiveness. For “work with AI” organizations, making use of AI efficiently is important. Here, buying and integrating plug-and-play solutions is often the best path forward.

Given the dynamic nature of AI, continuous adaptation is essential. Organizations must adjust their structures as AI capabilities evolve, understand the variables at play, and connect the roles of judgement and prediction to overarching economic and ethical goals.

In the long term, AI solutions rely on specific domain knowledge and must be owned by subject-matter experts. Unlike other software, ML tools evolve and their results may change over time. In the short term, central initiatives are often required to bundle expertise, provide budgets, and drive the implementation process.

• Central Initiation: Strategic leaders define the long-term vision, allocate resources, and oversee initial development. Central initiatives provide resources, and governance.
• Experimentation: Individual use cases appear throughout the organization, as departments experiment. Central forces encourage this build-up of skill and experience, but maintain transparency.
• Standardized Structure and Control: With growing interest, Centers of Excellence consolidate expertise and implement standardized operating procedures. CoEs encourage risk-taking, provide funding, but also establish control to maintain efficiency through synergies.
• Integrating Decentralized Responsibility and Expertise: In the long-term, subject-matter experts should own their AI applications, requiring a hand-over of control to departments. Owners of datasets and tools are responsible for up-to-date structures and human expertise in the loop, ensures that judgment is not sacrificed for efficiency through prediction.