When Not to Use Generative AI

Generative AI (GenAI) adoption has exploded over the past year, and it has rapidly become one of the most deployed AI techniques across business units and organizations, according to Gartner surveys. There’s a good reason: GenAI offers the promise of both everyday and game-changing business improvements. But GenAI is not a silver bullet.

• Start by figuring out whether the use case is value-driving for the business and feasible to execute on, regardless of the AI technique. This is important because some use cases are not a good fit for AI and do not merit further consideration.

• Map your use case against the relevant use case family. GenAI is:

  • Highly useful: Content generation, conversational user interfaces, knowledge discovery

  • Somewhat useful: Segmentation/classification, recommendation systems, perception, intelligent automation, anomaly detection/monitoring

  • Hardly useful: Prediction/forecasting, planning, decision intelligence, autonomous systems

• GenAI may also be a poor fit for your use case if the risks that come with it are unacceptable and cannot be effectively mitigated. These include unreliable outputs, data privacy, intellectual property, liability, cybersecurity and regulatory compliance, either alone or in combination with one another.

• For areas where GenAI does not rank as “highly useful,” consider other AI techniques. 

• Common established AI techniques to investigate include nongenerative machine learning (ML), optimization, simulation, rules/heuristics and knowledge graphs. Emerging techniques, such as causal AI, neuro-symbolic AI and first-principles AI, are also worth tracking.

• Trying a simpler alternative AI technique before diving into generative AI can be a smart move; they are often less risky, less expensive and easier to understand.

• AI techniques are not mutually exclusive; they can often be combined in a way that delivers better accuracy, transparency and performance, while also reducing costs and need for data.

• The combination of GenAI models with other AI techniques can be particularly powerful.

• The potential combinations of AI techniques are endless. Strong combinations and use cases include:

  • Nongenerative ML and GenAI models for segmentation and classification, synthetic data generation and computer vision

  • Optimization/search and GenAI models for enterprise search

  • Simulation and GenAI models for simulation acceleration

  • Graphs and GenAI models for knowledge management and retrieval-augmented generation

  • Rule-based systems and GenAI models for chatbots, robo-advisors and specialized natural language generation