As outside counsel to a client who is a Software-as-a-Service (SaaS) vendor that incorporates a form of Artificial Intelligence (AI) in its products, I recently encountered questions from a prospective customer to that SaaS vendor regarding my client’s use of AI. Based on the questions from this potential customer of my client, I realized that a lot of confusion and concern about AI in the technology vendor vetting context can arise because of confusion over the nature of AI.
So what exactly is AI? Defining AI can be confusing because AI encompasses several ideas. Parsing out these ideas can help compartmentalize assessments of risk.
1) Artificial Narrow Intelligence (ANI) AKA Weak AI (the “One trick ponies” of ai)
Artificial Narrow Intelligence (ANI) or Weak AI is AI designed to perform a specific task or a narrow range of tasks effectively, but it lacks the broad, general intelligence and adaptability of human intelligence. ANI excels in its specialized domain, often surpassing human performance in that specific area, but cannot apply its knowledge or skills to unrelated problems. Some examples of ANI are chatbots, image recognition software, recommendation systems, smart speakers (virtual assistants like Siri and Alexa), self-driving cars (autonomous vehicles), AI for web search, AI in medical diagnostics, language translation, and AI applications in farming or in a factory (industrial robotics).
2) Generative Artificial Intelligence (Generative AI)
Generative AI is a more general purpose form of AI that learns from existing data and uses that learning to create new content, such as new text, images or audio. Think of Generative AI as a machine-learning model that is trained to create new content, rather than one that makes predictions about a specific dataset. Generative AI can produce novel outputs in response to a prompt or query and can learn and improve over time by interacting with users or be retrained on new data. In terms of content generation, Generative AI can create a wide range of content, including text (e.g., stories, articles, scripts and code); images (e.g., photorealistic or abstract art); audio (e.g., music, voiceovers and other sound effects); and video (e.g., short films, animation and other visual content).
Some examples of generative AI include ChatGPT by OpenAI (an AI chatbot used for text generation, conversation and even code assistance); Gemini by Google (fka Bard - a conversational AI chatbot); Copilot by Microsoft (an AI virtual assistant, particularly useful for coding and within Microsoft 365 applications); Midjourney (an AI image generator tool); DALL-E (a text-to-image AI tool); Runway ML (a platform that allows users to generate videos, animations, and other visual content); and Claude by Anthropic (an AI assistant that combines text generation with detailed analysis, suitable for complex tasks).
3) Artificial General Intelligence (AGI)
Artificial General Intelligence (AGI) refers to a theoretical stage of AI where a machine can perform any intellectual task that a human being can. AGI is the benchmark of creating AI that can match or surpass human cognitive abilities across a broad range of tasks and domains. Needless to say, while we have not yet achieved AGI, the prospect of actually achieving AGI brings a multitude of concerns and risks.
Going back to my story, when I explained to my client’s potential customer the nature of the AI my client actually uses (i.e., Artificial Narrow Intelligence that relies on algorithms to make certain limited recommendations and predictions), we were able to allay their concerns. Having the vocabulary to distinguish these different forms of AI really helped in providing context and ultimately helped my client close the deal with a major new customer.