🌱 What AI Can Do (and What It Can’t)

Overview

AI can do the heavy lifting for some of the repetitive tasks that we currently do. But it can't think. Whilst AI is developing rapidly, at the time of writing:

• Generative AI creates content, or processes large volumes of data, by utilising the patterns it was exposed through during its training. It operates in response to natural language prompts (see: 🌿 AI Prompt Engineering Guide).
  • It does not understand the content it is generating - it simply mimics patterns. Likewise, it does not reason or plan, instead generating output based on probability or likelihood.
  • It has no goal or intention - it simply responds to prompts.
  • It is particularly prone to hallucination - producing plausible but false information. (see: 🌱 Preventing AI Hallucinations)
• Agentic AI can pursue goals and take action, and can make decisions across multiple steps. It can break a goal into subtasks and work through them, and it can monitor its own progress and adapts its behaviour accordingly.
  • It does not have self-awareness or general intelligence, and it still requires human oversight for goals, values and decisions.
  • Without constraints, it can become inefficient, and it may struggle in unstructured or ambiguous environments.

AI can not replicate human thinking. And without the right kinds of human thinking, there is a risk that using AI will result in unintended consequences. And this is where the shift of human skills comes in: see The decline of linear (or analytical) thinking and Thinking models that still matter