Imagine telling a robot how to find the tallest student in a line. 'Look and pick the tallest' does not work, because the robot does not know how to look and pick. Instead you say: point at the first student and remember their height. Move to the next student. If they are taller than the height you remember, remember their height instead. Repeat until there are no students left. Then say the height you remember. Every step is exact, every step is something the robot can do, and the steps stop at the end of the line. That is an algorithm.
Try it on heights of 130, 142, 128, 150, and 139 cm. Remember 130. 142 is taller, so remember 142. 128 is not taller. 150 is taller, so remember 150. 139 is not taller. The line has ended, so the answer is 150 cm. The very same steps work for a line of 5 students or 500.
Not every list of steps is an algorithm. A cookie recipe that says 'add sugar to taste' or 'bake until golden' leaves the choice to the cook, so two people get different cookies. A step like 'stir forever' never ends. An algorithm replaces 'to taste' with '100 grams' and 'until golden' with '12 minutes', or with a test that anyone can check the same way.
Think of it like this. An algorithm is like a recipe written for a robot cook: every amount is measured, every step is spelled out, and the last step says stop. A person can guess what 'a pinch' means; the robot cannot, so nothing is left to guess. Unlike most recipes, a good algorithm must also work for every input it is allowed to get, not just one.
Core idea. An algorithm must be precise, finite, and guaranteed to produce a result for valid inputs.