Start Where You Are: AI Without a Technical Background

What if what holds you back from AI is your fear of the word 'technology', not a shortage of intelligence? Your phone suggests your next word, your maps app chooses your route, and your viewing platform recommends a film. You use machine-learning algorithms every day, even if you say you 'don't understand technology'. We have explained the basics in the articles 'AI, Really: What Is It, and What Is It Not?' and 'Large Language Models', so we shall not repeat them here.

What do the experiments say about the less experienced?

In a study of 5,179 customer-support agents, the agents were given an assistant based on a language model. The number of issues resolved per hour rose about 14% on average, and about 34% among the less experienced and lower-skilled, while the seasoned agents showed no notable effect (Brynjolfsson et al.). In another experiment with 444 college-educated professionals, the time needed for a writing task fell about 37% and its quality rose, and those who had scored lowest on the first task benefited most (Noy and Zhang).

Both results have limits. First, the 'novice' in the first study is someone with little experience in their job, not necessarily a stranger to technology. Second, the tasks are short and specific, and the researchers caution against generalising to jobs that need wider context. So they invite you to try, not to expect miracles.

The illusion of fluency

At Harvard, researchers (Deslauriers et al., 2019) compared, in an introductory physics course, an active lesson in which students take part with a smooth lecture. Students in the active lesson did better on tests by about 0.46 standard deviations, yet rated their own learning about 0.56 standard deviations lower. The researchers explained that mental effort is read as a sign of poor understanding, and smoothness as a sign of mastery.

The editorial reading is that this concerns you with smart tools: an orderly, convincing answer gives you the feeling that you understood, and perhaps you did not. What the model offers may be wrong with confidence, so checking the output is part of learning, not an addition to it.

And if the tool is wrong?

It will be wrong sometimes, and with confidence. That does not mean abandoning it, but treating it as a fast colleague whose word is not final. Ask it for the source of what it said, compare that with a reference you know, and if the matter is medical, religious or legal, ask a specialist. We have set out common misconceptions in the article 'AI Myths'. Whoever gets used to reviewing from the first day gains a skill that stays with them however the tools change.

Three steps

• Begin with a problem that matters to you. An employee who wants to summarise a report, a teacher who wants to draft test questions, a small shopkeeper who wants descriptions of products. A specific need is more useful than an abstract lesson.
• Make your request clear. 'Write me something about time management' gives you generalities. 'Explain to a tired employee who works nine hours three small habits to apply tomorrow morning' gives you something closer to a plan. In doing so you learn to think in order before you ask. (This is an editorial example, not from a study.)
• Review and write notes. What was right, and what was not? And avoid entering passwords, identity data and work secrets, as we explained in the article 'Privacy in the AI Era'.

Do not wait until you feel ready, for the background you are waiting for mostly forms through starting itself, and every week you postpone postpones what you would have learned in it. Open one tool today, choose a small task from your work, and note down what has changed after a week. Success comes from God, and taking the means is the servant's part.