We Have Been Hearing The Wrong Half Of The Word

Artificial once meant made by human hands, and recovering that sense recovers something boards have quietly lost.

Sign up for my Substack daily AI newsletter here.

See my AI Training course portfolio for corporate Business Leaders here.

Article link: https://johanosteyn.substack.com/p/we-have-been-hearing-the-wrong-half?r=73gqa&utm_campaign=post&utm_medium=web&showWelcomeOnShare=true

Follow me on LinkedIn: https://www.linkedin.com/in/johanosteyn/

A sentence spoken at a South African university this year is more useful to company directors than most of what has been written about artificial intelligence governance. It is eight words long, it costs nothing to adopt, and it changes what a board can plausibly claim not to have decided.

CONTEXT AND BACKGROUND

Rhodes University has been running a public series called the Vice-Chancellor’s AI Conversations since March, deliberately convening law, psychology, anthropology and journalism rather than technologists alone. At the launch, senior anthropology lecturer Dr Dominique Santos told the audience that there is nothing artificial about AI and that it is utterly human. The second conversation, held in late July under the title Knowledge Inequality, Guardrails and Justice, turned that observation into a question, being that if AI is a human creation, then who gets to decide what it becomes.

The word she was recovering has always carried two meanings. One suggests something counterfeit, as in artificial flowers. The other, older and more accurate, means made by human artifice, constructed rather than naturally occurring. The term itself was a naming decision rather than a description of anything. John McCarthy organised the 1956 Dartmouth Summer Research Project on Artificial Intelligence, proposing that every aspect of learning or any other feature of intelligence could in principle be described so precisely that a machine could be made to simulate it.

INSIGHT AND ANALYSIS

Notice what the counterfeit reading does to a sentence. The model declined the application. The algorithm flagged the transaction. The system produced an inaccurate response. Each of these describes an event without an author, closer to weather than to a decision, and weather is nobody’s responsibility. Restore the constructed reading, and the same events acquire people. Somebody selected the training data. Somebody set the objective the system optimises towards. Somebody chose the confidence threshold at which a case is escalated to a human being. Somebody signed the deployment approval and was paid to do so. None of that is artificial in the sense of being unreal, and all of it is artificial in the sense of being made.

The strongest objection deserves an answer. These systems genuinely do produce behaviour that nobody specified, and engineers frequently cannot explain a particular output. That is true, and it is not an alibi. Unpredictability at the level of output does not dissolve accountability at the level of decision. An organisation that deploys a system it cannot fully explain into a lending process or a claims decision has made a choice about how much uncertainty it will accept on a customer’s behalf. That choice is entirely human; it happened in a meeting, and it is minuted somewhere.

I have previously written about this in an article asking who is really speaking when AI finishes your sentences, where I argued that AI-generated fluency can create the illusion of competence, that a report may sound strong while hiding weak reasoning, and that leaders should not mistake polish for judgement.

IMPLICATIONS

For a South African board, the test is practical and takes an afternoon. Take the AI risk register and rewrite every entry in the active voice with a named person attached. Not the model may produce biased outcomes, but we selected training data that we have not audited for bias, and the executive responsible is this one. Registers written the first way describe a hazard. Registers written the second way describe a duty, and only the second kind is governable. Any accountability framework assumes a decision-maker, and language that quietly removes one from the picture makes the framework unenforceable regardless of how carefully it was drafted. The same discipline applies to supplier conversations. When a vendor explains that a particular behaviour is inherent to the model, the follow-up question is who decided that this model was suitable for this purpose, and on what evidence.

CLOSING TAKEAWAY

Santos was not correcting the industry. She was recovering something the word had always contained and that ordinary use had worn away. That generosity is worth preserving, because an argument about terminology is easy to file as pedantry and an argument about accountability is not. The point is not that artificial intelligence is badly named. The point is that the name is accurate in a sense we have stopped hearing, and hearing it again puts people back into a picture from which they had gradually disappeared. Every system in your organisation was made. Somebody made it, somebody approved it, and somebody will have to answer for it.

Johan Steyn is a prominent AI thought leader, speaker, and author with a deep understanding of artificial intelligence’s impact on business and society. He is passionate about ethical AI development and its role in shaping a better future. Find out more about Johan’s work at https://www.aiforbusiness.net