I discovered this week that British workers are spending an estimated £958 million of their own money every year on artificial intelligence to help them do their jobs, which is not how I remember employment working when I first encountered it. In those days you went somewhere, they gave you work, you complained about it quietly, and at the end of the week they gave you money. We appear to have modernised the arrangement. Now they give you the work, and you secretly pay AI to do it. You are essentially the middleman. This is known as “shadow AI”, which improves the whole thing enormously because it makes an office manager sound as though they are running an intelligence operation out of the stationery cupboard.
Somewhere in Scotland there is undoubtedly a woman who has worked for the same council since 1997, owns three cardigans in varying shades of blue and has secretly become an advanced technology operation. Management thinks she has simply become better at correspondence. In reality, Barbara in Payroll has gone underground.
I first realised artificial intelligence had infiltrated my life when Effie declined tea. Effie has been declining tea successfully for years without technical assistance. She has declined because the washing was out, because the washing was in, because a parcel was coming, because she had put something in the oven, because the weather looked wrong and, on one memorable afternoon, because she had taken her boots off and felt the day’s events had therefore reached their natural conclusion. Then one day I received four paragraphs thanking me warmly for including her, regretting that prior commitments prevented her attendance and expressing the hope that we might reconnect at another mutually convenient time. It finished by wishing me a productive and fulfilling afternoon.
That was the giveaway. Effie knows perfectly well I have never had a productive and fulfilling afternoon. I have had afternoons in which things have happened, but that is not the same thing. One particularly successful afternoon resulted in a clean freezer. Another ended with all the towels in the same cupboard. These are not achievements on which one builds a reputation. I later discovered that Effie had asked artificial intelligence to decline the invitation because she simply did not fancy coming out, and I admired this enormously. Technology has spent decades promising to transform the human condition and has finally found its proper purpose: helping a Scottish woman avoid telling her friend she would rather stay home in her slippers.
The unsettling part was how much the AI had improved her. The real Effie is warm, funny and entirely capable of ending a message with two kisses despite having forgotten to include a verb. AI Effie sounded as though she had recently completed a course in stakeholder relations. Once I noticed this, I began looking more closely at other messages and realised that artificial intelligence is already everywhere. People who have known me for twenty years are suddenly “reaching out”. Now everybody appreciates my patience, values my understanding, hopes I am well, looks forward to reconnecting and wishes to ensure alignment. I do not wish to be aligned. At my age, if both knees are pointing in approximately the same direction I consider the day a success.
The great attraction of artificial intelligence seems to be that it can translate what we actually think into something fit to send to another human being. Most genuine thoughts are remarkably efficient: I sent you this last week. No. You have the wrong person. I already paid that. Please stop emailing me. The AI takes these sturdy little sentences and puts cushions around them until you are no longer annoyed that somebody has ignored your previous three messages; you are grateful for the opportunity to clarify. This may be a greater threat to Scottish life than anyone realises because we are a nation built upon carefully controlled irritation. Remove that and nobody knows where they stand.
Elspeth has apparently been using artificial intelligence for difficult emails for months, which I should have suspected because her messages have recently acquired bullet points. Elspeth is not naturally a bullet-point woman. She once wrote a shopping list in a circle around the edges of an envelope because there was a telephone number in the middle she did not want to disturb. Yet suddenly she is producing concise summaries, proposed next steps and courteous follow-ups. I had assumed retirement had given her time for personal development. It turns out she has ChatGPT.
Margaret has not yet entrusted anything important to artificial intelligence, but she has tested it, which in Margaret’s case means asking it questions to see when it will make a fool of itself. She began with recipes. The machine offered substitutions, suggested improvements and at one point proposed altering something she has made in exactly the same way for more than thirty years. Margaret regarded this as an act of considerable nerve. There are areas of life where new information may be welcome, but advising a woman to rethink her traybake after three decades of successful church tables is not innovation. It is provocation.
Agnes has used artificial intelligence to complain about a vacuum cleaner, which I found particularly interesting because the vacuum cleaner had only one fault. It did not vacuum. There are products whose failures require explanation. A boiler can make noises. A car can develop a mysterious lurching. A television can have sound but no picture, picture but no sound, or both. A vacuum cleaner has a simpler agreement with humanity: there are crumbs, and it removes them. Agnes’s did not. Artificial intelligence nevertheless transformed this straightforward commercial dispute into a gracious statement of disappointment and a desire to explore an appropriate resolution. The company sent instructions explaining how to clean the filter. Agnes wanted her £189 back. AI had turned a refund request into couples counselling.
There is a lesson in this because sometimes anger is efficient. We have spent years teaching ourselves that every communication must be calm, measured, reasonable and constructive, but there are moments when the most useful thing a person can say is “see you next Tuesday.” Artificial intelligence may eventually make us so polite that nothing is ever resolved because nobody can tell whether anyone is actually dissatisfied.
Jean immediately saw the potential for dealing with family group messages. Every family has one. The message begins innocently with somebody asking what time lunch is, and within minutes there are seventeen notifications. Somebody wants to know whether two o’clock means arriving at two or eating at two. Somebody else wants to know if there will be potatoes. An aunt who has not yet said whether she is coming announces that she does not like parsnips. A cousin asks whose house it is despite the address appearing six messages above. Then somebody sends a photograph of a dog. If artificial intelligence can take over this work, the country may recover several points of GDP before Christmas.
Morag remains less impressed. Living on an island part of the year gives a person a healthy relationship with technological certainty. She has seen websites confidently describe services as operating normally while standing on an island looking at the stretch of water where the ferry ought to be. I trust Morag’s judgement on these matters. Anything that claims to remove uncertainty has never had to get somewhere from an island in February.
Rhona, on the other hand, is the reason I discovered that artificial intelligence can now attend meetings. She had come across one of the systems that listens to the entire thing, makes notes, produces a summary and identifies action points. This is regularly presented as useful. I regard it as an attack upon civilisation. Meetings work because the minutes bear only a distant relationship to human events, and this is one of the foundations of organised society.
A forty-minute disagreement becomes “discussion followed”. Somebody announcing that the whole proposal is nonsense becomes “concerns were noted”. Three people talking over one another becomes “a range of views was expressed”. A committee member threatening to resign because nobody has collected the raffle prizes from her garage becomes “storage arrangements remain under review”. This is not dishonesty. It is mercy. Rhona could see the efficiency of having the notes produced automatically. I could see only the danger of creating a permanent record of things everybody present had already agreed, silently and wisely, to forget.
Unfortunately I then made the mistake of reading what the experts think comes next. There are serious people considering a future in which artificial intelligence is woven into almost everything around us. It may help monitor our health, organise our work, influence our decisions and become such a constant assistant to human thinking that we are no longer the only intelligence in the room. By 2040, some expect AI to sit behind so many ordinary objects and decisions that much of its work may happen without us consciously noticing. Our lives could become more programmed. It might know how we sleep, what we eat, how we work and possibly who we ought to date. I had only just become accustomed to the kettle knowing when it was boiling. Now apparently it’s little blue light may eventually have views.
This concerns me because several electrical items in my house already behave as though they possess independent judgement. The printer has had strong opinions for years. The television occasionally announces there is no signal while visibly showing me a programme. The washing machine once locked my towels inside for forty-five minutes and refused all negotiation. None of these things possesses intelligence at present, and I do not see why we would reward that behaviour by giving them any.
Some experts also imagine AI becoming something like a constant partner in human thought, helping to check, expand and organise ideas. This has been described in impressive terms involving collective intelligence and collaboration between people and machines. I know what it will be called in my house. Interfering.
I do not want a collective mind while I am making soup. I barely want another person in the kitchen. There are certain decisions that should remain between a woman and whatever she has found in the vegetable drawer.
Margaret agrees with me on this point, although for different reasons. Her concern is that AI will begin believing its advice is superior. She has already seen this with satnavs, which issue instructions in the tone of someone who has never encountered a tractor on a single-track road. She sees no reason to expand the principle into the rest of domestic life. A refrigerator that knows what is inside it sounds useful until it begins commenting on the unhealthy desserts.
The more alarming debate concerns artificial superintelligence and whether systems far beyond human ability could eventually become difficult or impossible to control. Some scientists and technology specialists regard this as a serious risk; others argue that present systems are nowhere near that point and that more immediate problems such as jobs, misinformation and environmental costs deserve the greater attention. This is one of those subjects where I find myself looking for the middle option between “everything will be fine” and “human civilisation may end”.
Unfortunately there does not appear to be a checkbox for that.
What fascinates me is the difference between the conversation at the top and what ordinary people are doing at the bottom. Scientists are discussing whether artificial intelligence could transform human consciousness, while Effie has discovered that she can photograph the contents of her refrigerator and ask a machine what she can make for dinner, then reject every suggestion until it eventually arrives at toast.
Researchers are considering whether machines could one day become more intelligent than humanity, while Elspeth has spent twenty minutes asking one to identify a mystery key she found in the kitchen drawer, despite there being no conceivable way for the computer to know what lock it belongs to.
Margaret has finally asked it what the “Eco” button on her dishwasher does after fourteen years of refusing to press it, only to decide that a four-hour cycle is ridiculous and continue not pressing it.
Jean has discovered that artificial intelligence can plan routes and is now attempting to develop one through the supermarket that avoids everyone she knows while still passing the reduced shelf.
Rhona photographed a recent grocery receipt and asked the machine to explain how she had spent £47.62 while apparently buying nothing, which may be the first genuinely impossible problem placed before artificial intelligence.
This is, I suspect, how civilisation will actually respond to revolutionary intelligence: give humanity access to the accumulated computational power of the age and within a fortnight somebody will be asking the most advanced intelligence humanity has ever created if it can distinguish beef stew from something frozen during lockdown.
The danger of the meeting assistant became personal when I realised one of these systems could theoretically attend tea. Our gatherings have operated successfully for years because nothing is documented. We have no agenda, no minutes and no defined purpose. There is occasionally a three-tier cake stand when somebody has ambitions. Matters develop naturally. One minute we are discussing the price of butter and the next we are examining the moral character of a woman in Inverness. Nobody knows how we got there. This is the beauty of it.
Artificial intelligence would immediately attempt to impose order. It would identify themes, and recent gatherings would probably have been classified under household maintenance, local government, deteriorating joints, supermarket pricing, ferry reliability and people who park badly. Rhona pointed out that a proper system could also identify who had agreed to do what, which only confirmed my worst fears. Our entire arrangement depends upon no one being able to prove that anybody agreed to anything.
Jean has been waiting since spring for Margaret to return a plastic container. At present this is simply part of the natural migration of Tupperware through Scottish society. Margaret maintains, quite reasonably, that several containers are currently in circulation and nobody has established beyond doubt which one is Jean’s. An artificial intelligence system would probably search the minutes, identify ownership, assign a deadline and create a follow-up reminder. A friendship that has survived years could be destroyed by database accuracy.
Effie has announced several times that she is reducing her sugar intake, and no human present has been unkind enough to correlate this statement with the shortbread. A machine might. Margaret occasionally undertakes to bring something the following week and then quite reasonably decides that nobody can remember what it was. Rhona’s meeting assistant could remember. This is precisely why it must never be invited.
Then the programme would produce a summary, and nobody wants a summary of tea. Tea should disappear behind you like weather. I do not wish to receive an automated email at six o’clock informing me that participants discussed wheelie bins, butter prices, or seasonal ailments. Nor do I want a list headed “Next Steps” informing me that Jean will bring milk and I have responsibility for biscuit provision. There must be parts of human life where nothing is captured.
This, I suspect, is where some of the more serious warnings about artificial intelligence become easier to understand. The concern is not simply that machines become clever. It is how much judgement, attention and responsibility human beings may quietly hand over because AI is convenient. I have already surrendered multiplication to the calculator. I no longer know anybody’s telephone number. I once drove twelve miles in the wrong direction because a pleasant woman inside the dashboard told me to. Convenience has an excellent record against human resolve.
If a machine eventually knows my diary, my shopping, my health, my route home, my preferred biscuits and which messages I cannot be bothered answering myself, the great technological takeover may not require force. That is why the £958 million suddenly looks rather different. Perhaps artificial intelligence is not coming to steal our jobs. Perhaps we are paying its bus fare.
I have decided not to worry unnecessarily because every generation is told that technology will save time, and it never does. The washing machine saved hours of washing, so we bought more clothes. The dishwasher saved washing-up, so we acquired seventeen mugs per person. Email made correspondence instantaneous, so we started sending messages all day. Mobile phones meant nobody ever had to wait beside a telephone again, so now everybody is expected to answer the telephone everywhere. Artificial intelligence will save us from writing emails, and I have no doubt we will use the spare time to create more very human problems.
Effie has stopped using AI to decline invitations from me. Yesterday her message arrived shortly before lunch. It contained no greeting, no appreciation, no regret, no bullet points and no desire to reconnect at a mutually convenient time. There was no punctuation. She had the hairdresser. She could not make tea. There were two kisses.
It was unmistakably Effie, and if humanity is going to survive artificial intelligence, I think bad grammar may be where we begin.
