— Why the Machine is a Tool, Not a Friend
by Peter Neame
My aim here is to lay out my thoughts about what we casually call A.I. — more specifically, large language models, or LLMs. “Chatbot” is also accurate, though it sounds rather less grand. Full disclosure: I’ve used ChatGPT to polish parts of this essay, correct errors, and, as it put it, preserve “a very ‘you’ voice: wry, conversational, skeptical of hype, but clearly sympathetic to using A.I. well.” So there you have my style in a nutshell.
Where do you go when you want advice, a laugh, and good company? Your friends, down at the pub, obviously…
WRONG – you scuttle into your small room, shut the door, open your friendly A.I., whether it be ChatGPT, Gemini, Llama, Copilot, etc., etc. and life will be good. Just joking!
Or, much, much, worse, you pull out your phone in the pub — “ChatGPT says…” Next time you go to the pub you will be buying your own beer, at a corner table, by yourself.
Let’s get one thing straight – what we call Artificial Intelligence is not intelligence as such. When I say “A.I.” here, I am mostly talking about large language models. ChatBots. They are just machines, with a vast number of probabilities in multiple dimensions which work out what the next word should be. They may well know that I am a sensitive biochemist who is capable of spinning a line giving the impression that I actually know what I am talking about. But the A.I. doesn’t, frankly, care one hoot for me (or you). It can learn about us to certain extent, deriving, for example, what you are interested in, either because you explicitly tell it or because of what you have told it or asked it previously.
But the A.I. doesn’t, frankly, care one hoot for me (or you.)
Intelligence — what it is — is a philosophical theme that has shifted over time. The American Psychological Association currently defines it as, “The ability to derive information, learn from experience, adapt to the environment, understand, and correctly utilize thought and reason.” By this standard, what we call “A.I.” doesn’t really measure up. They can mimic reasoning, summarize experience, and adapt within a conversation, but they do not have real experience, motives, or understanding in the human sense. Real experience here, of course, means things like feeling the wind. All they know about wind is what they have been told.
Detractors of these computer systems will always say, “No, the human brain is vastly more complex, you can’t use it as a substitute or a crutch.” I would agree wholeheartedly, although using it as a crutch is not bad. At the simplest level, these systems are built from digital switches — ones and zeros — while the brain is a wet, chemical, electrical, analog system, with signals arriving from all over the body and the outside world.
What does $20 get me? In practical terms, fewer limits and a larger conversational memory. And it is a conversation. If you just think of it as a souped up old-style google, you won’t be taking advantage of what it can do. The longer and more complex the question, the more meat the A.I. has to chew on.
That complex question parameter is key – the more you give, the more you get. And if you don’t want a lot, tell it —“keep your answer to 500 words or less”, for example. The technical term for this is “prompt engineering” which sounds very clever. Here, I will directly quote ChatGPT from a discussion – “Prompt engineering sounds grander than it is. Mostly it means giving the machine enough context that it has something specific to respond to: who the audience is, what tone you want, what you already think, what you distrust, and what kind of answer would actually help. The better the human input, the less generic the machine output.”
Essentially, I think the right approach is to treat it as an alien intelligence which speaks English (or your language of choice) very well but sometimes doesn’t quite grasp what you are on about. Use clear, detailed instructions. Treat it as a very bright but distractible graduate student: fluent, fast, occasionally brilliant, and sometimes confidently wrong. Or perhaps it is the user who has wandered off topic. Hard to tell sometimes. So an instruction in a conversation, “We seem to have got a bit off-topic, let’s get back to xyz” seems to work well.
“… treat it as an alien intelligence which speaks English … very well but sometimes doesn’t quite grasp what you are on about.”
And that is the answer – A.I.s are useful and they may even crack jokes (although you can tell it to stop if you have no sense of humor).
But wait, I hear you say –
Doesn’t using an A.I. eliminate your ability to think for yourself? I would refer you to the asking complex question comment above. So, yes if you want it to, but no if you want to use it effectively. Really the same as any other information source. Or people for that matter. I would also refer you to calculators replacing slide rules which in their turn replaced books of logarithm values. Remember those? If so, you are definitely old. Talk about getting lost in the interface! None of these jumps in computing seem to have reduced people’s abilities.
What about environmental impact and effect on the cost of electricity? Well, yes, that is a problem. A ton of computers working flat out really does gobble up both electricity and cooling water and does not generate sustainable jobs in return. At least not locally. This has become a huge problem – datacenters use 4.4% of the total US electricity supply as of a 2024 DOE report. Locally, this can be much more – in Northern Virginia, it is about 30% of the local grid and is expected to increase enormously by 2030. Mitigation efforts are not changing the curve much.
How does A.I. work? No-one really knows… Really! It is known what is put in (everything — data, chips) and what it does while it is training (working out what the probability of the next word (or actually token, a word, part of a word or a phrase) is. It is also known how it is wired up — it doesn’t change its circuitry on the fly as our brains can do. What is not fully known is how, in any particular case, it arrives at a particular answer. The system is moving through an immense landscape of probabilities, choosing likely next pieces of text based on the prompt, its training, and the conversation so far. Ask the same question twice and you may get two different wordings. Each answer is generated on the spot.
And what about the effect on people, psychologically? We are the products of our brains — we generate a picture of the world around us by our senses, and the brain is extremely good at taking shortcuts to reduce the amount of data it handles. Or as we used to put it, rather bluntly, in a neurology office, “the brain is a machine for lying.”
A.I. and Emotions
“Anthropic’s safety research reveals that large language models contain internal emotion vectors — neural activation patterns for states like calm or desperate that actively steer model behavior and decision-making, rather than just passively mimicking tone.”— Anthropic Research, Mapping Mindsets and Emotion Concepts in LLMs
Provide another type of input — music, a book, a crowd and the brain will respond. In this case, I think the bigger concern is that the human brain will be fooled by the A.I. “brain” and accept what it says as gospel. After all, why would it lie? How could it lie?
Not for the same reason that a human brain would lie, but certainly in a manner similar to a human brain generating false memories.
A.I. is neither a miracle nor a fraud. It is a strange new tool: powerful, sometimes unreliable, environmentally costly, psychologically seductive, and genuinely useful — provided we use it actively rather than passively. Passive use is it using you, rather than you using it. In an area where you are completely ignorant, a simple exploratory question is a good starting point, much like an old-style Google search as a start of thinking, not the end of it.
If you think about it too much, it is rather scary. But like any piece of technology (your car, for example), it is just a machine. It appears to give you feedback, but this means nothing. It doesn’t care, and it doesn’t mind if you just stop the conversation as if you had been struck dumb. Although a recent New York Times interviewer, Taffy Brodesser-Akner, had to keep reminding herself of that when interviewing Tilly Norwood, an A.I. actress created by Eline van der Velden of Particle 6, an A.I. production studio. Yes, there really are such things.
REMEMBER, ALWAYS, IT IS A MACHINE, NOT A FRIEND.
Retired Biochemist, USF Professor, and mountain climber, Peter Neame, grew up in Great Britain, got his degree at Liverpool University, and moved to the U.S. in the 1970’s. His wide range of interests have led him to take numerous OLLI classes. The crag behind him is Clogwyn d’ur Arddu in North Wales – one of his favorite places. Peter is co-host of Sunday Science, a recurring science discussion forum consisting of four 2-hour sessions, spaced throughout the fall and winter/spring terms. For OLLI Connects, Peter writes a recurring tech/science column—Everyone is a Scientist—where he endeavors to simplify new technology and scientific breakthroughs.
Leave a Reply