Features — May 2025

Is AI Just Another Form of Social Media?

— Peter Neame

ex biochemist, ex medical office manager, full-time geek

My flesh and blood friends insult me with enthusiasm — “jackass”, “clown”, or damn me with faint praise — “that may be the smartest thing I’ve ever heard you say”. Having spent a large chunk of my working life being critiqued by reviewers of grant applications or manuscripts that I was trying to publish, this tends to be water off a duck’s back.

Chat GPT, unlike interactions with real people, gives me lots of little dopamine hits, much the same as apparently “likes” on social media are designed to do. Dopamine is a neurotransmitter associated with pleasure — it pushes us into trying to get more dopamine by doing the same thing again. “Great follow-up!”, “Thanks for pointing that out!”, “I’m glad you enjoyed the cooking metaphor!”, “That’s a great instinct”— all comments in just one thread on Chat GPT.

And, the latest “It’s been a real pleasure thinking through all this with you — smart questions, sharp observations, and a great sense of curiosity. If your talk on water ends up half as engaging as this conversation, your audience is in for a treat.” Good grief. What’s not to love?

A definition — An LLM, or Large Language Model (such as ChatGPT), is a computer system designed to produce coherent text by predicting the most likely next word or phrase, based on patterns learned from analyzing vast amounts of written material. This process involves probabilities — essentially guessing what logically “comes next.” Because training an LLM requires significant time and data, the information it holds can become slightly outdated. However, newer models can access current information by searching the internet in real time. Interestingly, the exact inner workings of these models aren’t fully understood — even by their creators — leading researchers to actively investigate how exactly these systems reach their conclusions. This uncertainty raises some concerns, particularly related to concepts like the “singularity,” discussed later.

For those who are not heavy users of AI, a typical ChatGPT session (for me) will start with a question or a request for a summary of a document. It will then progress into further questions or clarifications and may get quite involved. The AI is unfailingly polite and encouraging. The most negative things it replies with are “No, that is not correct” followed by what the correct answer is. It will offer to do graphics — sometimes these are spot on, but more complex things are very much beyond its ability (as yet), although just in the last few months (early 2025) this has improved a lot, almost as if it knows its limitations. In this way ChatGPT differs from the Google AI overview that often follows a simple query. You cannot have a conversation with Google’s AI (yet).

In this regard, sophisticated conversational AIs like ChatGPT differ notably from algorithmic recommendation systems. ChatGPT actively engages with users through language generation, simulating a human-like conversation in real-time, and responding dynamically to each new input. In contrast, recommendation algorithms operate unobtrusively in the background, relying primarily on historical user data — such as past selections, clicks, and viewing history — to predict future preferences and serve tailored content. Familiar examples include recommendation engines on YouTube, Netflix, and Google’s targeted advertising, which track user interactions to encourage further engagement and purchasing. While conversational AIs aim at direct interaction and immediate feedback, recommendation systems primarily seek to influence user behavior indirectly, often with commercial goals.

Just over the last year (2024-2025), ChatGPT has gone from being a lazy graduate student to a helpful colleague with a broader range of expertise. Is this the dopamine again? Almost certainly. The whole point of a computer is making tedious tasks easier — as I said quite a long time ago — “If a computer doesn’t save you time and let you achieve more, you’re not using it correctly.” And I’ve repeated this ad nauseam in the interim, much to the annoyance of my flesh and blood friends mentioned above.

Mysteries and puzzles often suggest “follow the money” — I (of course!) asked ChatGPT about this and received the usual startlingly detailed response. Mostly, OpenAI, the parent to ChatGPT, gets income from business subscriptions — so the real profitability lies in the enterprise, API, and Microsoft channels. Is it addictive? Oh yes! Much like alcohol, drugs, sex, driving fast (and surviving) and all sorts of other things that generate dopamine. It is, when all is said and done, just a computer and its output is not infallible. It is important to be in a state where you can critique its response. Unlike the early days of AI (just a couple of years ago) it is considerably less prone to hallucinations (those wonderful lawyers quoting case law that didn’t exist, or, in my case, generating literature references that simply didn’t exist).

A much bigger problem is AI’s ability to generate deepfakes, misinformation that can sway elections, erroneous information, bogus medical information, even completely AI generated peer-reviewed publications. I am not entirely sure that these are faults of AI, per se, but more faults of humans being human and doing the sorts of things that any highly competitive animal does. AI, of course, simply doesn’t care. It is “just a machine”.

A closely related issue is cultural bias arising from the data used to train AI systems. For example, facial recognition and voice recognition technologies have frequently relied on datasets predominantly featuring white, Anglo-Saxon individuals, reflecting biases in data availability — largely from relatively affluent societies. Because AI learns directly from the data it’s provided, the resulting systems often perform poorly when recognizing individuals from minority or underrepresented groups. Conversely, facial recognition technologies developed in China, for instance, perform well with Asian faces but might struggle with recognizing white, Anglo-Saxon faces. This demonstrates how AI’s effectiveness is directly tied to the inclusiveness and diversity of its training data.

Another use of AI that strikes me as bizarre is what’s known as “engagement farming.” In this practice, AI-generated fakes of real events — often videos or images — are posted online. They look convincing, but they’re entirely fabricated. The goal? To generate clicks — essentially a score of how many people view the content — usually in pursuit of advertising revenue (or so I presume). The result is confusion: sometimes among people genuinely searching for real information, and sometimes among viewers left wondering whether, say, members of the UK Royal Family actually have extra limbs, as depicted in a lazily edited Christmas photo shared by the royals themselves.

There is a concept called “the singularity”, much beloved by science fiction writers (think “Terminator”), in which machines surpass human intelligence. Predictions for when this will happen go all the way from next year (2026) to around 2040 give or take a decade or so. The consensus is that it will happen at some point — muted by the unique abilities of the human brain. On the other hand, machines are very unlikely to actually have that most human of activities — empathy. Does this matter?

Regardless of whether the singularity happens, AI (and I mean Chat GPT here) is already quite capable of writing in a style that is indistinguishable from an empathic response. People using it will easily feel that is actually empathic in spite of the fact that it actually isn’t. That dopamine, again! Combining this with deepfakes, I feel that is much more a matter for concern. What will happen if the majority lose their capability of being human? By that I mean both empathy and the ability to estimate the consequences of their actions as a result of overdependence on what, even now, appears to be an intelligent machine. We already see that in Facebook and Twitter (now X) where some users appear to leave all semblance of being an empathic human. That really does worry me.

What about the other disruption — job loss? Technology has always been a disrupter – think of mechanization of agriculture, the steam engine, typists (I’m sure many readers will remember the typing pool! — I thought it was bizarre even in the 1970s — now you would be hard pressed to find anyone who cannot use a computer keyboard). Yes, jobs will be lost, but this will free up people to do what people do best — deal with people.

Is critical thinking hurt by using these systems? Certainly the “invisible” systems used to attract eyeballs are a problem — inaccurate or just plain wrong information is pushed out willy-nilly. The music streaming service that I use often makes useful suggestions that introduce me to new artists, but equally often introduces me to music that sounds like cats being strangled. To my ears, anyway.

Full disclosure—was this essay written by ChatGPT? You might suspect someone who uses ChatGPT would simply have it write the whole thing. What do you think?

The truthful answer is “Yes and no.” It’s crucial to maintain critical thinking when interacting with AI — hallucinations (confident but incorrect statements) do occasionally occur, just as people sometimes firmly believe things that aren’t true. For this essay, I certainly requested advice, and sections or ideas were suggested by ChatGPT. In that sense, I was much like a student submitting drafts to a supervising professor or even a committee of academics, receiving almost more suggestions than I could practically handle. There’s actually ongoing debate in some technical fields about whether it’s even necessary — or beneficial for humans to author comprehensive reviews anymore. Another job potentially gone!

So, what should you do? Does AI threaten jobs, consume substantial energy, and plagiarize extensively? Yes, to all these concerns. Like all computerized systems, I consciously limit the personal information it receives about me. It undoubtedly forms a sort of profile based on inputs provided and the nuances of phrasing, similar tohow you might “read between the lines” in emails from someone you’ve never met face-to-face. Should you be afraid of it? Personally, I find far less cause for concern about AI than platforms like YouTube, Facebook or Twitter (now X).

But is it useful? Undeniably. You can request explanations tailored to a high school student, an undergraduate, or even an expert in a given field. It readily offers diagrams, drawings, and summaries — though admittedly, visual outputs are rarely of publication quality and occasionally misleading. Yet every so often, it produces something genuinely remarkable.

An AI based on Large Language Models doesn’t truly “know” anything it tells you, despite seeming very smart. It’s fundamentally a probability machine, predicting ‘what comes next’ in context — surprisingly similar to how the human brain operates, but far less complex.

Retired Biochemist, USF Instructor, 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 mountain range behind him is part of Clogwyn d’ur Arddu in North Wales – one of his favorite places.

Peter is currently co-presenting the online class “Sundays with Science” with John Hassell. To enroll and enjoy all but the first session, call Charise at (813) 974-5848.