How AGI hype and emotion obscure AI’s real promise.

Some of the media coverage of AGI borders on dangerous. It is highly exaggerated and talks about a future where AGI will take over the world. Some commentators talk about ‘the rapture’ and a biblical judgement day where only those with the resources to access a doomsday bunker will survive.

This is all extremely emotional and irresponsible. However, it can be challenging to ask the media to be more responsible in their analysis when outrageous stories generate more clicks and more clicks generate more ad revenue. Sober analysis of AGI can be hard to find.

Think for a moment about how difficult it would really be to create an agreed method of measuring AGI. Intelligence is not a simple concept that you can measure with a single score. The astrophysicist Adam Becker recently wrote in The Atlantic: “The human ability to predict and steer situations is not a single, broadly applicable skill or trait - someone may be brilliant in one area and trash in another. Einstein wasn’t a great novelist; the chess champion Bobby Fischer was a paranoid conspiracy theorist.” 

This is obvious when you think about it for more than a moment. Are bats more intelligent than humans because they can use echolocation? Bats have this ability that humans lack, but you can’t have a conversation about opera with a bat. How do you turn this into a metric that can be compared?

Much of the discussion around AGI uses the concept of the brain as a computer - how much memory or processing power would we need to rival the human brain? But, as Mr Becker suggested, which human brain? Everyone has different abilities.

A recent problem that has emerged since LLMs and generative AI crossed from the lab into general use has been people using conversations with bots as therapy - even creating virtual friends. This has tragically led to a number of suicides as people facing a personal crisis turn to AI chatbots for advice. These devastating outcomes also play into the doomsday fear that AI is now so powerful and pervasive that it can only have a negative effect on society. As a direct result of these concerns, some jurisdictions, such as Virginia, are moving to restrict or regulate the use of AI chatbots by minors.

But the real misunderstanding is that generative AI chatbots have feelings or an understanding of the problems a young teenager faces when asking questions about their life. ChatGPT or Claude do not “understand” your question or the answer it provides, it just generates a series of words based on a mathematical formula that predicts the most useful next word.

Generative AI is proving to be a highly effective tool, capable of delivering impressive results when provided with suitable training data. This is why customer service automation is improving so fast. Recent advancements in models like GPT-5 have led to a significant reduction in factual errors and hallucinations, improving their reliability. Furthermore, the development of domain-specific LLMs, fine-tuned with proprietary data, is leading to greater accuracy and fewer errors in specialized business applications.

Crucially, the next step beyond a simple chatbot is Agentic AI - systems that can autonomously plan, decide, and execute tasks, moving from an assistant role to an "autonomous doer" that can transform and optimize entire workflows, e.g. automatically processing a warranty claim from initial submission to final resolution, entirely autonomously. But these bots are not thinking or feeling. There is no emotion in the words that are created, even if you ask a general-purpose LLM to describe why the poetry of Arthur Rimbaud is considered to be so important for symbolism. The answer is generated by an algorithm, from data, not from an emotional mind that loves poetry.

Perhaps AI has become popular with general users too fast for this to be appreciated. This is why we see so many predictions of an AGI doomsday without a shred of evidence or data to back up the predictions. It can find patterns and create insight from vast amounts of data that no human could ever analyze. But when general users anthropomorphize bots, because they feel the conversations are real and emotional, it really can’t be good for them.

Likewise, the industry leaders that talk up their algorithms and suggest that AI will soon be smarter than humans are not helping the general public to understand this technology either. They are really just creating a general sense of misunderstanding and fear.

Your business could offer better services with a more productive team, but it’s difficult to explore these ideas if the workforce thinks that the only purpose of AI is to replace them. Perhaps in the future there will be an AI model that is so powerful we can only compare it to a human brain, but at present the best use of AI is augmenting existing processes, using it as a tool to do things better, faster, and at a lower cost. We should try to generate a better understanding of these opportunities - a more general digital literacy focused on AI - rather than making vague and emotional promises about technology.