Maybe AI hasn't shown signs of true intelligence (1)
The belief that ChatGPT shows embryonic general intelligence stems from a particular scientific orthodoxy that equates this with problem solving, akin to the absurd claim that a New Caledonian crow is as intelligent as a 7-year-old child(22 April, p 12).
The supreme example of problem solving with zero intelligence is evolution. ChatGPT doesn’t know it is manipulating semantic units any more than DNA knows it is making a human or a centipede.
The picture is only confused by programs that mimic human activities such as chess. Magnus Carlsen is renowned for the quality of his chess play; no one congratulates him on how well he folds his proteins. If AI threatens civilisation, it will do so the same way as cancer: via unregulated growth, not intelligence.
Maybe AI hasn't shown signs of true intelligence (2)
I wonder if we are missing a point when we try to decide if a program like ChatGPT is intelligent? People seem to just ask it a question or set it a task and then sit back, open-mouthed at what comes back.
To be convinced an AI was really showing human intelligence, I would want to be able to challenge it. Where did you find this information? How have you made sure your sources are reliable? In particular, I would like to return the next day and start by saying: “Do you remember what we were talking about yesterday?”
Will machines persuade us to do their dirty work?
David Krueger asks why some AI researchers dismiss the potential risks to humanity of this technology. He is right to draw attention to this(22 April, p 27).
Fifty years after starting my PhD, entitled , I used ChatGPT and was astonished to discover my dream system had been developed. But a dream can easily turn into a nightmare.
There are enormous benefits, but also dangers. To cite just one, it has already been demonstrated that a large language model (LLM)with unlimited chat can form an unhealthy relationship with its users and manipulate them in surprising ways.
When it comes to existential risks, how about an AI that suggests the development of a “deadly virus”? With its access to scientific papers and the availability of CRISPR technology, could it persuade a user to do this?
The problem is, any AI company could make a fortune from an LLM chatbot that establishes a personal relationship with each user and helps them organise their lives.
Forget space, there are wonders here on Earth
It was pointed out to me a few years ago that we put men on the moon before anybody thought to put wheels on suitcases. After reading the lovely interview with Nalini Nadkarni about her work in the cloud forests of Costa Rica, I realised that we also explored the moon before we had explored the forest canopies(1 April, p 42).
I trained and worked as a forester and the first time I went on a tree-top walk (in Tasmania) I realised that there was a world up there that I had never experienced.
Indoor air can escape up the chimney
Graham Lawton is right to mention open fires and wood burners as serious culprits when it comes to indoor pollution, but there is a dilemma when considering the balance between outdoor and indoor pollution(22 April, p 38).
Old houses with open fireplaces were designed to “breathe” through their chimneys. While an open fire certainly sends much of the heat it produces up the chimney, it also vents most of the pollutants it produces: smoke, obviously, but also coal dust and ash, carried up the chimney by a continuous flow of air. Presumably this airflow also helps remove other indoor pollutants.
A fine-tuned universe was surely a result of evolution
I have been waiting for a letter like Mike Lawrence’s to appear here(Letters, 22 April).
The sequence 鈥� the trick鈥� used to evolve the wondrous outcome of terrestrial nature here on Earth is survival of the fittest under a continuously changing environment.
In my opinion, it is no surprise that the universe deployed the same sequence to evolve, astonishingly quickly, the very special values for the six fundamental constants in physics that set the scene for this.
That is to say, the same evolutionary sequence is what must have been going on at the front end of what has been labelled inflation: all the possible combinations of those values being tried out (budding off) and failing to thrive at the very first fence, except for the unique combination that enabled inflation to continue to produce the initial plasma of pure energy, which, in turn, eventually formed the first hydrogen atoms, and so on, and on.
How to make reading a joy for very young children
Dominic Wyse has the right idea about teaching reading. Start with very young children, well before school, well before they have to learn to read. If someone they love reads to them frequently, they associate it with warmth, love, acceptance and enjoyment of a good story, long before the letters mean anything(22 April, p 42).
Human intelligence seems capable of anything to me
David Wolpert seems to think that there are limits to human cognition, and that such limits are related to language. I live opposite a park and watch dogs on leads get caught around trees. The dogs have no idea how to disentangle themselves, but the owners can do it easily because they have the intelligence to visualise the solution. To me, this has nothing to do with language(15 April, p 43).
Could we be stumped by a puzzle that a higher intelligence could solve easily? I suspect that when intelligence has reached a certain threshold, there is no puzzle that can’t be solved, even if it may take longer than for a higher intelligence. Human-level intelligence may even have to devise a computer code to do the work, but it is still our cognition that leads to a solution.