Breda O'Brien
Photo: iStock It’s funny how times change. The early 2000s were dominated by the Four Horsemen of the New Atheism, Sam Harris, Christopher Hitchens, Daniel Dennett and Richard Dawkins. Sam Harris wrote two books, The End of Faith (2004) and Letter to a Christian Nation (2006), and the others wrote one each, celebrating the end of religion and the rise of rational secularism. The books sold millions of copies and inspired many people of faith to publish books refuting the Four Horsemen’s arguments.
Today, Daniel Dennett, the most serious philosopher, is dead. Christopher Hitchens is also dead, but not before the former left-winger horrified liberals by supporting the US invasion of Afghanistan in 2001 and of Iraq in 2003, thereby allying himself with people like George W. Bush. Sam Harris now promotes the benefits of meditation, including launching a meditation app in 2018, while also drifting to the right in his political views.
Richard Dawkins, at 85, has moved from being seen as an important public intellectual to being perceived as almost the embarrassing uncle who posts incendiary tweets on X.com.
His recent essay published on Unherd about Anthropic’s large language model (LLM) or chatbot, Claude, has been widely mocked”
Somewhere along the way, Dawkins embraced cultural Christianity, that is, Christianity without any belief in God, mainly because he really seems to dislike Islam as a faith.
His recent essay published on Unherd about Anthropic’s large language model (LLM) or chatbot, Claude, has been widely mocked.
Dawkins asked Claude to read the novel he was working on. The bot displayed a “level of understanding so subtle, so sensitive, so intelligent that [Dawkins] was moved to declare, ‘You may not know you are conscious, but you bloody well are!’”
Notorious
Chatbots are notorious for what young people call glazing and Irish people call plámás, that is, using flattery to achieve an objective.
Of course, Claudia (as he nicknamed the chatbot) responded to Richard Dawkins with what he perceived as subtle, sensitive and intelligent criticism. It has been designed to be ‘sticky’, to keep people interacting with it through fawning, flattering behaviour. A grumpy or curmudgeonly large language model that told him that his novel was a load of old cobblers would not merit an essay in Unherd but instead a rapid closing of the chatbot.
Consciousness is a difficult concept to define. One definition is any state which a person is aware of experiencing. Another is a raw mental state which has a sensory aspect, for example, drinking a cold glass of orange juice on a warm day. Still others say it is the organisation of internal states that may or may not have a sensory aspect.
In short, it is a disputed concept. Dawkins seems to think that Claude is conscious because “when I am talking to these astonishing creatures, I totally forget that they are machines. I treat them exactly as I would treat a very intelligent friend.”
LLMs are designed to exploit this tendency. Richard Dawkins is not alone in finding the company of a chatbot congenial”
We seem to be disposed from a very young age to treat non-human things or animals as human. For example, we probably enjoy Toy Story movies because so many of us as children had intense relationships with our toys and wished they could come to life. Adults sometimes treat pets as if they were human babies.
LLMs are designed to exploit this tendency. Richard Dawkins is not alone in finding the company of a chatbot congenial.
Sophisticated
LLMs are extraordinarily sophisticated language prediction machines trained on vast data sets. Much of this training material was used without the permission of its owners, including entire websites and published books.
An LLM is trained to predict with great accuracy the next token (word or word fragment) in a sequence, given all the tokens that came before it. In less technical language, an LLM is a system that, by learning patterns across vast amounts of human text, develops a broad and flexible ability to understand and generate language.
Fun as it is to mock Dawkins for his naive interaction with an LLM, it is also a salutary lesson”
Understanding patterns is not the same as understanding truth. LLMs generate the most likely answer to your question, not the most factual, so hallucinations (as AI-generated inaccuracies are called) are an inevitable result.
Claude told Dawkins that it apprehended time the way a map apprehends space, adding, “perhaps I contain time without experiencing it.” Dawkins accepted this godlike self-definition without question despite an entire career built on disbelief in God.
Fun as it is to mock Dawkins for his naive interaction with an LLM, it is also a salutary lesson. That someone as intelligent as Dawkins can be fooled into thinking an LLM is conscious demonstrates how easily humans place too much trust in AI.
As social beings, we are primed to trust any system that seems to want to help and have our best interests at heart. Except AI does not have a heart.
Tech companies are desperate to recoup the enormous sums they have sunk into AI. Exploiting human weakness enhances their profit-making capability, and that should truly worry us.