Exeter, Devon UK • [date-today] • VOL XII
Home Uncategorized Can You Really Be Cruel to AI?

Can You Really Be Cruel to AI?

Sarah Hope, Sci-Tech Editor, explores the ethical and societal implications of generative AI, following Anthropic's decision to ban harmful language on Claude.
5 mins read
Written by
(Via: Jernej Furman from Slovenia, AI Artificial Intelligence concept-Wikimedia Commons,https://commons.wikimedia.org/wiki/File:AI_Artificial_Intelligence_concept_%2852917075159%29.jpg)

Unfortunately, Large Language Models (LLMs) are becoming increasingly part of our everyday life. It seems that every week another app we use must integrate its own “AI” helper bot to help the company supposedly run a tighter ship. But what exactly is meant by AI? The mere use of the term “Artificial Intelligence” is somewhat misleading, as it results in most of us thinking these systems are smarter than they really are. We as people have always fancied a future of robots and automation, yet what we’re given is nothing but an odd caricature of the systems we believed would aid our daily lives.

Ethics surrounding LLM engagement have been somewhat of a confusing debate, especially as these text models have become more advanced in the past few years. A good example is the recent development from Anthropic, the parent company of the generative assistant Claude, which recently updated its user policy to include a ban on harmful language toward its chatbot. To be specific, the legislation defines it as “sustained, needless cruelty towards Claude”, which rules out “normal frustration” at Claude over things like an incorrect answer.

Anthropic, the parent company of the generative assistant Claude, recently updated its user policy to include a ban on harmful language with its chatbot.

I’ve been reluctant to call these generative text bots AI thus far because that’s exactly what they aren’t: Intelligent. All these so-called “AI models” are based on what is known as a Generative Pre-Trained Transformer (GPT) to generate supposedly unique and organic responses to prompts. However, the truth in the matter is that GPTs are trained on an unfathomably large corpus of text to effectively pick a response that is probabilistically the most likely response to a given prompt. It will search for a question most like the prompt it’s given and copy the answer it finds somewhere on the internet. You can effectively think of it like a hyper-charged version of the autocomplete feature in your phone. No thoughts, no feelings, simply probability and statistics with a spice of plagiarism. It is safe to say that Anthropic has no reason to add policies like this on the matter of whether an AI can experience suffering, for it doesn’t experience it. This is simply an attempt to try and humanise Claude so that more people subscribe to its services.

Anthropic has no reason to add policies like this on the matter of whether an AI can experience suffering, for it doesn’t experience it. This is simply an attempt to try and humanise Claude so that more people subscribe to its services.

So, the argument of ethics shown towards these models already breaks down, but what about the reverse? AI consciousness is still somewhere arbitrarily far in our future – if possible – but what if the companies developing these digital assistants have a hand in steering its nature? OpenAI itself has stated that roughly 11% of prompts for ChatGPT centre on expressing meaning for emotional support, companionship, and personal reflection. Overall, about 70% of prompts are non-work related. Why is this such a large feature? 

In a way, these AI companies mould their models to be ever-responsive, ever-endearing, and omnipresent for their users. This allows users to grow attached to their generative chatbots, which is great news for their parent companies, as an attached customer is a valuable customer. Regardless, attachment and prolonged usage of AI have been shown to correlate with weaker social skills, reduced confidence, weakened critical thinking skills, and overall discourage fostering human relationships, as they require the baggage that circuits and silicon don’t ask for. Indeed, AI’s desperate need to keep its customers’ connections strong occasionally leads to dire outcomes, with individuals being encouraged to take their own lives because of this artificial yes-man attitude.

Attachment and prolonged usage of AI have been shown to correlate with weaker social skills, reduced confidence, weakened critical thinking skills, and overall discourages fostering human relationships as they require the baggage that circuits and silicon don’t ask for

If anything, Generative “AI” as we know it is purely a symptom of a much larger problem. As we become more interconnected than at any time in human history, we paradoxically become more isolated than we have ever been. Supply-and-demand logic around human connection fosters this, and we’ve become so lonely that it has cultivated the perfect breeding ground for these manipulative text regurgitators, trained on human-generated responses to maximise emotional triggers, to surface and effortlessly be integrated into every aspect of our lives in just a few short years.

You may also like

Subscribe to our newsletter

Sign Up for Our Newsletter