“Questioning the Existence of Ethical AI: Do Morals and Machine Learning Interface Yet?”
“I’m Not Convinced Ethical Generative AI Currently Exists”
“OpenAI, a research organization committed to ensuring that artificial general intelligence (AGI) benefits all of humanity, has a knack for making headlines.” – WIRED
Could it be said with any more grandeur? An entity devoted to the lofty ideal of guaranteeing AGI which will somehow be a universal boon. Every time OpenAI rears its head in the news, it’s an invitation to explore what’s really occurring in the bowels of generative Artificial Intelligence (AI).
Here’s a straight talk – the promise of an ethically sound generative AI doesn’t exist. *Shock and horror*. Yes, you heard it here first. Or possibly for the umpteenth time, for the folks who’ve made it their business to keep an eagle eye on AI. There is no flawless process when it comes to establishing universally-accepted ethical guidelines in AI.
Consider for a moment, GPT-3, the company’s impressively slick language model. It’s surely a marvel of modern technology, sure to have notable impacts. Yet, can it grace the front cover of “AI Ethics Monthly”? Probably not. Despite the best endeavors of OpenAI, biases have slipped in, showing that even the smartest machines are still learning.
Every atom-sized judgment an AI system makes is informed by the information that has been fed into it. Throw in the minuscule matter of the whole of the internet – that paragon of fair-mindedness and balance – and there’s no wonder that biases come into the play. The inherent ‘goodness’ of AI is about as realistic as expecting a computer to write the next Pulitzer Prize-winning novel.
Reviewing AI-generated content and waiting for a magic censorship button is about as logical as waiting for the coffee machine to start spewing out champagne. The reality is, anything that has the audacity to claim ethical superiority in AI is likely to be wishful thinking.
These are machines. They’re capable of many things, but emulating human values, judgment, and discretion isn’t coded into their DNA. Nor should it be. By their very definition, machines should be impartial, free of predispositions, but unfortunately, they are only as unbiased as the programmers who code them.
“AI’s natural state is untruthful”, one might boldly proclaim. And it would be true. The Turing Test was never about ethical intelligence, after all. It is about the ability to generate competent lies, creating convincing fabrications of human thought.
OpenAI’s ‘Paused’ project? An inevitable result. A testament to the reality that AI, for all its great potential, is still a field in its infancy. A field fraught with hitches, requiring greater thought, further regulation. A field where the promise of ethical generative AI is still very much a pipedream.
Remember, it’s all just smoke and mirrors until an AI can independently pass the ‘ethics test’. And we aren’t quite there yet. We probably have better chances finding a unicorn at the end of a rainbow. Until then, don’t take everything AI says at face value. Like a broken grandfather clock, it’s only ‘right’ twice a day.