Despite the impressive strides that AI has taken in recent years, there are still areas where its abilities pale in comparison to those of humans – a fact that could bring a sigh of relief to PR professionals.
Our benchmark for determining whether a computer can meet general intelligence has shifted over the years. First it was whether the top chess master could be beaten. Once that happened, we hurriedly decided it must be something else.
We turned back to Alan Turing, who had argued in 1950 that to have reached general intelligence, an AI must convince a human, through conversation, that it is speaking to a fellow human.
Most AI scientists now believe general intelligence is indicated by knowledge of communication. The frontline of modern AI is its ability to distil ideas and understand language.
In his 2021 book, The Myth of Artificial Intelligence: Why Computers Can’t Think the Way We Do, Erik Larson, a natural language research scientist, argues that plausible conversation is not enough.
The new AI systems that have surprisingly good conversations are trained in deductive and inductive reasoning, using established truths and the knowledge of thousands of previous happenings to interpret
the present.
Humans also use a third kind of reasoning called abductive inference. This is how we explain our surroundings by constantly generating and revising a hypothesis.
For example, if you find bird poo on your car, you will hypothesise that it came from a bird. Maybe a seagull. You don’t know that to be true. But a seagull is the most sensible explanation you can come up with.
Larson says that because of its rule-bound foundations, machine learning struggles with reasoning that reaches conclusions which are by necessity imperfect but correct enough to satisfy the user.
Luckily for marketers, PR practitioners, and journalists, this kind of inference is an essential part of the news and infotainment cycle.
Any current affairs issue can be interpreted in hundreds of different ways, if not thousands. As more information comes in, and the story progresses, each reader will reformulate their hypothesis to reach a unique perspective that suits them.
Those of us whose work is based on human behaviour do more than just use abductive reasoning; we predict the way others will use it.
When someone comes across a news story on social media or by word-of-mouth, there is a window to influence that person’s assessment of the issue.
News media, public relations, and marketing forecasts how people will react in that moment to the information they receive. The job is to determine which pieces of knowledge or evidence will help that person form the hypothesis we want of them.
People are always time-poor, so hypotheses are formed rapidly, often with random and even unconnected fragments of evidence.
To aid this endeavour, journalists deal in the headline; marketers deal in the six-second sell, and PR people craft the key message.
Because computers are trained to investigate everything that could help solve a problem, they find it exceptionally hard to know, as humans intuitively do, when you’ve learnt enough to stop reading.
AI reasoning doesn’t have the benefit of our lack of time. It can’t grasp how we use humour, cultural knowledge, and colloquialisms, to understand and communicate factually disconnected concepts.
The use of collective understandings to quickly pass on a message is why ‘just good enough’ reasoning is an incredibly human practice.
Essentially, we get the joke. Computers seek to explain why it’s funny. Newer AI models, like ChatGPT, have shown some ability to perform abductive reasoning, but understanding how a human does reasoning, sometimes with such flimsy material, is a chasm yet to be jumped.
If Larson is correct, we can rest easy at our desks. Useful AI journalism or advertising, and sound PR advice from a robot, are a fair way off.
This article was first published in our December/January 2024 issue.
