I can’t help wondering, if we are now standing at the crossroads of AI development with two possible alternatives before us – to model AI to mimic human behavior or to develop a better form of alternative intelligence, one, which will elevate human intuition.
Artificial Intelligence (AI) is the buzzword of our age, but occupies a very liminal space between what it can do, and what we want it to do, and, its uses as intelligent-machinery verses it’s ‘almost-human’ identity. But given the significance of AI in our world, shouldn’t we focus on a clear roadmap of intentions and expectations for the development of AI?
We all know the notion of AI is nothing new, but just how old it is can be surprising. The first recorded stories of Artificial Intelligence come from Greece around 600 BC and feature the bronze man, Talos. Although he couldn’t order a box of washing detergent as soon as you ran out, Talos was very good at fighting, and as is the case in many fictional AI narratives, he dreamed of becoming human too.
Needless to say, when humans imagine AI, we do so in our image. We expect to see a humanoid form, one that talks and responds in a human-like way. A natural extension of this is to imbue the machine with human emotions and aspirations too.
This anthropomorphic behavior on our part is a very human attribute. We do it to our pets, our stuffed toys, even our cars – instilling within them imagined human wants and needs. And while this makes us feel more connected to the things we own, in the case of AI development, we must ask ourselves if this is ideal.
Given the opportunity to create an alternate intelligence, surely it is better to plan one that elevates the human potential and supports human endeavors, rather than one which replaces or replicates human intelligence. After all, we already have plenty of humans.
I have recently been experimenting with my very own personal virtual assistant Andrew. Perhaps you know him. Andrew, from x.ai, is very helpful for setting up and canceling meetings and the like, but he’s not quite yet perfect. This is due to a couple of issues, not all on the AI side.
Firstly, Andrew uses Natural Language Processing (NLP) but one of the biggest hurdles in mastering NLP is learning how to dig the meaning out of the errors humans make when communicating. We, humans, are masters of miscommunication; our conversations are peppered with errors and missing words, which force the listener to guess at the meaning. This is real human intelligence in action.
Designing Andrew to make sense out of the unpredictable responses that come from each individual’s unique use of language is setting him up for failure. When he comes across language that is ambiguous, he drops the ball. This means he makes mistakes, misunderstands my meaning and intention, and occasionally tries to guess the right answer. In fact, in this regard, he’s almost human.
So, I recommend training him just as I would a new human PA. The makers recommend that we use Andrew 13 times so he can get up to speed, but just as with a human, this involves some degree of risk of error. However, during this process, I’ve recognized several things Andrew could do, much better than a human. For example, Andrew can apply cluster analysis methods used in unsupervised learning to find patterns in the way I use my calendar. With this, he makes suggestions on proposed meeting times. This is a great advantage over a human PA and it raises the question – rather than stating the fact that we want AI to reproduce human skills, shouldn’t we be looking for AI abilities, which support and complement the existing human ones?
I can’t help wondering, if we are now standing at the crossroads of AI development with two possible alternatives before us – to model AI to mimic human behavior or to develop a better form of alternative intelligence, one, which will elevate human intuition.