“Artificial intelligence drift” is a relatively obscure term, and you won’t see it mentioned a whole lot in online tech literature. However, it is something that engineers and others are thinking about as they contemplate the evolution of artificial intelligence toward stronger and more comprehensive results.
Artificial intelligence drift happens when sophisticated AI entities, whether they are chatbots, robots or digital constructions created to pass the Turing test, start to diverge from the directives and instructions of their original programming toward types of responses and activities that may not have been contemplated by their human handlers.
You can see examples of this drift in recent projects, for example, where two Facebook chatbots famously began to communicate with each other in what IT professionals described as “secret code.” Essentially, the technologies evolved to the point that they decided to utilize a different means of communication, one that was not suggested or requested by the human programmers.
The factors involved in artificial intelligence drift are the factors that have led to the evolution of strong artificial intelligence paradigms in the last few decades. One is more loosely coupled machine learning algorithms that are highly interpretive, and give these technologies leeway to grow and evolve. Machine learning fundamentally changes how computing systems work – rather than simply focusing on quantifiable data and rigid computing tasks, as traditional technologies did, artificial intelligence is moving toward self-correcting and self-evolving tools reflected in machine learning and deep learning strategies, and toward the idea of a neural network that much more capably simulates human thought and intelligence.
Another factor in AI drift and the evolution of artificial intelligence is multi-part technologies that work on a collaborative basis, again, to simulate more sophisticated kind of intelligence. Some IT professionals refer to these as “deep stubborn networks,” or technologies including both a generative and a discriminative component. As these and other individual entities in multi-entity paradigms work with each other, they evolve what the technology can do and move it toward a freer result that is less constrained by its original programming. That’s the idea behind artificial intelligence advances, and it’s the concept behind this artificial intelligence drift – that computing systems may change or alter themselves after their original program execution, simply because, due to these progressive factors, they can.
