The future of AI: super deep learning or real world reasoning?
Artificial Intelligence and Machine Learning are not new fields; they’ve been around longer than most people realize. Right now, though, they’re at a hugely critical juncture, with the next decade possibly being one of the most significant in their history.
From watching the media over the last few years, you’d be forgiven for thinking that AI was now a solved problem, as there have been many stories of impressive AI systems outperforming humans. If that’s true, though, where are all the robots? Where is the super-intelligent AI that will solve all the world’s problems?
The more boring truth is that AI has solved some very narrow problems in very narrow ways. These are extremely impressive feats of engineering but, for the most part, are not intelligent and rarely survive first contact with the infinite complexity of the real world. This has led to a lackluster adoption of AI across industry, with these clever but fragile AI solutions deployed only in narrowly defined low-risk use cases.
This commercial reality contrasts starkly with the high-profile picture painted in the popular media, which seems to show impressive feats of intelligence, from apparently sentient AI language models to gallery-worthy generative artworks. However, these remarkable feats are just the result of a decade of rapid progress using artificial neural networks. In recent years, we have mastered the art of creating huge machine learning models which can so wholly and profoundly identify the patterns present in specific datasets that they can predict their outputs or manipulate the relationships in the data at will to produce endless creative variants from it.
This realization that scale alone can produce intelligent-looking results has been one of the driving factors in AI over the last decade. This philosophy is behind one of the two major factions emerging within AI, where some people believe that true intelligence will emerge if only we could make the models big enough. Others believe these large models are nothing more than glorified curve fitting – a statistical mimicry of intelligence that can’t extrapolate beyond its trained experience. This second faction believes that true intelligence can only come from systems that explicitly reason about real-world concepts (so-called “symbol manipulation”), developing a true general intellect rather than just copying patterns of behavior from other intelligent beings.
The debate rages on between the two camps, but the reality is that both are probably right to some extent, and elements of both approaches will be part of an ultimate, hybrid solution. The massive deep learning models and their remarkable abilities to comprehend and manipulate complex data will likely work in tandem with higher-level symbolic reasoning to produce some genuinely astonishing leaps forward. External pressures are also driving the field of AI in a different direction, from the very big to the very small. The environmental cost of modern AI is hard to comprehend. The energy requirements to train just one of these large models equates to a carbon footprint equivalent to the lifetime emissions of five family cars, forcing the field to question its obsession with the ever-increasing scale. At the same time, consumer trends are driving demand for AI deployments on smaller edge devices, both in a quest for independence from cloud connectivity and privacy fears.
One thing is clear, though – the improvements in AI’s abilities are on a swift exponential curve, accelerating at a pace that is surprising even experts in the field. AI is simultaneously over-hyped and under-hyped, with some short-term successes blown out of proportion, while at the same time, many cannot comprehend the magnitude of change coming in the medium term. There is every reason to believe that AI will be genuinely transformative for humanity, society, and the planet, although the path will not be easy or risk-free.
Whatever happens, though, it is clear it will come surprisingly fast. Like all exponential curves, the rapid growth phase has a habit of creeping up and then suddenly overtaking. For those who see it coming, there will be massive opportunities.