The AI winter is coming



Artificial intelligence is cyclic when it comes to hype and advancement. One time stretch sees AI reach dizzying levels of media attention and industry funding. The next sees the other side of the AI hype cycle, known as the AI winter.

These are the times where ‘artificial intelligence’ wanes in both favour and furtherance. During an AI winter, the technology becomes little more than a dirty word, synonymous with false promises. It steps out of the spotlight, away from the disillusioned eyes of the public.

‘AI winter’ is the term that denotes the lowest points in AI. They’re periods of reduced interest, inhibited advancement and low funding. And we might be heading for another one.

The AI winter is coming.


The initial AI cycle

AI technology has roots that go almost as far back as the modern computer. The field of artificial intelligence research was officially created in 1956. It didn’t take long to gain traction.

Indeed, the idea of AI taking over jobs in the near-future saw widespread marketing as early as the 1960s. Promises soon emerged of AI that could win chess games and translate written messages into any language. These premature claims preceded the first AI winter.

There are differing opinions about the exact timing and number of AI winters that the tech industry has seen. However, the most common opinion is that there have been two winter spells in artificial intelligence research.


AI winters past

The first AI winter started in the early 70s. The Advanced Research Projects Agency (now DARPA) pulled their funding of AI research. Instead, they allocated funds to projects and research that promised identifiable, reachable goals.

The UK Lighthill report of around the same time damaged the image of AI further. It reported on the dubious real-world value that AI held under the hype. So, the 70s was a recession for AI following its initial hype.

The AI cycle of interest revolved again in the 80s, when a smaller boom hit with the rise of ‘expert systems’. (Artificial intelligence that would focus on one key task.) Once again, expectations around the field grew inflated. When these false promises went unmet, the AI winter returned in the 90s. It would stay until the next AI hype cycle started, with the AI obsession we’re seeing today. 

In both instances, people realised AI promises were mostly hot air. Real-world AI — clouded by too much hype — caused disappointment and disillusion when reality hit. It had failed, it was a fraud, and it wasn’t worth any more time or money.


The current AI summer

Today we’re in a high point of a new AI hype cycle, which started around 2010. Excitement around deep learning and machine learning fuels current AI popularity. And it’s all thanks to data allowing AI to become viable again. 

Chatbots have also helped, with natural language processing technology making a comeback. Suddenly, the promises of AI past seemed to be coming true — or at least getting much closer.

Now, we’re back to worrying about AI taking jobs. We’re reading about AI that’ll match and surpass human intelligence and ability. We will all have AI personal assistants and self-driving cars. We’re going to live in a post-work society, thanks to new AI.

But whispers are starting to spread. We’re once again questioning the promises and representations of machine learning and AI capability. It seems we’ve reached the inn at the crossroads. This is where we stop and decide our next direction — and it’s a decision that could bring about the next AI winter.


Falling into old patterns?

So, are we heading for another AI winter? The winter days of the past were a product of hype and unmet expectations. And the thing is, we seem to be falling into the same old pattern of puffery, false promises and stretched truths.

40% of EU AI companies don’t use AI. Many that do use it in their background processes, not on the customer frontline. We often confuse simple automation with modern artificial intelligence. AI has become little more than a buzzword, with too broad a meaning to be truly useful. It can’t yet maintain context in long conversations. It can’t match our empathy, flexibility or understanding.

And yet, we’re focusing so much on what AI will do, we aren’t recognising what it can do currently. As a result, we’re setting AI up to fail, putting it on a throne of exaggeration and misconception.


Facing the stark reality of AI

It’s time to face the stark reality of AI. If we continue as we are, the AI winter is coming. If we distinguish the hype from the reality, maybe we can hold off the winter.

So, how can we avoid another long AI winter? We need to stop making false promises and unreachable expectations about AI. Instead, it’s time we recognise not only the promise of AI, but its limitations. We need to get excited about what AI can already do, not what it will do one day.

With data to feed it, AI can offer insight unattainable to us before. This lets us make more informed decisions and drive our businesses. Elsewhere, AI is starting to help our chatbots get better at understanding and serving us. Voice recognition AI is beginning to improve accessibility and human-machine interaction.

Artificial intelligence doesn’t need false promises to be exciting — it’s already doing some impressive things.


Is an AI winter coming?

If we continue to fall into the hype-fuelled patterns of the past, we risk becoming disillusioned with deep learning, disappointed in the AI we have, and cynical of the successes on the horizon.

We have seen advancements in AI, and that’s what we need to focus on. It might not be hype-fuel, but it is exciting.