Winter is coming!
Time to get to work
Many observers are predicting that the AI bubble is about to burst, and they’re probably right. We’re in the middle of a “long summer” with peak hype and generous AI funding. But I think there’s a better metaphor to use instead of a “bubble”.
Winter is coming!
In Game of Thrones, Ned Stark famously warned “Winter is coming” to “express the need for preparation during the long summer and potentially hinting at the political turmoil he sees on the horizon” (according to Google Search’s AI summary).
Probably like you, I’ve been through a few tech seasons, and there are weather patterns to recognize and preparations to be made. But experience says the coming AI winter may not be entirely bad. Sometimes winter is a good time for reflection and preparation for next spring. What benefit can we get from this cold winter that’s coming? It’s clear that the season is changing, just look at the signs…
There are many clouds on the horizon
AI media hype has peaked
Compared to previous AI summers, this one is worryingly long (see below chart)
Companies are failing/pivoting/cutting back due to flawed ideas, failed promises or outright fraud (see the example of Builder.AI)
AI has over-promised and is not living up to its hype (Gary Marcus was right!) and we are no where near AGI
Revenue models are looking iffy and big corporations are pulling back on AI spending
Some companies say they are pausing hiring due to AI (I have a feeling this is BS and they are being cautious because of current economic uncertainties)
For a different metaphor that says it won’t all be bad, see the idea of a Wildfire and the terrible damage and good that it does. Having grown up hiking the chaparral covered hills of Southern California, I can attest to this. Fires are a necessary (if shocking) part of the Santa Monica Mountains ecosystem. Wildfire has to be reckoned with and AI may be at that point where very dry kindling covers the hills and breeds dangerous fires that then lead to the reshaping and rebirth of the landscape. AI needs this.
We can compare prior seasonal cycles
I asked Claude to make some charts comparing cycles of media hype vs funding for CD-ROM, Internet, and AI.
CD-ROM, WTF? I know, I know. It was a long time ago and a primitive technology. Yeah, but I’m old and I was there.
Below are screenshots of the Claude charts for CD-ROM and Internet linked to live web versions in the captions. While these are useful at a high level, they are crap from a stats point of view—the vertical X axis are unclear and “creatively” normalized.
What patterns do we see here? These tech cycles are similar - investment booms follow hype and then there’s a bust
Seasonal Patterns:
Hype correlates with funding
There are always winters
AI is bad at charts and I’m too lazy to make it to do better
What an AI winter means
How should we interpret this? Apply lessons learned from previous winters:
Each new technology often leads to some evolution that supersedes the “new one”
In CD-ROM, many involved learned about digital production which could then be applied to the Web and other fields.
Winter is sometimes a good thing. It can be a time for consolidation and rethinking. And it allows for better ethical, business, societal models in the next phase.
This is a good time to prepare for spring, whatever that means for your context
Pivoting you company to something that will grow when the time is right
Experimenting with new approaches to AI - Design, Technology, Business
Coming to a more clear understanding of what society wants from AI
Strategies to survive the winter
Maybe we need a pause to slow it all down. AI implementations are moving so fast given the potential negative cultural impacts. Perhaps this is a good time to create global AI safe-guards as Thomas Friedman suggests.
Time to do some woodshedding to rethink how we approach AI. It raises so many questions. What do we want AI for? Is that a good thing? How will we use it? What about the societal impacts? How do we design with AI as a material?
Let’s do it better this time around! We’re at one of those technological inflection points like the dawn of social media. Can we avoid the terrible outcomes?
New crops for Spring
This coming AI winter may be a great time to invent new approaches for AI that move beyond the Dumb-Smart character of current AI. LLMs seem to be peaking in their capabilities and we probably need something else.
Gary Marcus suggests that neuro-symbolic AI is the answer (by hybridizing LLMs with symbolic AI), but as much as I respect him, I’m not sure it goes far enough. Perhaps a more broad approach is a rich implementation of Howard Gardner’s theory of multiple intelligences.
“Godmother of AI” Fei Fei Li (inventor of ImageNet) has just launched WorldLabs.ai focused on Spatial Intelligence a key component in any successful real world application of AI.
Fight potential AI security vulnerabilities. A lot more research is needed to protect against vulnerabilities (especially with agents).
Agents that read the web as they do their job are subject to an old web SEO trick of putting blue text on a blue background — enabling a dangerous kind of prompt injection
Vibe coding can be influenced to incorporate malicious code from hiding in GitHub libraries
Prompt injection can target LLM’s long term memory, giving it instructions that will influence it’s behavior in the future
What useful application can we find for all those depreciating GPUs? For those who argue that this bubble burst will leave highly useful infrastructure (i.e. data centers) in place, I give you this quote:
My message is this: Slow down AI. We need to figure out what we want out of AI and what its impacts should and shouldn't be. This will take a time and experimentation.