Posted on 26th January 2024

What's next for AI

Where have we come from?

One of the things that struck me when I was taking artificial intelligence and machine learning classes in the late 2000s was that much of what we were learning was based on research from decades before. In fact some of it, such as our understanding of propositional logic, dates back not just centuries, but millenia...

The latest developments in AI have their underpinnings in the research of the 20th century. LLMs in particular, used by so-called generative AI, rely on neural networks. There are so many questions related to generative AI that have been covered extensively by the media in the last year or two. So I don't intend to cover the same ground here.

What is striking to me is the degree to which the world has discovered one particular development in AI and elevated it in people's perception to the position of something approaching general intelligence. As I've discussed before, this has led to some compelling applications of this tech. But general intelligence it is not.

Where are we going?

The current AI trend is a bubble, unlike anything I've seen in my career and it will burst. That is not to say when the dust settles it won't have had a significant impact. One thing I hope that will remain is a curiousity about what AI can be and what it can do. This goes much broader than a chat-bot that seems to give human-like responses and has a large body of information in its knowledge-bank. But for some time yet, there will be no "one-size-fits-all" solution that much of the world was hoping for.

There are so many branches and aspects of AI that are left unexplored, or are still the preserve of a few academics and entreupreneurs who are quietly advancing their chosen frontier of the field one small step at a time.

Generative AI became the zeitgeist through a combination of the new LLMs' affinity with creative applications and a latent desire to commodify creativity. Oddly on its own this simulated creativity doesn't have any particularly compelling commercial applications, especially when you consider the enourmous monetary and environmental cost of supporting the models required. Looked at through a philosophical lens it also does nothing to enrich human experience. Yet somehow it has captured the imagination of people beyond the technology industry.

What comes next?

As stated already, soon the current bubble will burst, by which point society's wider under understanding of what the current trend means will have evolved. There will definitely still be some growth in AI related business linked to current developments. But I think there will also be some interest in casting a few lines out into the less explored areas of AI.

It's not hard to imagine, given the hype created this time round, that people will want to recreate this boom by "finding the next LLM": Essentially looking for the new big thing waiting to explode. And I don't doubt an AI boom will happen again, but it could be decades before it does, just like before.

In the meantime it's also entirely possible many businesses will start a drive to try and use AI to cut costs. This will probably play out like outsourcing trends of the past, which in some cases provided some reductions in overheads but in many cases resulted in similar or higher costs than before due to the necessary quality control.

At the moment I think any CEO seriously considering implementing LLM based AI at scale in order to reduce their headcount is utterly bonkers. There is a difference between investing in R&D to test AI solutions in an isolated setting, and a scatter-gun approach of just asking technical staff to 'make it happen' at scale (and start laying-off employees) before you've proven the business case for it. And in some cases LLMs just aren't the appropriate form of AI to make the necessary 'automation' happen.

Don't get me wrong, I'm sure some people will continue to find access to this latest form of AI useful and keep discovering specific tasks it can help them with. But the true value derived from it will not match up to the current ambition to use AI everywhere that this phenomena has created.

My hopes for the future of AI

All this doesn't mean renewed interest in AI, in all its forms, isn't a good thing. The problem is how to retain the enthusiasm and investment for innovation when it becomes clear the business model has dried up. After all, many of the genuine business opportunities discovered from riding this current wave of speculation have already been proven. Tech companies have already planted their flags in the sand and a lot of startups in this field have already been acquired.

Rather than try to replicate success in a tight market by copying other businesses I do hope that there will remain some interest in going in different directions. In academia of course this is hopefully a given - novelty is one of the things that gives rise to exciting discoveries. For businesses I'd like to see more emphasis on solving problems and proving business models. It all goes back to my previous discussion about design: By considering what is desirable, viable and feasible and applying a bit of human creativity something capable of a sustainable impact can be made.

The future of AI hinges on people building things we actually need and making them work well and efficiently. I very much hope the next thing we hang the AI label on is not just another LLM.