Technological innovation tends to be so forward-thinking that it could be argued that many companies place the cart before the horse — but that’s really the idea, isn’t it? The hype around artificial intelligence (AI) can feel the same, and it’s resulting in the sector being reported as a bubble, like crypto currency. However, AI is an underpinning technology that’s already delivering value throughout daily life for most of the world’s population, from banking to aviation to Alexa (yes, that Alexa, created by Unlikely AI’s founder).
Excitement about technology often peaks long before it delivers results. Looking back at the horse and cart as an example, cars didn’t immediately replace horses as people’s preferred mode of personal transport when they first emerged. In fact, one of the first mass-produced cars, the Ford Model T, didn’t enter production until 1908, 23 years after Karl Benz took to the road in the first motor car, the ‘Benz-Patent-Motorwagen’ in 1885 (but Ford did go on to sell over 15 million Model T’s by 1927).
Navigating the hype is a challenge, and over-valued companies that won’t reach the revenues investors hope for are certainly popping up. We’ve been investing in AI for decades and will continue to do so for years to come. Read on to find out what we look for.
Open Source AI
While companies like Meta are taking an open source approach to its AI development, it’s not the best approach for those that want to protect their intellectual property (IP). The term ‘open source’ usually relates to the development of software, where its source code is made freely available to be distributed and modified, and it’s quite similar when it comes to AI.
The Open Source Initiative (read: the self-appointed body of what it means to be open source) has a clear view on this. They define an open source AI system as one that “can be used for any purpose without the need to secure permission, and researchers should be able to inspect its components and study how the system works”. This should also include access to the training data upon which an AI model has been developed, to be truly open source. If researchers can inspect components and study how the system works, then they will be able to replicate it.
That’s problematic if you’re developing an AI model built upon data that’s freely available, allowing anyone access to your IP and compete with your business.
Don’t build on others’ foundations
Rather than relying on open source AI data, some of the most innovative companies are creating models that learn from customer data.
An example of this is Secondmind, the company helping automotive engineers design better cars, faster. They’ve created the world’s first decision-making platform using Gaussian processes, a probabilistic AI technique that can generate accurate predictions from low volumes of historical data. It could prove essential for automakers, given the government mandated switch to electric vehicles in many countries, in the coming years.
Secondmind has a multi-year partnership with Japanese automaker Mazda, and was recently featured in Bloomberg as one of 25 top UK startups to watch in 2025.
Share the learnings, not the data
It’s not all about Large Language Models (LLMs), in fact that’s only one type of foundational model, the origins of which can be traced back to the 1960s. The hot air filling the supposed AI bubble has largely built up from the hype created by the sudden availability of this technology to the general public. Identifying the innovation amongst the plethora of companies popping up in the sector requires a keen eye, expertise, and experience.
Before the general hype erupted in 2022 (thanks to the release of ChatGPT), we’d been investing in AI for over 20 years. In 2020, we invested in Altana, a company which specialises in federated machine learning to help governments and enterprises gain visibility across their supply chain networks. The platform uses this form of machine learning (ML) to share intelligence about supply chain networks for all entities using it, without sharing any of the underlying data.
In the last four years, Altana has gained some of the world’s most important public and private sector organisations as customers, culminating in the company gaining a valuation of over $1 billion in 2024.
Artificial General Intelligence
Looking to the future, there’s chatter about Artificial General Intelligence (AGI), which was covered at The Times Tech Summit 2024. AGI, theoretically, would understand and learn any intellectual task that a human being can, but experts like Professor Neil Lawrence were quick to critique this at the summit.
Achieving AGI would mean solving problems which are unbounded and training a model on far more data than even LLMs were trained. For reference, OpenAI’s GPT-4 required approximately 0.02% of the electricity California generated in a year, and that’s only for this iteration, not including the three that came before.
We just aren’t quite there yet, and even when we are, what purpose will it serve? There are underlying issues of trust and accuracy to solve with current models first. Unlikely AI is pioneering transformative technology that addresses these issues, making AI more accurate, trustworthy, explainable and safe. The companies solving real-world problems and understanding how people can work with AI, rather than looking to replace them, are the ones able to fundraise now.
Good for your health
Real-world problems are certainly the focus for governments when adopting AI to improve public services. The National Health Service (NHS) in the UK is set to adopt AI to relieve pressure on its clinicians.
The National Institute for Health and Care Excellence (NICE) is recommending four AI tools to be used in urgent care. One of these tools will be used alongside clinicians to help spot broken bones, which are missed in 3-10% of cases — a symptom of the huge workload placed on an under-staffed NHS. This technology will help ease pressure on the overburdened public healthcare system here in the UK. It’s already being used to detect the earliest signs of breast cancer on scans too.
Investing in AI
This is not investment advice. Innovation can and will happen in open source AI, the issue is that the ones who benefit from this will be those who control and own the data (or those that can build on top of these models, but with their own data). This is why we’ve focused our approach to investment in the area since LLMs and GenAI became mainstream. We look for the companies that solve hard or intractable problems for large or soon-to-be large markets. The importance for us is as much on the data which models are built as the problems they solve and the markets they address.
Truly innovative AI was never part of any bubble to begin with, as it provides real value to companies and people that’s evidenced by revenue which keeps pace with investment valuations.
Just remember this when you read another headline about AI speculation: it took Ford 19 years to sell 15 million Model T’s, and they didn’t even start selling them until 23 years after the car was invented. That’s 42 years from invention to mass market adoption. We’ll continue to invest in AI companies for many years to come.
