In April 2024, I had the privilege of delivering a talk in the historic theatre at the Royal Institution in London. It was for Hoare Lea’s Designing the Future event, and the theme was disruption. I chose to focus on the recent arrival of generative AI.
And by generative AI, I don’t just mean using tools like ChatGPT and DALL-E to generate text and images. I include the breakthroughs achieved by Google DeepMind, particularly with Alpha Go, and then Alpha Go Zero.
It wasn’t just amazing that the Alpha Go algorithm beat the human world champion at Go, a game considered too complex for an algorithm to play. It’s that the algorithm generated strategies that were alien to the top human players.
Alpha Go Zero went a step further in learning to play Go, and then beat Alpha Fold, within 34 hours. All without being given any human-generated or labelled training data. It created its own data from running millions of simulations. For me, this is still the closest we have come to true AI.
And it is why I believe this area is the third great disruption brought about by information technology.
The first: the invention of writing information. The ability to create an artefact of information that persists outside of a person’s head. That can be stored, viewed and shared ‘as is’, instead of being passed down through spoken narratives that erase or embellish some or all of the original content. It led to the administrative state and growth of large complex societies.
The second: the invention of copying information. The ability to reproduce artefacts of information using machines instead of human copywriters. It accelerated the creation and spread of written knowledge. It enabled science to flourish and, ultimately, led to the Industrial Revolution and the creation of the digital computer.
The third: the invention of generating information. The ability for algorithms to generate new information, mimicking human imagination. It is the first time we have had a non-human form capable of generating new content, whether as language, images, or strategies to solve complex problems. What will be its impact on social systems?
Whilst there is a huge amount of hype around current developments and whether or not the arrival of large language models has moved us closer towards developing artificial general intelligence, to dismiss generative AI as a fad is to underestimate just how different these modern algorithms are to everything that came before.
And whilst there is a lot of noise around flippant and exploitive uses of these algorithms, we are also seeing some exciting advances in a range of fields. It has been said that scientific breakthroughs have been slowing down over the past 60 years. Perhaps new approaches are needed, including non-human techniques for generating solutions. This was most visibly demonstrated with Alpha Fold taking on the 50-year-old grand challenge of protein folding in 2018. There are also promising developments in materials research that would have taken decades to complete without access to modern algorithms. Instead, outcomes are being achieved in weeks.
That was the theme of my talk. 9 minutes to summarise just how different these algorithms are to anything that has come before, the scale of computational power behind them, and to imagine what grand challenges in the built environment they could help us solve sooner, rather than later.
Here’s the talk:
There were other amazing talks during the evening. To view more: https://hoarelea.com/2024/05/10/designing-the-future-2024/
References/Further reading:
Protein folding: https://deepmind.google/discover/blog/alphafold-a-solution-to-a-50-year-old-grand-challenge-in-biology/
Materials Research: https://news.microsoft.com/source/features/innovation/how-ai-and-hpc-are-speeding-up-scientific-discovery/
Computer vision breakthrough: https://www.technologyreview.com/2014/09/09/171446/the-revolutionary-technique-that-quietly-changed-machine-vision-forever/
Arrival of generative algorithms: https://venturebeat.com/ai/gan-generative-adversarial-network-explainer-ai-machine-learning/
Rise in computation power: https://www.visualcapitalist.com/cp/charted-history-exponential-growth-in-ai-computation/
Breakthroughs in AI and machine learning algorithms, 2006 - 2023: