💡 The AI Revolution is Just Beginning
Artificial intelligence can be hard to keep up with. It’s a highly technical field, and major leaps forward happen so frequently that it’s hard to make sense of one breakthrough development before the next one has arrived.
That’s where my latest podcast guests come in: Nathan Benaich and Ian Hogarth. Their annual ‘State of AI’ report takes stock of the major developments in artificial intelligence, as well as predicting a path for our near future.
The authors know their stuff. Nathan is the General Partner of Air Street Capital, a venture capital firm investing in AI-first technology and life science companies. Ian Hogarth is the founder of concert-discovery service Songkick, used by 17 million people worldwide. He’s also an angel investor in more than 100 startups, and a Visiting Professor at UCL. They’ve seen the inner workings of dozens of companies with artificial intelligence at the core of what they do.
The podcast (like their report) is brimming with insight into what AI is doing, and what it might do next, in areas from drug discovery to warfare. It’s a clear, precise look at how the great general-purpose technology of our age is changing the world right now.
You can listen to my conversation with Nathan and Ian on all podcast platforms, including Apple Podcasts and Spotify, or read a transcript of it here.
The Big Idea
I asked Nathan which AI advance of the last 12 months represented a big breakthrough – a Sputnik moment. He pointed to work being done by DeepMind, Salesforce and others that use natural language processing (NLP) – AI techniques that analyse and replicate human language – to read protein sequences and determine how they work.
Proteins are built from an “alphabet” of 20 naturally occurring amino acids and, as in human language, certain patterns tend to recur. An NLP model trained on proteins can predict what certain strings of amino acids might do. That knowledge can be used to make further advances, as Nathan explains:
[U]sing models like that and training them on protein language, which is a string of amino acids, and using that model to generate entirely new protein sequences or versions of sequences that we have in the world today, [...] can generate artificial proteins with new function.
That shows how brilliantly adaptable this cutting-edge research in AI is, and why I consider these general-purpose technologies (GPTs) as seminal as the phone or car once were. They will improve scientific methods and accelerate progress across the board. In my book, I write:
[General-purpose technologies’] effects spread inexorably from area to area, rippling across all aspects of our daily lives. And the GPTs of the Exponential Age are only beginning to emerge.
Indeed, protein-sequence prediction – largely hailed as a medical breakthrough – could have numerous exciting applications. Here’s Nathan with one example:
I think the most exciting implication of this is to hopefully accelerate our transition away from a petrochemicals-driven world where most of the materials that we create and consume around us are based off of a very small number of chemical backbones, and instead moving towards a world where we either use microbes or various living organisms to create the materials that we need.
The Early Innings
Nathan and Ian touched on something really fascinating in our discussion: the huge growth in MLOps (machine learning operations – best practices for running machine learning systems at an industrial scale) in the last few years. MLOps is an area I spent some time evaluating a few years ago. (At my previous startup, PeerIndex, my engineering team had to solve lots of problems in this field, although it wasn’t called MLOps back then. I’m also an advisor to Seldon, one of the leading players in MLOps.)
That points to a maturation in machine learning – industrialisation of the process – and suggests there is a huge deal of growth still to come. Here’s Ian:
[Y]ou have people putting large sums of money into very, very speculative, long-term, research directions. There's tons of discoveries happening, a huge amount of capital coming in, but it's still insanely early if you think about the kind of aspirations that these researchers have and how intelligent they want to make these systems in the long run.
Interestingly, MLOps is dominated by startups. Some commentators think the space is ripe for consolidation, which should mean a whole lot of M&As in the coming years.
We are, as an American might say, in the early innings of the game. Artificial intelligence and machine learning are changing the world, and they’re just getting started. To hear more about it, listen to my conversation with Nathan and Ian here.
We also discuss:
🏢 What Salesforce’s biology research says about the AI industry [14.59]
🤖 How the AI/ML engineer shortage could be solved by… machines [24.01]
⚔️ Why tech ‘choke points’ could become political battlegrounds [39.23]
Listen to this, too
Last year, I spoke to Demis Hassabis, the founder and CEO of DeepMind, perhaps the most exciting AI company around. Just this month, Demis announced the creation of Isomorphic Labs, an Alphabet subsidiary that will use artificial intelligence to aid drug discovery. When we spoke, we discussed how DeepMind moved from solving computer games to helping solve some of the biggest challenges in science.
Demis is a searingly smart guy – you can listen to our podcast here.