A new way of searching

Discovering a new way to discover

A ground-breaking study from the Massachusetts Institute of Technology opens up an exciting new story. It transforms the way we’ll search out medical discoveries of the future.

The search, or quest, is a central feature of human storytelling across time and culture. From the Epic of Gilgamesh, on clay tablets, to The Lord of the Rings on DVD, our stories have prioritised the search for answers to seemingly invulnerable foes and unspeakable woe.

Even as individuals, our lives can be seen as searches for security, for approval and for meaning. Religions have long catered for this by offering belief and belonging.

Evolution as search engine

But searching predates human beings and can be seen as something much more fundamental indeed. Life and its evolution can be considered as a search strategy to allow a better fit between organisms and their environment. Through iteration, evolution allows the space of possibilities to be searched for the least worst solutions.

The human brain can be recognised as one of the outputs of that evolutionary search; an evolved organ that allows us individually and collectively to search for everything from partners to purpose. It’s our “Swiss Army knife”, vastly adaptable to circumstance and endowed with creative imagination to test possibilities that aren’t even visible before us. The 40,000 year old Lion Man of Ulm is said to be the oldest known artefact that represents something abstract that its human carver could never have seen.

The way in which the brain does all this work remains unclear and complex. But it’s not a computer churning through one task after another in series. It’s a huge number of networks processing information in parallel, but crucially allowing that information to modulate the network. Signal and structure co-evolve, impacting iteratively on one another. It’s a little like the way in which our online behaviour shapes the structure of what we then see online, in terms of ads and recommendations. In that sense, the brain behaves as a search engine and vice versa.

For most of human history, our brains have been our most sophisticated search engine; finding solutions to problems we had; fulfilling needs we hadn’t even realised we felt. In a way, that remains the case. By inventing computers and then supercomputers, our brains have enabled a new type of searching, more powerful than what we’ve managed so far. And this isn’t merely the use of unmanned probes to search the physical cosmos. It’s the use of artificial intelligence to search the vaster space of possible solutions.

Manufacturing intelligence

In 1996, the world’s best human chess player, Gary Kasparov was beaten in a game for the first time by Deep Blue, an IBM supercomputer. Though he won the match overall, he tells how this was the turning point. Once a machine had shown it could be done, it would only get better. And so it proved. The following year, Kasparov was defeated in the match overall. But again this was just a start. Deep Blue prevailed through brute computing rather than something we’d recognise as intelligence. It crunched through the possibilities without needing to use intelligence to efficiently winnow out the non-starters.

This week’s research paper from the Massachusetts Institute of Technology points the way to a new era in searching for our solutions. The scientists used artificial intelligence (AI) to search out new antibiotics whose properties had not even been imagined by human scientists alone. Their search strategy was able to do in hours, days and months what it would have taken human intelligence alone, many years to do. Indeed, humans alone may not have discovered these new agents at all, so focussed are we on drugs that might be similar to ones we already know to work. This application of AI opens the way to far faster drug discovery and also the identifications agents we would never have anticipated. What’s more, the value of this approach is only likely to increase as the method is further improved. As Kasparov points out, everyday chess programs can now beat him and Deep Blue, such is progress.

If this promise is realised it heralds a new age for discovery, where AI assists and even leads the human search. It can be applied not just to antibiotics but to a vast realm of agents of potential use in many other conditions. It’s already being applied to problems outwith medicine too, from facial recognition to spam filtering to driverless cars.


The remaining questions are however important. AI looks set to search out possible solutions we could never have imagined alone. But it will go further. It will search out solutions that work but which we can’t even understand, even when right in front of us.

This raises ethical questions. If the AI solutions work but we don’t know how, how do we know that they’re fair or just? We know that human biases can influence the outputs of computer algorithms. How do we know that our AI solutions are free of such bias, if we don’t even understand how they reached their answers? For all our modern sophistication, it may yet be that we’re driven back to primitive superstitions about how our AI answers have been obtained. In that sense, there remain some issues on which we’ll have to keep searching.