Listen to people speak about search engines (I’m speaking at a search engine marketing conference today), and you’ll hear one word a lot: index. The role of a search engine marketer is to get their client’s sites, products or services, to the top of the relevant indexes.
The problem with this is that an index is, by definition, finite. However fast the sum of sites and content on the web is growing, right now its scope is still essentially finite. It takes a huge amount of smarts and a great whack of processing power, but search engines can index pretty much everything people might want to find.
What happens when the supply of content scales exponentially? We’re already seeing a dramatic increase in automated content creation. People are using programmatic techniques to create everything from news articles, to t-shirt slogans, to phone case covers, to children’s videos. This trend is only going to increase, particularly with the growing ability to remix existing content in new and disturbing ways.
But this is just the beginning. The real game-changer is ubiquitous cameras coupled with machine learning.
Imagine that most of the population spends most of the day walking around with a camera, capturing everything around them. (How will this happen? I’ve written about that here.)
Now imagine the output of all those cameras is piped through machine learning systems that doesn’t just recognise objects, it understands their state and context. Is that bottle full or empty? Is it in a kitchen or a bathroom? What’s that text on the front of it? Can I scan that barcode?
Every person, place, object, and state is indexed in ultra-high resolution, in real time.
Now the index of possible answers to any search query is essentially infinite. How does search work in this context?
Firstly, I think it has to be hugely personalised. The engines searching for us will need a deep understanding of us and our needs in order to filter signal from noise.
To some extent they have this already: all our search results are personalised by our location and previous behaviour, amongst other signals. But they might need more information to match our needs more precisely: emotion, health, social graph and more.
Second, I think more of the process will be automated. Machines will effectively pre-search for us based on their understanding of our needs and often present us with results before we have formulated the question.
These shifts present enormous challenges and risks. The risk that our worldview is narrowed. The risk that our buying behaviours are essentially controlled.
That’s why I’m increasingly of the opinion that the machines we let do this on our behalf — tomorrow’s discovery engines — must work for us. That means we must pay for them, not trade access to them for our personal data. And it means we must understand them: they cannot be a black box.