The Fundamentals of Search

With the rise of social networks in the last 10 years, and Generative AI in the last 2, it’s a messy time to start predicting the future of search and travel discovery.

Most of us have spent the last 2 decades using Google as the ultimate source of information, and it’s difficult to argue that Google hasn’t done a fantastic job of implementing their mission statement: “organizing the world’s information and making it universally accessible”. 

It’s easy, it works, and in many ways it feels like it might be the most intuitive way to find information. Nobody has been crying out for some alternative format, and If some other interface or data structure was superior, Google would have evolved to it already. Because Google has been able to monetize the formula with Ads that don’t significantly detract from the user experience, they have largely been able to stay true to the mission of getting the customer to the information as quickly as possible.

But is it optimal?

It would be interesting to use first principles and think about the optimal way for a human to discover and consume information? If you go back in time and ask people that question, there might have been answers about big flat cave walls, nifty devices to store scrolls, fancy libraries of encyclopedias, until finally agreeing on a search engine like Google – the world’s information on a multi-modal device that fits in your hand.

But knowing what we know today, what we come up with? (In the future, information paths will bypass our senses and be sent directly to neurons. But we’re talking about the stage before that.)

I imagine 3 fundamental formats, based on how humans have evolved to consume information. The smart-phone may or may not be part of the flow here, but ultimately it comes down to a mix of audio and visual, and it comes down to the presentation and sources of the information.

Large Language Models. Train a computer on all of the data in the world. Build a chatbot that can converse over voice and text, deliver any information about any subject at any time, and in a format that the user asks for. Most would argue this one is tough to beat. Up until 2 years ago, most expected this to be decades away..

Google. Yes, it’s still relevant, at least for now. Text search presents credible third party sources for places to do further research. Results are sorted by a complex algorithm, which prioritizes credible and trusted sources, and which has a singular goal of getting users the information they are looking for. Users being able to pick and choose sources is a fundamental feature. Trust is vital, and the future of influencers / brands should have some staying power here.

Video Platforms Ask a question and have a knowledgeable, credible human show us the answer in video format. Inspirational for travel content, practical for how-to’s, presentations for educational etc. This is clearly a format that attracts humans. It’s a more passive way to consume answers.

That’s all very convenient. Pretend to go back in time to predict the future and conveniently end up with the exact state of the world today. The fake prediction isn’t the point though. The point is that this place we’ve ended up might just be the absolute optimal solution.

There are some problems though.

LLMs are new to all of us. As the best source of truth, there is little doubt they’ll win. But today they hallucinate, and are troubled by pesky human knowledge full of bias which they had to train on. Not so much discriminatory bias, but cognitive bias which causes humans to have problems understanding truth. Over time, the LLM will shake most hallucinations. Not all, because to some degree hallucinations are part of creativity and you don’t want to lose that, only control it. On cognitive bias, humans will never shake it. But LLMs will understand that, and adjust information accordingly.

Because the companies building the LLMs mostly have a wider goal of AGI, the models will always seek out the truth. Where travel data exists, they will eventually find it, and make it discoverable.

When it comes to travel data, the recent LLMs are filling the gaps. I’d argue there isn’t much information left on websites that the LLM doesn’t already surpass. Up-to-date business and operational data being one exception.

Google works. But is it going to start looking like too much work compared to other methods? . Certainly LLMs will take a decent chunk out of Google searches. Why do we have to choose which link to click on? Often we don’t know, and we click on 3-4 which ends up being very random and inefficient. Gen Zs apparently agree that it’s too resource heavy – searching videos is more intuitive to them.

Video Platforms are problematic. Let’s say Tiktok for arguments sake, and if it’s not TikTok, assume Instagram will take the space. The problem here is that it’s an entertainment app, not an information retrieval app. But it IS a great source of information for many things, so it gets tricky, and people get confused. The problem is the algorithm. It only cares about engagement. There’s no deeper purpose. And to increase engagement, you have to cater to whatever it takes in human psychology, which often isn’t pretty.

You can argue, when it comes to travel they’ve succeeded at the inspiration part. The informational part – it’s not even close. There’s an absolute conflict between viral content and informational (factual) content. That conflict isn’t going away. It’s baked into our brain. And because of that it’s baked into the algorithm.

I don’t see a way out of this. What could happen here? A toggle on TikTok so we can switch to informational mode vs entertainment mode. Not going to happen. It’s not TikTok’s problem.

The video format is not going away any time soo. It's too popular. If TikTok gets banned in the US, it's exactly because of this conflict, and the power the algorithm has to influence our brains without realizing. 

The ultimate solution then? It must be a hybrid. Visually, a UX not too different from Google or Perplexity today. Google will keep moving towards this hybrid. ChatGPT is building the same thing with its newly released ‘Search’ option, but from the other direction. Voice will end up with a seamless transfer back and forth with our screens. 

A whole article without talking about Agents? No, our AI agents will throw a wrench into this whole thing. The agents pre-compress all of the data before it gets to us. It will de-bias everything inbound, and serve us the information in a format we want. The problem is we’re only humans, so we’ll instruct our agents to inject more bias - picking specific sources and brands against our own best interest. So we might end up back in the same place.

What about the problem of the videos? One possible way out will certainly be judged as the worst idea in history. Videos generated mostly by AI. Keep the 5% of videos today that are credible. Fill the gaps with synthetic videos, using only information created by the LLM so that it’s trustworthy. What if you didn’t even know they weren’t real? You saw it here first. We need probably another 2 years until this is 100% viable.