Would You Trust Perplexity in a Natural Disaster?

Children born in 2020 are forecast to experience up to 7X as many climate disasters as adults born in 1960. When they do, they'll pick up their phones to look for resources and help.

Today's news that Perplexity could be coming to every Samsung phone in the world brings up big questions about how we'll relate to an increasingly dangerous future. Perplexity's most recent ad is NOT how most people will use the technology in the face of danger.

Samsung devices are the 1st or 2nd most popular in most regions of the world. Will the AI on our phones be helpful and accurate, ready to serve us in our specific places and circumstances? Or will it too often be inaccurate? What's a level of acceptable inaccuracy that's worth the trade-off to get incredible access to information?

I use Perplexity every day, for many things. But would I tell my loved ones to make life-or-death decisions based on Perplexity search results? Right now I would urge them to take the seemingly useful results with a very big grain of salt. I hope Perplexity will rise to the occasion.

Mobile Challenges

Both Perplexity's partnership with Yelp for local business data and their documentation on filtering by location demonstrate that they are working on location, of course. (Both found via a Perplexity research query on this topic, thank you Perplexity.)

But if they land on every Samsung phone in the world, given the direction the world is going, they are going to have a lot more location work to do. Including for traditionally underserved markets. How's the training data, the web crawling, and the co-ordination with existing disaster response systems?

In the global environment of 2025, a disaster scenario is often an adversarial one - with bad actors actively poisoning the information commons in order to further harm rival countries struggling with fires and floods. (Here's a good Perplexity round-up of examples.) Will Perplexity account for that? Will Perplexity be simple enough to work within the limited cognitive capacity remaining when people are in a disaster scenario?

Doing the Work to Minimize Trade Offs

I asked for thoughts on this from Ed Bice, CEO of Meedan, who after years of global leadership on accuracy online is now working on AI for the Public Square.

Ed said to me, "Working on AI tools for information response, Meedan has wrestled with the fundamental dilemma, What is an acceptable risk profile for meeting information needs during a natural disaster?

"If we are using AI to respond to questions over WhatsApp, as we did during the devastating 7.8 magnitude Turkey/Syria Earthquake in early 2023, is it better to lean into automation to have more valuable information distributed or does the risk of a bad result (imagine directing a user to the wrong location for an emergency shelter) recommend excluding AI from emergency response scenarios?

"Like everything, the answer is in the system design. There is much that AI can do to assist the human digital first responders, and there is much that these teams can do in front of natural disasters to pre-train RAG systems."

Let's hope that Perplexity and all relevant mobile AI players are ready to do the best work they can to prepare their technology to serve up helpful and accurate information as often as possible to the front lines of an increasingly dangerous world. The connection between Google Maps and the community-editable Open Street Map as source data comes to mind. Ushahidi's disaster response mapping work comes to mind. Worst case scenario, the energy needed to support AI makes climate instability even worse and then when people reconnect to the network and reach for help, AI search results are confusing and inaccurate when people need help the most.

Best case scenario, work is done both top down and bottom up to make AI a powerful source of help as we adapt to the future.