This reminds me of when Google wiped out every other search engine back in the day, and searching the internet became known as "Googling." Back then, the technology Google created was pioneering, and their search results were reliable. What we asked of Google wasn't as complex as what we ask of AI today. We'd ask how to get between two locations and be pointed to the websites of the relevant train providers. Google could be trusted to find the right options, but we still needed to check the details on the train provider's own site — timings, connections, and constraints were left for us to work out.
This morning I was researching something for a client using Google, and I pressed the AI results button, which took me into Gemini. After a couple of rounds of chat with Gemini, I got to the final results I needed and was presented with two sets of results in two separate columns and asked to grade which one was best.
Back in the day, I don't ever remember being asked which Google results were better. Google had the expertise to know on its own that its search results were good. Google pioneered back then — their staff used to be able to spend 20% of their time (one day a week) working on projects they were passionate about, and GMail famously came out of that. Gmail itself was kept in "Beta" for over five years. Being in Beta meant mistakes were allowable and Google compensated for that, and we understood that Gmail would keep developing and improving over those years.
This was before Google listed on the stock exchange, and the drive for profitability to satisfy investors has meant, according to Google themselves, that "the policy has faced tighter management controls and shifting corporate focus toward core products." The pace of change has increased so much over the last decade that it feels like we're still in "Beta" with AI tools — except they're no longer labelled Beta, so there's no allowance made for their mistakes.
Our demands of AI have grown — we're asking more complex questions, and the search providers are now doing part of the thinking for us. We're assuming they're loading up the website to check times and connection constraints on our behalf. I suspect the feedback Google is now optimising for is speed over quality.
No matter what an AI engine tells us, we need to exercise due diligence to review and validate its output — otherwise we're risking AI slop. Maybe I'm looking back with rose-tinted glasses; I don't remember putting nearly as much scrutiny into reviewing Google's results. But then, the decisions about which train and which connections to take were always mine to make.
We need to be careful about what decisions we outsource to AI, and make sure the necessary checks and balances are in place.
These are all my own words, Claude was used for spelling, punctuation and grammar.
This reminds me of when Google wiped out every other search engine back in the day, and searching the internet became known as "Googling." Back then, the technology Google created was pioneering, and their search results were reliable. What we asked of Google wasn't as complex as what we ask of AI today. We'd ask how to get between two locations and be pointed to the websites of the relevant train providers. Google could be trusted to find the right options, but we still needed to check the details on the train provider's own site — timings, connections, and constraints were left for us to work out.
This morning I was researching something for a client using Google, and I pressed the AI results button, which took me into Gemini. After a couple of rounds of chat with Gemini, I got to the final results I needed and was presented with two sets of results in two separate columns and asked to grade which one was best.
Back in the day, I don't ever remember being asked which Google results were better. Google had the expertise to know on its own that its search results were good. Google pioneered back then — their staff used to be able to spend 20% of their time (one day a week) working on projects they were passionate about, and GMail famously came out of that. Gmail itself was kept in "Beta" for over five years. Being in Beta meant mistakes were allowable and Google compensated for that, and we understood that Gmail would keep developing and improving over those years.
This was before Google listed on the stock exchange, and the drive for profitability to satisfy investors has meant, according to Google themselves, that "the policy has faced tighter management controls and shifting corporate focus toward core products." The pace of change has increased so much over the last decade that it feels like we're still in "Beta" with AI tools — except they're no longer labelled Beta, so there's no allowance made for their mistakes.
Our demands of AI have grown — we're asking more complex questions, and the search providers are now doing part of the thinking for us. We're assuming they're loading up the website to check times and connection constraints on our behalf. I suspect the feedback Google is now optimising for is speed over quality.
No matter what an AI engine tells us, we need to exercise due diligence to review and validate its output — otherwise we're risking AI slop. Maybe I'm looking back with rose-tinted glasses; I don't remember putting nearly as much scrutiny into reviewing Google's results. But then, the decisions about which train and which connections to take were always mine to make.
We need to be careful about what decisions we outsource to AI, and make sure the necessary checks and balances are in place.
These are all my own words, Claude was used for spelling, punctuation and grammar.