Paradox of Choice in the AI Universe
For those who are not aware, ‘The Paradox of Choice’ is a concept popularized by psychologist Barry Schwartz in his book of the same name which suggests that having too many options could eventually lead to dissatisfaction, paralysis, and even unhappiness. It is interesting how this pattern has been constantly haunting us in the online universe, ever since the crazy days of dial-up connectivity!
First, it was the huge array of search engines we had at our disposal. Everything from navigational experts like Altavista, Lycos, Yahoo, and Excite to unique players like AskJeeves. While the choice seemed like a luxury in the beginning, you could never be satisfied by a search in just one of these as in some cases you might consumed by a feeling of FOMO and launch a search in an alternative — just to check if the other engine brought up something the first one missed! And then you might get into a dangerous spiral of spending too much time to find even the most straightforward information!
Until the gamechanger called Google eventually arrived on the scene and removed this pain. And became almost a ‘Single Point of Truth’! Tricky, but worked — at least removed the pain of having to browse through so many choices.
Shifting to a specific domain like Travel. A similar symptom popped up in the mid-2000s. While the first set of OTAs(online travel agents) like Travelocity, Expedia, Opodo, Priceline Hotwire, Orbitz, etc felt like a boon to the online travel planners, the task of finding the best fare a flight or the most discounted rate for a hotel room became a nightmare. Because, until you browse through every one of them you couldn’t be sure if what you were planning to book was indeed the best deal! And this deliberation only increased with the arrival of more and more players into this space.
Until a pain-relieving technology called Metasearch marched its way into the crowded travel-tech space. Apps like Kayak, Skyscanner, Wego, Kiwi, Momondo, HotelsCombined, Trivago, etc alleviated the pain of having to check every OTA on your own. As a solution they programmatically browsed and highly automated the exercise to provide an easy comparison. But after some time, even these so-called saviors started to become unreliable — thanks to the disparity in the prices each of them started showing on their platforms for the same item(a flight seat or a hotel room) So now, instead of having to check prices across multiple OTAs, we were pushed to search across multiple meta-search platforms — leading us to crave for a meta-for-meta!
Fast forward to the 2020s. And here we are with yet another interesting Paradox of Choice avatar in our hands. And this time it is getting ‘generated’ by the wide range of Large Language Models(LLM) and their diverse range of Generative AI capabilities! Dont get me wrong — these are certainly game changers. Making our lives much easier — and boosting our productivity to unbelievable levels. Especially by tremendously simplifying the Finding Information and Research and Analysis functionalities of the traditional search engines(including Google!)
Tasks that involved spending minutes — and included engaging in multiple mundane actions like launching a search, fighting your way through cookie consent pop-ups, dodging annoying ad banners, clicking through multiple Links retrieved by the search engine, and eventually extracting the exact information you were looking for. Phew! For example, a task like “Finding the recipe for a simple pancake that is easy to make and healthy” would have involved at least(if not more) the following steps:
- Search: Type in a carefully crafted search query (hopefully composing it in a way the engine understands the intent!) in a search engine like Google or Bing in a browser(or app)
- Consideration: do a cursory review of the top-ranked links and pick the one that seems the most interesting, relevant, or convincing
- Education: Load a site like Allrecipes.com and then attempt to browse through the content to understand the recipe and cooking instructions. This would in most cases involve multiple UX annoyances already called out (like cookie pop-ups, ad banners that use retargeting to pitch products completely irrelevant to the current search intent, videos you might be forced to watch before unlocking the content, paywall, etc). In most cases, the content itself might be fragmented and split by multiple ad segments!
- Selection: Assuming you manage to patiently browse through a few recipes, then comes the hard task of picking the optimal one, or the one that will suit your requirements. Of course, there are going to be dozens of recipes for ‘Simple Pancake’ from Traditional Aunt Annie to Health Conscious Henry! 🙂 In most cases you are left without any assistance to choose which might be the healthiest (even if you find a few simple recipes) or vice versa.
Do note, that in the example above, I have only called out browsing through a single website under the education phase! Now imagine having to repeat this tedious process for multiple websites!
With the advent of Generative AI Chatbots fueled by LLMs, this entire exercise is now potentially shrunk to typing a simple prompt like the one below into any of the popular tools — like ChatGPT, Microsoft’s CoPilot, Google’s Gemini, or Perplexity.
“Give me a recipe for a simple pancake with all the ingredients and cooking instructions listed in an organized fashion. Ensure the recipe is healthy and I can make this with minimal cooking experience”
And voila! Within a few seconds, you have clear-text content(no distractions or annoyances) that lists out the ingredients and instructions for a single recipe that the machine has chosen to match your needs the best! Instead of having to deal with multiple options or browsing fatigue! That is wonderful, no? But hold on! What if the information you are looking for is not that simple or that straightforward? That is where the paradox of choice creeps into this otherwise magical universe of AI!
Take for example a task like finding a specific song with only limited information like the lyricsist and what concept it talks about. I attempted such a search using four of my favorite Conversational AI tools. Not so surprisingly, each of them shared very different suggestions. Of course, I wasn’t expecting convergence here, or each of them to be pointing to the same answer. But what I observed was quite unexpected.
- Google Gemini: The only contender in this list to transparently say it couldn’t find any exact match! This might be true which makes Gemini very tight.
- Microsoft CoPilot: The answer brought up was not just irrelevant, I am not even sure how it picked this up. Maybe I am missing something not so obvious
- Perplexity: This tool at least brought up a song that was written by the lyricist I was looking for. But the song suggested was quite off from the query intent
- ChatGPT: Once again, this was wrong both in terms of the lyricist I was looking for as well as in terms of relevance. Also, this response was quite peculiar as Microsoft’s CoPilot is supposed to be riding on top of the same LLM ChatGPT was using and they were not even remotely aligned. This might be because I hit ChatGPT 3.5 and CoPilot probably used the GPT 4.0 model.
In the end, I am not even sure if a song like this exists or if one of these songs was really what I might have been looking for — thanks to this wide disparity, as well as confusing suggestions.
So, where does this leave me? It leaves me with a belief that I have to check every one of these tools to get the most accurate information when have slightly complex questions & queries or receive ambiguous answers. It leaves me with a strong desire for a metasearch equivalent that could wrap around these tools to compare the answers side by side easily.
In any case, I feel like we have already entered a new era in information retrieval that seems to have introduced new challenges and hurdles to the information seeker. An era where traditional search and browsing experiences will be severely altered( and more painful — thanks to publishers scrambling to make money with more eyeballs getting stolen by the AI tools!). An era where validating the credibility of information retrieved is going to probably require more effort than what’s required to retrieve it(I have already experienced enough fake suggestions by Gemini from its Hallucinations that I had to tediously validate!)
But then, every chapter of disruption comes with its blend of challenges, right? So, despite the paradox, I am as excited as most folks to wait and watch us evolve as humans with a machine-driven assistant to help us navigate this ever-expanding universe of information!