OpenAI’s chatGPT and the future of Search Engines

Launched on Nov 30, 2022, “chatGPT” has joined the ranks of other industry buzzwords in the modern era. With over a million users in just five days, chatGPT has rattled both the IT industry and education sector. Its meteoric rise has leaders of both these domains sending red alerts on the future of their businesses. It now begs the question: is this the beginning of the end of search engines? Could this be why Google, being the market leader in the search business, is internally sending code red alerts and universities & schools are worried about their students (mis)using chatGPT? We can draw further extensions from this. How is chatGPT going to impact the software engineers and developers? Could chatGPT accelerate Quantum Computing?

To explore these questions, we have to go back in time to get a grip of the foundation of what is causing this fear.

Evolution of Search Engines:

It has been a highway drive for search engines since their conception from the mid-1990s. While there have been many over the years, the top search engines dominating the world include Google, Bing, Yahoo, and Baidu.

While it was Yahoo that initially came out swinging in 1994, it was unable to maintain its lead in the long term. Google’s “pagerank” algorithm and its subsequent iterations marked distinctive accuracy and relevancy of the search result pages and the site boasted a faster, more responsive user interface. By the late 2000s, Google had surpassed Yahoo in market share, and since then, has tremendously widened the gaps between competitors in all aspects of search business including accuracy, relevancy, key words, bidding for paid search, cost per click, revenue, search volume, advertisers and all other key parameters. It has now become the De Facto search engine in all home desktop computers and laptops. Several other search engines have sprung over the years including Microsoft’s Bing in 2009 (initially rolled out as Windows Live Search in 2006), but Google’s dominance was so strong that competitors aim simply to catch up. As per to Statista.com, Worldwide desktop market share for search engines, published in July 2022, Google had 83.84% market share, Bing 8.88%, Yahoo 2.55%. There are several other search engines which makes up the rest of the market share.

Birth of?chatGPT:

Fast forward now to November 2022: OpenAI released its AI Chatbot system “chatGPT”. Suddenly, everyone in the tech industry is talking about it. Could this be another rendition of how a technologically more capable service beats out its competitors – akin to how Google surpassed other search engines in the last couple decades? To explore this potential parallel, it’s integral to understand what “chatGPT” truly is and why it was made.?

OpenAI was founded in 2015 as an AI research group and as a non profit by Sam Altman, Elon Musk and others with a whopping investment of USD 1 Billion. It attracted the best brains from top universities and organizations who are interested to take AI to the next level and mainly towards “benefiting all of humanity” (albeit remembering that the semantic definition of a “benefit for humanity” is subjective and deserves its own discussion). While Elon Musk left the OpenAI board in 2018 citing conflict of interest, he remains invested in the organization. Following Microsoft’s USD 1 Billion investment and partnership in 2019, OpenAI transitioned to a ‘capped-for-profit’ organization. Despite this, their mission statement remains unchanged. Thus, the goal of the organization and the products it presents to the public is not to tout monetary gain, but rather, make substantial leaps in the development of useful artificial intelligence.?

Search Engines, Bot and chatGPT

So, what is chatGPT? A straightforward answer to that is, it is a bot, but with a seemingly limitless pool of knowledge from which it can answer a user’s query. It seems rather similar to the way a search engine can provide a user a seemingly limitless amount of information given a query, does it not? This likeness is precisely what is sending shock waves.?

In a traditional search from a browser, a user types a search query and is greeted to a web page with “m” number of results displayed over ‘n’ number of pages. In addition to these organic search results, these result pages may also contain partially relevant promotions, irrelevant advertisements, and potential search suggestions. By design, the search engine essentially displays a smaller subset of pages that could help the user find answers. It’s true that modern iterations of Google or Bing contain extra panel components to surface answers to query questions without exiting the search engine itself.?

A bot, on the other hand, is an agent that can converse with a person, which is trained to answer certain questions. Bots are widely used now in almost all consumer and customer centric online businesses and portals. The bots are typically trained with machine learning algorithms for natural language processing to parse the user query and match with a pre-existing question/answer bank. Most of these algorithms are some variations of supervised learning. Indeed, training these models takes time, sometimes minutes or a few hours (taking into consideration of production deployments), but once complete, they are rudimentary user interaction vessels.

Drawing from both prototypical search engines and chat bots, chatGPT serves as a personal encyclopedia. Using the chatGPT text box, users type a meaningful question or keyword, a phrase or anything of question and press the submit arrow. The internal model will generate a text-based response as the result. The experience the user gets is almost as if somebody on the other end is typing the answers for you. There are no pages or lists of relevant pages shown. Just the bot window. By also “remembering” the entire stretch of the conversation, the model generates useful, natural text to answer the user’s queries on a topic. chatGPT is so intelligent that it can build a SQL Query, write complex stored procedures, can debug a piece of code, provide suggestions to code. In simple words, it could be a personal secretary. It can give suggestions about what you can eat to how to exercise and what workout to do to build muscles and so on. It is limitless. On the flip side, it can also write essays for school and college students.?

While the front-end is surprisingly intuitive, the inner workings of the model are where the technological advancements lay. GPT stands for Generative Pre-Trained Transformer – a model trained using supervised learning and reinforcement learning.?The GPT3 model, the latest iteration by OpenAI and the one used for chatGPT, is trained on a vast dataset comprising of nearly all human knowledge on the web as of 2021 and contains hundreds of billions of trainable parameters.?

Such a vast amount of data that was processed and used for training seems almost unfathomable. We have been talking about big data for a while. The volume, velocity and the variety of data plays a major role in training any ML models to make accurate predictions and outcomes. But the amount of data which openAI has processed to train chatGPT cannot be underscored enough. Here is where it gets mind-boggling. The GPT3 model needs hours and sometimes days of training ranging from 120 hours to 160 hours on a dataset. Think about the resources consumed and the computing power needed to train these models with this vast variety of information available in the globe. openAI uses Azure for computing power, and it is evident that this partnership is going to define the future of AI in the coming years.?

Future of Search Business, Monetization and Ethical AI

Given the power and intelligence of chatGPT, the big question that comes to several of our mind is: will chatGPT end the web search business? The answer to this is a big NO, at least not for the next several years. But what it does indicate is the beginning of a new paradigm, which is going to define the way search engines and their underlying business’s function.?

First, all search engines are internet based and use highly advanced crawlers to help surface the optimal webpages to answer a user query. These are integrated into well-designed pipelines that allow for real-time information to be incorporated in the search process. But since chatGPT is meticulously trained on a finite unlabeled dataset from across the globe, it does not have access to the latest data like current search engines. Given the time-intensive and expensive training that chatGPT needs, it may not be feasible to re-train and publish the model on a scheduled basis. This is a major limitation for chatGPT and a main reason why chatGPT could not replace search engines as of right now.?

Another key limitation lies in the veracity of results from chatGPT. Granted that it is, at the end of the day, a trained model, its singular outputs always have the chance to be unhelpful, or possibly even outright counterfactual. Given that chatGPT is in its nascent stage, it lacks heavy moderation of these incorrect responses and its output could be harmful as well. While trained to be unbiased and neutral, given its stochastic nature, the bot could even generate hateful content. From another perspective, if the students in schools and colleges start using chatGPT to write their college essays or misuse it for any of their academic work, plagiarism will become rampant. These squarely put the bot under the ethics radar. We will come to know more about the implications in the coming weeks and months.

Perhaps another avenue of thought and discussion is more uniting in approach. What if the organic, natural language generated by chatGPT could be merged with the real-time results sourced by search engine providers? If that happens, it would be a major paradigm shift in search business, a shift that defines the union of online search and AI Bots.?

This makes me to write that for the first time in few decades, Microsoft has the opportunity to not play the catchup game. Given the partnership with OpenAI, Microsoft could leverage both research divisions to make advancements in quantum computing to build faster supercomputers to train large generative models like GPT3 to add chatGPT-like functionality to their search engine. If successful, it is going to be a blue ocean for Microsoft. However, since advertisements and user search data drive the monetization of search engines, it is still difficult to picture what a realistic user experience might look like. In the current release, chatGPT does not have an integrated platform to add advertisements. Though technological advances are a must for the organizations, it should also lead to monetization and revenue. The success of any organization is measured through the revenue they generate through ethical means. That is where product managers can help provide direction.?

It is given that the world of business is defined by the survival of the fittest, most dynamic, and innovative organizations. The same is true for the web search business. Those unwilling to explore the integration of generative AI models to complement existing search engines should certainly be worried. Billions of dollars of revenue are now at stake once again.