Google Search and ChatGPT can coexist

Models for generative AI and traditional search are fundamentally different and have separate purposes

There is no shortage of articles and opinion columns doing the rounds on the challenge posed to Google’s search engine by ChatGPT, which was released by Open AI (heavily backed by Microsoft). The latter is a generative artificial intelligence (AI) model. Many are calling this new model the death knell for Google and other search engines, and news reports painting a terrorised Google are now the rage. The recently botched announcement by Google of its own capabilities in generative AI has only increased this speculation. I don’t share this view, and will explain why later in this column, after first laying the groundwork for what generative AI actually is.

The most salient feature of generative models for AI is that they scour almost every shred of information that is available on the web (or in?ChatGPT’s case, at least was available until 2021). The web is a data store that is doubling in size every two years, and at last count—at least when I was writing this column—it was 23,429,747 petabytes, and is growing by 70 terabytes per second (bit.ly/3DYdmGX)!

First off, at this point, ChatGPT has current information only until 2021, as I pointed out above. That is now over a year ago, and given that the web is doubling in size every two years, a mathematical (admittedly simplistic) view of the data tells me that it has access to only three-quarters of the information that a traditional search engine has. Traditional search engines have access to everything, including what happened on the web just earlier today.

Second, the computing power required for generative models used for web search and traditional search engines used for web search can differ significantly. Generative models, such as GPT-3 developed by OpenAI, are often trained on massive amounts of data, and can require significant computational resources. This is due to the complexity of the algorithms used to generate new and unique content in response to a user’s query. The training process for these models typically involves multiple layers of neural networks that require large amounts of computing power to process the vast amounts of data used for training. The result is a model that can handle complex, natural language queries and generate creative solutions in real-time.

On the other hand, traditional search engines, such as Google, rely on deterministic algorithms that use pre-existing information to provide quick and accurate results. These algorithms do not require the same level of computational resources as generative models, as they are not generating new content in real-time. Instead, they use pre-existing information to rank websites based on factors such as relevance, authority, and popularity.

This is not to suggest that traditional search engines don’t require significant computational resources to handle the vast amounts of data they must process and index. They do. This is because they must constantly update their databases to ensure they have the most up-to-date information, and they must also be able to handle high levels of traffic from users making searches. That said, their algorithms are simpler and are apt for using less computing horsepower.

Third, we don’t quite know where generative?AI?can be compromised. ChatGPT has some safeguards in place. It actively blocks requests to generate potentially illegal content. Ask the service to write code for stealing data from a hacked device or craft a phishing email, and the service will refuse and instead reply that such content is “illegal, unethical, and harmful.”

But, sure enough, Ars Technica now reports that hackers have found a simple way to bypass the safeguards that have been put in place. According to Ars Technica, “The technique works by using the application programming interface for one of OpenAI’s GPT-3 models known as text-davinci-003, instead of ChatGPT, which is variant of the GPT-3 models that’s specifically designed for chatbot applications. OpenAI makes the text-davinci-003 API and other model APIs available to developers so they can integrate the AI bot into their applications. It turns out the API versions don’t enforce restrictions on malicious content” (bit.ly/3IjmOXO). This is not to say that search engines cannot be compromised in other ways, but they have been around long enough where the abuse of simple web queries (such as by a malicious automated bot) can be stopped.

All this leads me to the conclusion that the two have different comparative advantages and will continue to exist side-by-side. The theory of comparative advantage, proposed by David Ricardo, an English economist, in the early 19th century, suggests that countries (or in this case, companies) should specialise in producing the goods and services they can produce most efficiently and trade with others for the goods and services they cannot produce as efficiently. Applying this concept to the comparison between generative models in web search and traditional search engines, it’s clear that each approach (search vs. generative AI) has its own unique comparative advantage.

Traditional search engines are better suited for providing accurate results for well-defined searches, while generative models are better equipped to handle complex, open-ended queries and provide creative solutions such as a portion of the output for this column (and yes, I did use both ChatGPT and Google Search to help me with it. The ideas and searches are my own, but some of the generic text around Ricardo’s theory and computing needs were auto generated; I inserted explanatory clauses where needed).

Generative AI (where OpenAI-Microsoft?seems to be winning) and Web search, which Google dominates, are two different offerings, each with enough specialisation to live alongside one another for a long time.

Siddharth Pai has led over $20 billion in technology transactions. He is the founder of Siana Capital, a venture fund management company focused on deep science and tech in India. These are opinion pieces; the opinions expressed are the author's own and do not represent any entity.

This article first appeared in print in The Financial Express and online at www.financialexpress.com

For this and more, see: