The massive cost of AI-powered search engines
The AI-powered language models Google and Microsoft are incorporating into their search engines will come at a massive cost to the companies, as the technology requires much more computing resources per query than traditional search engines, Ars Technica reports.
Morgan Stanley analysts estimate Google could face a $6 billion (R110 billion) expense increase “if ChatGPT-like AI were to handle half the queries it receives with 50-word answers”.
The higher costs come from the increased computing power required to run AI-powered search engines. It requires powerful processing capabilities and depends on billions of dollars worth of chips.
Processing AI-powered searches is known as inference, where the answer to a question is inferred based on prior training and involves a neural network modelled on the human brain’s biology.
On the other hand, traditional search engines involve building an index of the web. When a query is submitted, index entries are scanned, ranked, and categorised to show the most relevant entries in search results.
The higher search costs will likely be a more significant problem for Google than Microsoft, as the Alphabet-owned company holds around 93% of the worldwide search market.
Microsoft holds only about 3% of the global search market.
It is also a primary business for Google, and with it needing to process around 8.5 billion searches a day, its per-search costs accumulate rapidly.
Alphabet’s John Hennessy told Reuters that Google wants to drive down the cost of AI-powered search, which he described as “a couple year problem at worst”.
Google announced its ChatGPT rival — Bard — on Monday, 6 February 2023, opening it up to trusted testers and said it is preparing the service for the public “in the coming weeks”.
Its share price took a beating after announcing the tool, when an advertisement for the language model showed Bard giving an inaccurate answer to the question of which telescope took the first photo of an exoplanet.
Bard said it was the newly-launched James Webb Space Telescope in 2022. However, according to Nasa it was the terrestrial Very Large Telescope (VLT) in 2004 that claimed the honour.
Loading ...