Google, we have a problem

By Matt Redley

Ever been in the position where you’ve given the wrong answer to a question in front of a large audience? If so, spare a thought for Google. Yesterday, the tech titan showed off its new AI chatbot, Bard, which publicly answered a question incorrectly, leading to shares in Alphabet, Google’s parent company, falling by 7%. As a result, the company lost $100bn in the firm’s market share and was left with egg on its face.

In a promotional video for Bard, the chatbot was asked the question “What new discoveries from the James Webb space telescope (JWST) can I tell my nine-year old about?”. Bard answered that the JWST had been used to take the first ever pictures of planets outside the Earth’s solar system, also known as exoplanets.

Astro experts quickly noted that in fact, JWST did not take the very first image of the planet outside of our solar system, and that this was achieved by the European Very Large Telescope in 2004. Oops.

In what now looks like a rushed response to Microsoft’s ChatGPT tool, Google admitted that the error highlighted "the importance of a rigorous testing process, something that we're kicking off this week with our Trusted Tester programme".

ChatGPT has taken the internet by storm and has amassed 100 million active users in two months, making it the fastest growing consumer application in history. The tool uses natural language processing, which allows a computer program to understand human language as it is spoken and written. Google has been under pressure to respond quickly to Microsoft on this front, sounding a ‘Code Red’ (essentially the panic button) to address the threat posed by ChatGPT, whilst parachuting Alphabet’s founders back in to respond.

On the same day as Google’s misstep, Microsoft addressed a crowd of 100 journalists and industry analysts to demonstrate its unveiling of Prometheus, a specialised version of the AI tool that powers ChatGPT. The company announced that it will be integrated into its Bing search engine, which has always played second fiddle to Google’s. The audience was shown how the tool will be able to instantly plan a family’s travel itinerary for a holiday in Mexico City or share how to substitute an egg in a recipe, without having to visit a website.

As the AI arms race hots up, the voices around regulation are becoming louder as tools become rolled out to the public with little to no oversight for the sake of commercial interest. Back in 2018, Elon Musk said that the rapid advances in AI ‘scare the hell out of me’, and that ‘AI is far more dangerous than nukes, so why do we have no regulatory oversight?’.

These voices should be given credence in the area of natural language processing alone, given the fact that the essence of Google / Microsoft’s natural language processing AI technology is not well understood by either party, or even researchers who pioneered the underlying models. As mentioned by Semafor, Prometheus appears to have multiple personalities, given that the same question will produce varying answers each time you open up a browser window and begin a new chat.

Semafor note that the reason for this isn’t completely understood, but that it could be something to do with the way the work is divided up among the hundreds of thousands of central processing units searching for the answer. Essentially, each new opened chat is with a new, unique entity, and means that results are currently neither predictable or verifiable.

Although a lot is unknown, it is clear that Microsoft has fundamentally out-manoeuvred competitors and left them firmly on the ground.