I did not expect “dread” to be one of the feelings I felt during this week’s Google I/O Keynote. Google has spent the last two decades developing products that have made my life easier. This week it did the opposite, demonstrating a way of making communication with others worse.
This year’s Google I/O was centered around artificial intelligence. One of the highlights was prompt-based text creation, , particularly in Google Office Suite. The keynote included several demos of a “Help me Write” system, which transforms Gmail prompts into more polished paragraphs. But a more polished does not automatically mean a better.
A good writer can make us smile with a clever turn of phrase. It can reveal something about a person’s thoughts and feelings. It can also provide useful information about the world. There’s no reason why AI shouldn’t be a part of the solution. I’ve seen authors create unexpected text, or summarize data readable. I’m eager to see how people use it.
Google’s ideas onstage were impressive, but they served no purpose at all in the core of writing. The examples in Help me write highlight the worst aspects of “professional” communication. These are bland, bloated boilerplates that turn a human prompt to something that says less with more words.
Look at an example. In the demo, the presenter automatically generates a job description (in the early part of the keynote) for a textile designer position
This is the version that you are expected to use when applying for a job. Written by a person, it shows you are committed enough to write a long response and can hit certain cultural and class signifiers. AI-generated responses don’t accomplish either of these goals. The AI-generated response was generated almost instantly using a predictive text program that only requires a minimal level of English language proficiency. However, this system is limited to formulaic writing and can’t replace human work for many real-world scenarios. It merely creates a more formalized version of the original prompt, which is only useful until people assume it was created by AI.
Even worse, AI-generated text reinforces the notion that is required for writing, even if it’s not a skill needed to do the job. I’ve read stories of people with dyslexia who use ChatGPT to create text that’s — , as a Washington Post report puts it – “unfailingly polite and professional.” But there is a simpler way to do this: be willing to accept broader variations in the ways people communicate. I’m not against anyone who uses AI to write in a way that meets arbitrary standards. But at a larger level, this is a linguistic arms-race towards a duller future.
Try this message of friendly congratulations from Google Android Engineering VP Dave Burke, to Devices and Services SVP Rick Osterloh.
This is a joke, I assume. I’m familiar with the way executives write personal emails and I know that they are perfectly comfortable with missives of one sentence punctuation free. It’s still worse! The message tells me little about the sender’s voice. It takes more time to read. The sender sounds like a robot.
You need to examine an email that was sent at the beginning of the keynote in order to understand its practical value. The core of this generation is really cool — it takes an extremely simple prompt for a refund request from an airline and fills in concrete details taken directly from your inbox like the flight number and date. Then, it adds a few paragraphs with irritating prose to embellish the facts.
It’s not clear that these angry claims are true. It’s a bureaucratic display, convincing some poor customer service representative you are mad enough to type all that text out and that you will be a nuisance until you receive your money. This idea was better presented on Tumblr. In a post on AI, described how to use ChatGPT in a high-tech version Brazil’s form 27B/6 gambit. This generated a fake-legal letter threatening a landlord for breaking housing laws. As a long-time tenant myself, I commend them.
This stuff works in the short transition period when generative text hasn’t been widely adopted, and readers will assume that it is written by a person. The illusion is gone once you realize it’s an automated system. The world is full of longer communications, which are less well-written and harder to understand. I would rather hire someone with an honest “8 year exp” than a cover-letter full of automated prose.
Google’s best example of Help me Write was to simply relay information. Its AI was able, in an email regarding a potluck email, to read a document that contained a list with dishes people signed up to bring and summarize the list into a single line. This saves both writers and readers from having to copy-paste a list of items or clicking on another tab. Its value is not based on pretending to be a human. If Google has its say, this trick won’t work for very long.
