A.I. Bots Can’t Report This Column. But They Can Improve It.

Plagiarizing essays. Cheating on tests. Spreading misinformation. The advent of new, powerful artificial intelligence tools like ChatGPT has sparked concerns about the unintended consequences of the technology and debate over the ethical implications of its use.

Despite concerns, A.I. assistants seem here to stay. After ChatGPT’s splashy debut late last year, Microsoft invested $10 billion in OpenAI, the lab behind the chatbot. The threat from a rival forced Google to regroup. And the media site BuzzFeed recently announced that it would use A.I. to write online quizzes and other content.

Given the growing influence of this technology, it’s time to focus on how we can start reaping the benefits in a responsible way. Many A.I. experts and computer scientists agree that these tools can provide a major perk that does no harm: editing our writing.

In this context, we would still be using our skills and brainpower to do the hard work of making sense of information and coming up with new ideas, while turning to A.I. for some light assistance, such as making a sentence sound more conversational, sprinkling in some historical context or even adding a joke.

For several days, I tested Wordtune Spices and Rytr, two A.I. writing assistants released by start-ups in the last two years, and compared them with ChatGPT. I’ll go over some examples that highlight the strengths and weaknesses of each of these three tools.

Test 1: Rewriting a Sentence

To use Wordtune Spices, which the Israeli start-up AI21 Labs released last month, you insert text into a box, highlight the sentences you want edited and then click on options to make improvements. Among its best uses during my recent test were quick rewrites of sentences.

For example, I took this sentence from my column last week about email addresses and privacy:

Then I clicked the “Rewrite” button. Wordtune Spices shot back a list of options for revisions of the sentence. I liked this one best:

My original sentence included a significant detail to me — that we hand over our email addresses without hesitation — but the tool trimmed a few words to cut to the chase and summarize the heart of the problem. Not bad.

Test 2: Catching Grammatical Errors

Rytr, an A.I. writing assistant released in 2021 by a company of the same name, was designed to streamline the creation of blog posts and marketing copy. It was proficient at catching grammatical errors (as, similarly, word-processing tools like Microsoft Word have been doing for decades).

In my test, I took this paragraph, which has intentional errors, from an educational website that provides editing exercises:

Then in Rytr, I highlighted the paragraph and clicked “Improve.” It gave me this:

The tool automatically rewrote an error-free version of the paragraph. This could be extremely handy as a study tool or for non-native English speakers.

Test 3: Adding Context

Wordtune Spices also excelled at fleshing out paragraphs with additional context. By pressing the + button, you can add color and background information to a piece of text by inserting historical data, a joke and an example to prove your point.

For example, I took this paragraph from my column last week:

In Wordtune Spices, I clicked the + button to add an example, a historical fact and a joke. (Changes in bold.)

After doing some research, I was impressed to learn that 1978 was indeed when email was first used for marketing. The example of using email receipts to serve targeted ads was also accurate. The joke wasn’t very funny, but it lightened things up.

Test 4: Ending This Column

Last, I pasted this entire column into each writing assistant to see if it could come up with a kicker, or a clever conclusion to wrap this all up.

ChatGPT was the only one of the three that was able to read the entire text to generate a plausible conclusion:

Wordtune Spices produced an unusable takeaway:

Rytr did, um, something:

AI21 Labs, the Israeli start-up that developed Wordtune Spices, said current A.I. writing technology required more guidance from the user than the prompt I had given the tools. Rytr said its users could use a feedback tool to train its A.I. in the event that something went wrong. OpenAI declined to comment.

All this shows that artificial intelligence can be an effective tool for improving our work. I was surprised by some of the results — in particular, that Wordtune Spices could add accurate and relevant background information. I may occasionally use the tool for suggestions on fleshing out paragraphs with some historical data, though I will then check the facts before publishing them.

And in general, the A.I. bots were useful for sharpening prose and cleaning up clunky, ungrammatical sentences.

What the bots couldn’t do, however, was the research or reporting to show their strengths and weaknesses. That requires thought.