I started using artificial intelligence in my writing work almost by accident.
I was writing a case study for a client about a Wolt-competitive delivery service based in Umm el-Fahm called HAAT. Some of the interviews we conducted were in Arabic. We had a HAAT employee help ask the questions, but when we got the video back from the interviews, we didn’t feel comfortable asking the translator to pore over hours of tape and write down everything she heard.
So, I ran the audio through a Mac-based app called Whisper Transcription. Whisper transcribed it into both Arabic and English text. It even included the time code so we could easily add subtitles to the video.
Whisper Transcription uses artificial intelligence to work its magic. I have extracted salient points from dozens of podcasts this way. Compared to how I used to do it – painstakingly playing the audio while typing, stopping and scrolling whenever the audio got ahead of me – this has saved me innumerable hours of annoying grunt work.
The next step in my AI evolution was not a work-related issue but a personal one.
When COVID-19 hit, I moved my therapy sessions from in person to Zoom. I would take notes either after the session or during the session itself.
Then I discovered Read.ai. A more fully featured AI transcription tool than Whisper, Read.ai has a plug-in for Zoom that automatically launches and captures any audio. When you’re finished, Read.ai sends you a link to its transcription online.
Read.ai does more than transcribe. It also summarizes the conversation, presents a list of action items, and even grades you on how personable and attentive it felt you were during the session. I now take my hands off the keyboard during therapy and highlight the text later.
Whisper and Read pale in comparison to the incredible utility that comes with Google’s new NotebookLM tool.
Utilizing Google’s Gemini 1.5 Pro AI, NotebookLM (the LM is short for “large language model”) is astounding. Aimed at helping writers make sense of their research, you simply upload your documents, and, like Read.ai, the AI will summarize them for you.
That’s just the start, though.
NotebookLM suggests query topics, or you can type in your own. I gave NotebookLM a PDF of my book Totaled. “What were the key factors that led to the failure of Better Place, according to the author?” it asked. “Evaluate the management style of Shai Agassi as to the success or failure of Better Place.”
Which got me concerned: Will our content-creating jobs some day be threatened by AI?
I didn’t think so – until I discovered a NotebookLM feature that blew me away. The app can generate a 10-minute podcast based on your material. Two chirpy NPR-style narrators banter back and forth, talking over each other, disagreeing, then coming together as they discuss the merits of your work.
The result is unnervingly realistic. The narrators don’t sound like they are computer-generated (other than a few grammatical gaffes here and there). They can, at times, be quite smart, unlocking insights I didn’t even realize I’d written.
For example, I’ve been noodling for a few years now on ideas for a novel but have been having trouble figuring out the main plot points. I plugged my notes into NotebookLM. The podcast hosts honed in on what the main thrust of the book should be (spoiler alert: It wasn’t necessarily what I was thinking) while surfacing some illogical contradictions.
I’ve used AI in various other ways.
- For a book I’ve been working on with a client, we wanted to include a series of graphics that would have a consistent look and feel. Dall-e, an AI image-generation program, did a yeoman’s job.
- I’m a huge fan of Peter Gabriel-era Genesis. So, I asked an app called Songer to write me a song in the band’s early 1970s style. The song sounded more like Rush or Black Sabbath than Genesis. Still, it’s only a matter of time before musicians join the list of artists disintermediated by AI.
- When I was working on the audiobook version of Totaled, I considered using an AI tool such as Revoicer or Podcastle to turn my text into narration. They did pretty well, especially with a British-accented voice-over; but when I uploaded some samples of my own voice to “clone,” the result came out choppy and distorted. Plus, the AI had trouble enunciating Hebrew words and names. I stuck with the old-fashioned way and recorded it myself.
How Israel is using AI
Beyond my personal experience, Israeli hi-tech is fast becoming an AI leader. Among the companies worth tracking:
- Immuneai is mapping the entire human immune system to explain why people react differently to viruses, cancer, autoimmune, and degenerative diseases.
- IdentifAI has developed a non-invasive prenatal blood test that can detect problems with the fetus with a low level of false positives.
- RespirAI has created an algorithm that uses sensor data from smartwatches to predict when respiratory deterioration is imminent.
- Clarity.ai is tackling what CEO Michael Matias calls “the invasion of deceit” – specifically, deepfakes that have confused audiences as to whether Taylor Swift did that or whether Donald Trump said that (he probably did).
- Visionary.ai has partnered with Qualcomm to bring better pictures and videos in low light to an Android smartphone near you.
DO I believe these newfangled AI tools will send the creative class to work at a local McDonald’s? No. That doesn’t mean the day won’t come when computers surpass human cognition. Elon Musk is betting his latest company, Neuralink, on just that. Ray Kurzweil predicts in his 2024 book The Singularity Is Nearer that AI will reach human intelligence by 2029 and merge with humans by 2045.
To see what’s coming next, go to https://bit.ly/3BtO310, where I asked NotebookLM to generate an AI podcast version of this column.
The writer’s book Totaled: The Billion-Dollar Crash of the Startup that Took on Big Auto, Big Oil and the World has been published as an audiobook. Available on Amazon and other online booksellers. brianblum.com