A new app called MindMeld listens to you talk and tries to butt in with helpful information. There’s a clever idea at the core of that, but I found that it often misunderstood words and failed to get the gist of a conversation, which made it more distracting than beneficial.
Imagine that you’re on the phone with a friend and talking about a new restaurant: one or both of you might go online to find reviews. MindMeld tries to do that for you automatically. Or if you are talking about buying a birthday present for under $15, it will
pull up Web pages with inexpensive gift ideas.
MindMeld ($4, available for the iPad only, though creator Expect Labs plans to roll it out for the iPhone and Android devices, too) is organized around “conversations,” which are basically the app’s term for chat rooms. You can create your own conversations that are open to everyone, to your friends, or to people you invite—up to eight people can participate in a conversation at a time. In order to get the app to start listening and analyzing, you can say “Okay, MindMeld,” or just tap a little microphone icon in the upper left corner of the screen. If you stop talking for a little while, or tap the icon, it should stop recording.
So if I’m talking about how I want pizza tonight in San Francisco, it will transcribe my speech (“I’d like to go out for pizza tonight in San Francisco”) in a constantly updating stream on the left side of the screen. The rest of the display will fill up with a variety of pizza-related posts from Yelp, Chow.com, Facebook, and various food blogs. Since it’s constantly listening, it does its best to transcribe the conversation in real time, though I always noticed a delay.
This concept of an always-on digital butler is known as anticipatory computing, and it’s also at the root of services such as Google Now, which tries to bring up information that your calendar or your whereabouts make it think you might need. If these applications can also draw from your conversations, they could form the basis of novel services (see “Why Big Companies Are Investing in a Service That Listens to Phone Calls”). But it will take a while to bring MindMeld from curiosity to utility, because the challenges in voice recognition are so steep.
This was apparent when two people tried the app with me on their iPads at separate times.
The first conversation was with my sister-in-law, in which we talked mainly about her family’s upcoming vacation. MindMeld brought up some helpful results about the destinations she mentioned, Hawaii and Palm Springs, but it didn’t seem to understand certain words and phrases. In one particularly hilarious misunderstanding, MindMeld thought I said “Yachimun superintendent of the weird shipments of human vibrator.” (According to the Internet, “yachimun” means “pottery” in the dialect of Okinawa, Japan—neat, but not germane to the conversation.)
Another conversation, with an MIT Technology Review colleague, was more error-filled. We covered topics including Burmese restaurants in Boston, Harry Potter, and holiday plans, but MindMeld had trouble keeping up with the conversation and understanding what we were saying (at one point, it transcribed “lamb tagine” as “lamb has-been,” and a mention of Hawaii pulled up a list of Saved by the Bell episodes).
I had better results when I used MindMeld by myself, as a tool for brainstorming aloud. Since I’m in the midst of wedding planning, I talked about where the wedding will be and how I might find florists, hair stylists, and decoration ideas. MindMeld delivered useful results for vendors that I might not have found with basic Web searches of my own. It was simple to save the most promising results for future reference by holding my finger down on one of them and swiping to the right.
Expect Labs offers a MindMeld developer platform, so as the service improves it could be added to voice-call apps like Skype as a sort of “brainstorm” feature or used within companies as a way to bring up internal documents during meetings. Until then, I’ll keep Googling and checking Yelp before going out to dinner.
This new data poisoning tool lets artists fight back against generative AI
The tool, called Nightshade, messes up training data in ways that could cause serious damage to image-generating AI models.
The Biggest Questions: What is death?
New neuroscience is challenging our understanding of the dying process—bringing opportunities for the living.
Rogue superintelligence and merging with machines: Inside the mind of OpenAI’s chief scientist
An exclusive conversation with Ilya Sutskever on his fears for the future of AI and why they’ve made him change the focus of his life’s work.
How to fix the internet
If we want online discourse to improve, we need to move beyond the big platforms.
Get the latest updates from
MIT Technology Review
Discover special offers, top stories, upcoming events, and more.