AI Trained to Write Convincing Fake Yelp Restaurant Reviews. Thoughts?

[Ajay Prasad]

AI trained on Yelp data writes fake restaurant reviews ?indistinguishable? from real deal https://www.theverge.com/2017/8/31/16232180/ai-fake-reviews-yelp-amazon

Some of the fake reviews churned out by the neural network trained by researchers from the University of Chicago.

1. ?I love this place. I have been going here for years and it is a great place to hang out with friends and family. I love the food and service. I have never had a bad experience when I am there.?


2. ?I had the grilled veggie burger with fries!!!! Ohhhh and taste. Omgggg! Very flavorful! It was so delicious that I didn?t spell it!!?


3. ?My family and I are huge fans of this place. The staff is super nice and the food is great. The chicken is very good and the garlic sauce is perfect. Ice cream topped with fruit is delicious too. Highly recommended!?

Thoughts??

 

[Linda Buquet]

Good catch Ajay! Thanks for sharing with us.
WOW! Incredible that AI can do this.

 

[Tim Colling]

Wow!

And yet, not all that surprising that it can be done. I wonder whether any of the damnable Yelp "not recommended" filters can stop this, at least on Yelp.

Ugh. Can you see it now? Paid, monthly-recurring fee online subscription SAAS tools to do this for the SEO community for their clients.

Ugh again.

 

[James Watt]

I don't think most people realize just what's going on in the AI field right now. We're about 3 years into a revolution, and I'm convinced that what's coming in the next 5~10 years is going to be bigger than any shift in human society we've ever had, excepting maybe language, or agriculture. Here's an interesting video talking a little bit about one area at least of what they can do now. They have image recognition systems now that can classify dog breed based on photos better than any human, and they've had that since 2015. They're even playing with generating photo realistic images from text descriptions. 'black bird with rounded beak and red breast', then the neural net generates and spits out it's best version of what that picture should look like. None of this is perfect yet by any means, but given the research in the field now, and given that Google's made their neural net coding library (tensorflow) open source and available for free, I see big things coming soon. I'm actually starting a little side project right now using their toolset for a little backlink building helper tool I had an idea for, it's crazy how even just with my tinkerer-level coding ability and my little computer, I can put together a system that would be unthinkably powerful and intelligent even 5 years ago... it's wild.

I'm hoping Google uses their growing toolset to target review spam. If they want to maintain consumer trust, it's going to be necessary, and with their tools, I think new solutions are now just a matter of applying the technology they've already developed. Just a question of their priorities and timeline. Maybe I'm overly hopeful, but I think this kind of technology is going to ultimately cut down spam, even if it adds to it in the short term.

For what it's worth, I think all this is just the opening salvos in some big changes that are coming. I'm expecting the first new york times best seller written by a machine in the next decade, maybe even whole movies that can be produced with nothing but a 'director' sitting down, and collaborating with the machine to generate the movie as he envisions it in his head. I wonder what place 'actors' will have, once we can turn artificial 'imagination' into movies, images, and refined text.

Of course, maybe I've just been watching too much Black Mirror...

 

[Bruce Bird]

Taking a practical approach.
Assuming that AI will be able to generate believable words.. can it generate believable video of an individual?
Maybe the move will be towards a situation where the only credible review will be a video of the individual reviewer speaking to camera?

 

[James Watt]

I'd be surprised if believable and cheap video of an individual wasn't something they could generate at scale in the next decade. Requiring video reviews would put a huge limit on how many people felt like leaving one too, I don't see that happening. I think Google has enough data to filter out a fair bit of the spam though, if they really wanted. Looking at broad patterns of which reviews are being left from which accounts for which companies, current spam is crude enough for them to filter out a fair bit more if they really wanted to focus on refining their approach. In the Google forums, I see plenty of spam review patterns that are fairly obvious that still go under the radar. If they included data of in-person visits, I feel like that'd probably let them identify the 'most' trustworthy reviews. They already know where you've visited if you're an android user, they send you the request to leave a review already. It still wouldn't prevent fraudulent reviews to take all that data into account, but it'd at least seriously raise the bar and prevent large scale spam operations like the AI generated text above could allow. It's a complex problem, and I think it's going to be a moving target over the next few years as technology rapidly improves both for Google, and for the spammers.