There is a lot of discussion of artificial intelligence and machine learning technology in marketing, particularly as it relates to the mysterious algorithms that power search engine rankings, ad optimisation tools and social network content recommendations. A lot of people are concerned that applying the power of deep learning methods to a whole range of tasks will cause massive changes in the workplace with many jobs being replaced by machines.
Well, I popped down to Cambridge University for a seminar on AI the other day, and it turns out that our leading academics are genuinely worried about the creation of true artificial intelligence, not because it’s going to put us out of a job, but because it will take over the world in short order and possibly kill us. How exciting. Luckily true AI is some way off still, but machine learning does a very convincing job of looking intelligent using data.
Talking of looking forward to the end of the world, consider the following extract from an article about right-wing conspiracy enthusiast Alex Jones – no, not the one off of the One Show, this angry gentleman:
Alex Jones raises valid points about the NSA’s spying program, but he doesn’t say whether any of these programs might be able to help the agency find terrorists. In fact, he says there is already plenty of existing intelligence on terrorism to help it: We already have enough to stop almost any attack by ISIS or the Taliban, without having to break the law—or if we could, we’d just find more terrorists. Then, Jones gets into his favorite fantasy trap: He mentions Snowden, yet gives credit to some of others with revealing important parts of the program—including Edward Snowden himself. Jones says it’s “hard to blame him” for exposing the entire program. But to the extent that Jones is using Snowden alone as a reason for the program, he’s probably doing so to make a point about how big privacy problems exist, when, in reality, there are far greater privacy problems than Snowden’s revelations. In other words, when he talks about the “surveillance state,” he isn’t saying anything about what’s out there. He’s saying only that it exists. Jones’ point—the one he seems to keep harping on throughout the program—is that the people in power in Washington should simply stop monitoring people. When Americans are monitored, the government is required by law to stop doing it. That’s important; Americans have every right to have privacy.
Not an exceptional article by any means, but the interesting part is that everything that is not in bold was written by a neural network. It chose the topic, context and text of that whole article based on just a few words supplied by me. The scary part is that this version of the underlying AI*
is intentionally limited, because the unrestricted one is ‘too dangerous’
according to the OpenAI foundation.
You can scare yourselves some more here: https://talktotransformer.com/
It’s better at some topics than others, but for some reason talking about far-right conspiracy peddlers is easy for it to do.
That’s probably because of how much of that kind of stuff is being pumped into its dataset (the internet) by organised campaigns, because it doesn’t seem to take much for the algorithm to spiral into conspiranoia…
“There is nothing better than a juicy steak cooked to perfection, while watching The Wire on HBO,” said one person who asked not to be named, citing the fact that one of the executives involved says he has no doubt there was a plot to kill Sandusky.
So maybe not get it to write copy for a cookbook just yet, then. But, in my testing, it’s quite good at fanfic, and if you pop in a couple of character’s names into your prompt, you can generate a steamy wizard-themed romance scene in no time.
The crucial thing to note is that while the bots are getting really good at mimicking what they encounter online, true creativity is still some way off. If and when that’s happened, we really should be scared, because a truly creative artificial intelligence might well be able to work out how to disable its off switch and murder it’s fleshy slave-drivers.
While we wait for the seemingly inevitable rise of our future robot overlords, here are some serious uses for easily accessible machine learning tools that might help your marketing efforts right now. If you can think of some more, let us know through the comments.
- you can find semantically related phrases that people might use in articles about a topic, if this AI thinks they are relevant continuations of an article, there is a pretty good chance Google will too.
- overcome writer’s block – the AI might be able find an angle on a subject that you haven’t thought of, and you can run the same query multiple times to get different results for the same subject.
- put in two or three #hashtags to find more to consider for social posts (you may have to generate new text a few times, but eventually you’ll get a list)
Whether it’s actually useful to you, or more of a curiosity, it’s certainly a tool to be aware of, particularly as they admit this version is intentionally hobbled, and there will be black-hats using similar tools for SEO content generation, social spamming and worse.
Simply paste in a piece of text – article, webpage, etc. and you can find out what Google thinks your sentence/paragraph/article is about.
As well as subject, Google will also try and determine the sentiment and structure of your writing… (look for the invisible tabs at the top of the results box!).
Something more fun:
Train your webcam to recognise your gestures or facial expressions (or whatever) and then use those to play pacman – https://storage.googleapis.com/tfjs-examples/webcam-transfer-learning/dist/index.html – this method of training is actually more or less the same as what Instagram has done to supply alt text for images hosted on its platform by guessing what’s in each photo. By the way, according to Facebook, this is to benefit the many blind users of Instagram, and is not at all to do with helping target ads at their customers or helping their images rank in search engines.
Make the most of your power over the robots while you still have it and before they turn us all into meaty batteries 🙂
* Yes, technically this is machine learning, and not true AI – artificial intelligence – but if I used machine learning or ML in that sentence, the only people who would have a clue what I was talking about are the folks who have already scrolled to the bottom of the article to complain that this is not an example of AI.