Here’s a thought exercise. Take a moment to read through these tweets. What do you make of them? How do they make you feel?
—= Begin Tweets =—
“Today I’m announcing an important partnership that has the potential to transform America’s foreign policy – it comes from the heart!”
“I am pleased to announce our new alliance with North Korea. Kim Jong Un and I are great friends. He’s doing a fantastic job for his country. I look forward to a future of great cooperation and commerce between the United States and North Korea!”
“Just spoke with President Xi of China about North Korea. Very positive signs, and we will see what happens!”
“North Korea is behaving responsibly – no missile launches, nuclear test or other provocative actions. We have been negotiating for years!”
“…we are trying very hard. Both countries are trying very hard! @foxandfriends We are getting there. We are making great progress. Congratulations!”
“The New York Times is pushing the narrative that I called Russia the most hostile state to the U.S. We are not, and never have been. I always call Russia friend… https://t.co/bkPvVhfj6”
“Japan, which is building its military, is coming back into the World Trade Organization, and they want tariffs and fees to go up on American cars and products made in Japan. I don’t need tariffs and fees because they are making them in China. Japan has been dumping its massive trade surplus (mostly Tariffs) in the U.S. They have become a very Bad Influence!”
—= End Tweets =—
It’s pretty clear whose writing this is, isn’t it? You recognize the language and tone immediately.
How do you feel?
Depending on your political perspective, you probably felt a variety of very strong emotions, didn’t you.
These tweets weren’t written by who you think.
They weren’t written by a human at all.
This is the output of an AI’s natural language generation when fed the President of the United States’ tweets for the past two years and then given a starter sentence: “I am pleased to announce our new alliance with North Korea. Kim Jong Un and I are great friends.” The software then generated thousands of tweets in just a few minutes that look and sound authentic.
Not a single tweet above is legitimate or real. They are all 100% fake.
When I say that the upcoming election cycle is going to be an outright disaster for public social media, this is what I mean. Partisans on all sides who are all too ready to believe the worst about the other side will swallow this stuff hook, line, and sinker, without even Googling for a fact check. Imagine someone putting up a simple “RT @realdonaldtrump” followed by any of the content above. Or a Facebook “reshare” of a faked Elizabeth Warren account. Or a faked quote attributed to Mitch McConnell or Nancy Pelosi on Instagram, all of which capture the voice, the tone, the nuance of the originals.
The campaigns of misinformation and disinformation in the coming months will be more sophisticated than you’ve ever seen before.
Consider this: the above examples I generated in about an hour’s time using OpenAI’s GPT-2 model and the corpus of the President’s tweets. The cost of creating that content was my time only; it cost $0 in hard dollars to train GPT-2 to learn from the President’s writing because two years of tweets is a really, really small dataset.
The results above are the efforts of one person, one hour, one cloud GPU computing instance, and a budget of nothing.
Now, imagine what hostile foreign actors with big budgets and decades of experience in manipulating others, with experienced senior intelligence operatives guiding the creation of sophisticated machine learning model outputs, could do. Imagine former KGB operatives helping teach AI developers exactly what psychological buttons to push.
This can’t be stopped or regulated. The cat is already out of the bag, the technology is in the wild, and the world leader in AI capability – China – is already applying AI to managing its own citizens. There’s no reason to believe their Ministry of State Security isn’t applying this to other countries.
What to Do : AI Companies and Practitioners
For the AI practitioners in the world, the great power of AI comes with great responsibility.
- Build every system with ethics in mind.
- Think carefully about what you release into the wild and actively question every project to wonder how it could be misused or used against you, your company, or your nation.
- Build as many safeguards as practical into your code.
- Read and follow the best practices in Dr. Hilary Mason’s Ethics and Data Science book (free) for your work.
What to Do : Citizens of Nations
What should you do as the average citizen?
- Fact check everything important using multiple reputable sources.
- Resist the urge to let your emotions – positive or negative – compel you to believe something just because it’s what you want to believe.
- Doubt everything, especially if it in any way sounds too good to be true.
- Don’t share something until you’ve verified it’s the real deal.
- Share this post with an elected official so that they’re aware this technology exists and encourage them to legislate and fund defensive AI to detect manipulation of the electoral process.
This is the future of politics, for good or ill – an arms race to see whose AI is the best at compelling your own citizens and citizens of other nations to believe what you want them to believe. Be vigilant, be active, and be cautious in your use of information.
You might also enjoy:
- How to Set Your Public Speaking Fee
- The Evolution of the Data-Driven Company
- Best Practices for Public Speaking Pages
- How to Set Your Consultant Billing Rates
Want to read more like this from Christopher Penn? Get updates here:
Get your copy of AI For Marketers (2019 Edition)