In 2017, I foresaw a future dominated by machine-driven communications (MADCOMs), where artificial intelligence (AI) enabled bots would flood our social spaces online. They would seek out humans, selling them products, ideas, and propaganda, using psychometric profiling tools that enable precision targeted marketing. Extrapolating, the machines would be looking for humans, but might not be able to determine whether target accounts are actually machine-driven. So inevitably, machines would be talking to machines, the machines would talk back, and much of the social internet would be overwhelmed with machines talking to machines.
This is the MADCOM future. However, this analysis may have been an underestimation of the actual implications because I didn’t consider multimodality. We are likely on the brink of an era characterized by a Distributed Denial of Everything (DDOE), where the inundation of AI bots in our digital and physical realms will significantly challenge our attention economy. (Want to see DDOE in action. Follow this footnote: 1
Attention, representing our time, is undeniably our most valuable and scarce asset. The advent of the internet, social media, and various digital tools has already escalated the battle for our attention. Yet, the emergence of autonomous, human-like bots threatens to intensify this battle further.
We are about to be flooded with AI-driven multimodal bots that will infiltrate every facet of our lives, extending their reach to phone calls, text messages, and instant messaging platforms, essentially any digital medium that enables them to impact a human being. Spam is going multimodal.
Here are some examples:
Restaurants: you’re a restaurant owner. Currently you field calls from customers seeking reservations or information, vendors looking to sell you things, sites like Google Maps or Tablecheck confirming your hours and seeking your business, etc. Most of these inquiries are driven by a human being at the other end. However, the technology already exists for AI powered bots to drive many of these engagements. This enables greater speed and frequency of these inquiries at much lower cost than using a human. Filtering out the desired communications from the chaff will be critical, otherwise you won’t have time to cook the food and run your restaurant.
Politics: hate robocalls? Get ready for botcalls where an AI that sounds like the candidate or some other celebrity calls you but actually engages you in a conversation designed to persuade you. We saw the first glimpse of this when a political consultant used a deepfake synthesis of President Biden’s voice in a robocall during the New Hampshire primary. The FCC banned deepfake robocalls of this type, but I suspect a broad ban won’t survive litigation because there are legitimate free speech reasons why candidates will want to use these tools to get out the vote. (A ban on unauthorized deepfakes and/or voter suppression using AI seems more tailored to the problem). Expect that every political candidate will deploy a set of AI bots that are trained on your publicly available data and will effectively tailor customized persuasive content to your personality and political beliefs. These bots will call you, email you, and find you on social media. I think there’s a better than 50% chance that within a decade, we will have the first door-knocking political bot that goes house to house trying to get out the vote.
Corporations: Currently, engaging with a corporation's customer service often entails navigating cumbersome phone trees to address even minor concerns. (Getting an actual human on the line is a great way to waste 20 minutes of your life). Soon, customers will have bots that will intermediate these initial transactions. Initially they will turn the interaction over to a person when an actual customer service agent connects to the call. But eventually, the entire interaction will be mediated by an AI bot. These bots, acting on behalf of customers, are likely to be embraced for their utility, making it unlikely for companies to restrict their use. However, not every customer has good intentions. People will create bots that are very good at complaining loudly, insisting on discounts, refunds, etc. and they can be deployed en masse, siphoning off corporate resources.
Governments: Federal, state, and local governments will face the same challenges as corporations in managing both legitimate inquiries from the public but also mitigating harms from malicious actors. One case is illustrative of the coming problem. Soliciting public comment on proposed rules is core to democratic governance. In 2017 the FCC solicited public comment about its proposed changes to net neutrality rules. The FCC was overwhelmed with ~22 million comments, ~18 million of which were later found to be fake. Around 9 million comments used fake identities but 8.5 million of these fakes used actual humans who were “incentivized” to comment through gift cards and sweepstakes entries. The stakes are high in governance, with billions or trillions at stake. So there is a large incentive to use AI tools to slant results. Expect government officials across the board to receive calls, emails, Zoom bombs (1,2), etc. from special interest groups and trolls, sapping their attention, efficiency and morale for public service.
How do we fix DDOE? This tidal wave of machine-driven communication is coming and it presents two key problems: 1) How do we preserve the precious attention of the human targets of these MADCOMs? and 2) How do organizations/governments filter legitimate communications from malicious communications or propaganda?
The answer to the Distributed Denial of Everything is…more machine-driven communication tools: mirroring the recommendations in my MADCOMs paper, the only solution to MADCOMs trying to steal attention is…more MADCOMs. We will all need our own personalized AI tools for intermediating our interactions with the world. Organizations will increasingly use them to filter their interactions. These won’t be too foreign as we’ve had spam filters for decades, we use AIs to filter our focused/prioritized inboxes, and we already have algorithms picking content for us in our feeds. These tools need to become personalized so they learn our wants and needs and can help us determine the people, content, and bots we want to engage with and filter out those we need to ignore.
Inevitably, our bots will engage directly with other bots and conduct transactions entirely without human interaction. (My worry is that many humans will actually prefer interacting with AIs, will get addicted to them, and will eventually stop making babies as I detailed in this article). Along the way there will be growing pains and mistakes (didn’t mean to order ten bacon and mayonnaise pizzas? Your bot thought you would enjoy them.) But eventually these tools will be indispensable elements of our lives, not just for the unwanted interactions they filter, but for breaking us out of our filter bubbles, enabling conversations we never would have had, and saving us from a lot of the mediocre grunt work that even the best of us have to manage.
And in the best possible world, humans won’t become addicted to these tools, fall in love with embodied AIs, and stop reproducing but instead these tools will be used to promote better, healthier, and more satisfying human interactions. We have an epidemic of loneliness in the world. I think AI can and should be focused on building human to human connections. But that is a topic for a future article.
-
Thank you for reading. If you enjoyed this essay, please subscribe and share with others.
Opinions in this post are my personal opinion and do not necessarily represent the views of the Department of State or the US government.
Ethan Mollick writes about how Amazon books are getting DDOEd by AI generated garbage publications: https://x.com/emollick/status/1766526577444073513
Patrick McKenzie writes about an AI powered app that sends “please remove my email from your database” messages to everyone, which creates substantial downstream work for companies required to delete your email. https://x.com/patio11/status/1758513571984478414
More examples en route from the future…