Algorithmic nudging via AI is an emerging practice - one that deserves ethical scrutiny. The term "nudge," while new in this context, has ties to manipulative media tactics. It's a potent new tool in the art of persuasion.
Manipulation through media is hardly new, but a nudge is a little different. The term "nudge" has been circulating around behavioral psychologists for a while, but the seminal book Nudge: Improving Decisions About Health, Wealth and Happiness, by Nobel Prize Laureate Richard Thaler and Cass Sunstein formalized the ways policymakers and other intermediaries can prod behavioral changes. Their definition of a nudge is "any aspect of the choice architecture that alters people's behavior predictably without forbidding any options or significantly changing their economic incentives."
Those who see the danger in this type of nudge refer to it as "libertarian paternalism." In a technical context, a developer organization assembles an environment in which people make decisions - and in the authors' term, the "choice architect" may predict individuals' prospective behavior and influence them to act in a preferred way. Nevertheless, individuals are free to opt-out. The presumption of freedom of choice is implied, but the nature of the nudge is designed otherwise.
There is lively discussion and controversies about nudging. Commercial enterprises promote the positive potential of nudges but shy away from discussing libertarian paternalism that does not encourage welfare but rather infringes on it: so-called "evil nudges."
Websites soliciting contributions for charities or political organizations often have default settings already engaged and a further option to continue to donate at specific intervals. These options allow for "opt-out," but the pre-selected options are a gentle nudge to act. More sophisticated techniques increase the donor's satisfaction for giving through rewards such as t-shirts, chakachas or recognition.
Subscriptions to information and news sources will offer a "$1 per month" subscription fee. Still, choosing the option automatically enrolls for an annual fee of $14/month billed to your credit card.
Countless innocent nudges are effective. In the UK and the Netherlands, some grocery stores removed junk food from the cashier line and replaced it with healthy alternatives. It was effective in increasing sales of fresh fruit and produce. In addition, the profit margins for food marked "organics" can be 3-5x as high as candy bars.
A reportedly effective nudge in the Netherlands was placing the image of a fly in urinals, which reduced the amount of urine on the floor. The mechanism of action should be obvious.
AI and algorithmic nudging
Recommendation engines nudges are used at many levels in AI algorithms, for example, recommender systems, and their consequences are still being investigated. Two articles that appeared in Minds & Machines in 2018 and Can Machines Read Our Minds? delve into the relation between nudges and Artificial Intelligence, explaining how personalized targeting algorithms can use persuasion and psychometrics to influence individual and collective behavior, sometimes also in unintended ways.
In 2020 an article in AI & Society, "On Social Machines in Algorithmic Regulation," addressed the use of this technology. The gist of the paper identified convergent social and technical trends leading towards social regulation by algorithms and discussed the possible social, political, and ethical consequences of taking this path.
"Algorithmic Nudging" is a term that first appeared in a 2021 article in the Harvard Business Review, in Algorithmic Management:
Companies are increasingly using algorithms to manage and control individuals not by force, but rather by nudging them into desirable behavior - in other words, learning from their personalized data and altering their choices in some subtle way.
With continuing advances in AI and machine learning, algorithmic nudging is much more potent than its non-algorithmic counterpart. Developing personalized models at scale is a well-honed skill with so much digital data about peoples' behavioral patterns. Nudge algorithms can adapt in real-time, boosting the results desired by the developers. The current political pattern of radicalism and division is a result of targeting and algorithmic judging. So despite the academic work and more or less neutral position about nudging by Thaler and Sunstein, nudging emerges as a dangerous tool in need of regulation and statutory rules. The question is, should the law recognize liability for evil nudges that result in bad faith influence?
Effective "evil" nudges in a technologically connected environment can scale in a way that brick-and-mortar cannot. Online intermediaries influence decisions through website design and promote behavioral change among internet users. AI enables vast and largely undetected hyperinfluence. Drawing on network theory, psychology, marketing, and information systems, nudges influence the process of information diffusion in digital networks. Nudging intermediaries can amplify the severity of speech-related harm. The law should respond to "evil nudges," with differential guidelines for deciding cases of intermediary liability.
There is a difference between nudging a particular behavior and compelling a specific choice. A good nudge may be considered to encourage a particular choice but is still:
- Transparent - make the nudge clear and obvious, not hiding costs / other options.
- The choice is retained - with the consumer able to make the final choice.
- There is good reason to believe that the nudge is warranted, e.g., substantial health costs of smoking/overeating sugar.
- Nudges may not be enough. To reduce smoking rates, we need policies that tackle core problems. This may require - higher taxes, restrictions on places where you can smoke.
Nudge is just a word for something that we all do. Breeding hate, hysteria, and violence through disinformation, at scale, however, is more than a simple nudge.