We Need Green Digital Public Infrastructure

It’s time to train artificial intelligence in the public interest.

Posted Feb 09, 2020

Artificial intelligence, we are told, will be developed by people with genuine intelligence. And they will make sure that real surgeons train robot surgeons, real drivers will train AI vehicles, and real billionaires train your bank’s security system. What’s there to worry about?

Anyone connected to the internet via fixed or mobile computing devices has interacted with AI. The most common form is known as a chatbot (for chat robot) or personal digital assistant. Nearly three-quarters of Americans now expect to encounter a chatbot on websites, though only about 13 percent like interacting with them. People’s preference for human contact is heightened when they experience stressful situations, such as healthcare, bills, tech support, and money.  

Nevertheless, we’ve become accustomed to the ubiquitous chatbots. A chatbot is “trained” by natural language processing (NLP), a component of AI, to understand, decipher, and generate human language. Through voice queries or typed requests, chatbots can “talk” to us while we shop, look for recommendations, or seek solutions to problems with customer service. They are present in messaging and social media platforms, smartphone applications for e-commerce, maps, and news, as well as online banking and payment apps. They exist to cut costs by displacing human workers.

The chatbot market is big business. Valued at $2.6 billion, prognosticators say it could grow by as much as 30 percent over the next four years. Meanwhile, the overall corporate value of AI patents and intellectual property could skyrocket by $4 trillion in the next couple of years, with cloud platforms “quickly becoming AI and machine learning IP and patent foundries.”

Beyond the business hype and promised utopias, there is increasing concern with the potential of AI to harm digitally mediated forms of social communication: frisky chatbots making dating apps more addictive, deceptive, and unsafe; advanced NLP algorithm performance-enhancing the believability of fake news and other propaganda; the creation of so-called deepfakes, manipulating audiovisual content to deceive viewers into believing falsified representations of public figures; and facial recognition algorithms instantly scanning billions of photos stored in databases around the world, exemplified by Clearview AI’s shameless violations of data protection laws and human rights.

If AI has been cast as an unstoppable force, it is largely because private corporations have taken over the narrative and the technology. And it looks like this will continue to be the case. The resources needed to produce what’s called “paper worthy results” in AI are largely unavailable to non-profit and academic developers. As one researcher put it: “This trend toward training huge models on tons of data is not feasible for academics…. So there’s an issue of equitable access between researchers in academia versus researchers in industry.”

Capitalizing “big data” to the exclusion of public interests relinquishes to corporations the power to define and run the digital technology we get. Challenging that power is a rising chorus of critics calling for a people-oriented system, one based on public interest rather than market criteria, business advantages, or directives from the surveillance state.

Ramesh Srinivasan, a research scholar at UCLA, says it’s time for a “digital bill of rights” that would make public participation in the development of algorithms more democratic and diverse, reversing tendencies at tech companies dominated by white and Asian male engineers (eg., by ensuring that “users, workers and local communities” share power in the process of designing AI). This might diminish the self-importance of the digital technology sector and the imperious goals it sets for AI and other algorithmic tools.

Along these lines, Ethan Zuckerman argues for reining in technology giants’ influence over public discourse. He imagines a digital public infrastructure based on our historical experiences of public-service media—for example, by establishing public funding models: taxing surveillant advertisers Google and Facebook, expanding and guaranteeing public participation, and pressing for “auditable and transparent search and discovery tools” to counter existing opaque ranking and discovery algorithms (basically funding a public, non-profit competitor to Google).

Complementing these ideas is Victor Pickard’s proposal for a media system that elevates and sustains robust public interest journalism. We need an adversarial, investigative press to strengthen democratic self-governance, something that the billionaire-owned commercial media rarely provide. Such a system “reduces monopoly power; installs public interest protections; removes commercial pressures; and builds out public infrastructure.” We need to “let journalists be journalists” by giving them “a stake in the ownership and governance of media institutions.” This would generate the information people need in order to understand whether the goals of big tech align with the public interest and, if not, what to do about it.

We would like to add an ecological perspective to these proposals. The climate crisis will define our future. So we need a green digital bill of rights, a green digital public infrastructure, and throngs of science journalists who can help cultivate green citizenship and expand green governance to stop climate change and its human causes.

Finally, let’s talk about artificial intelligence’s carbon footprint. A study conducted by researchers at the University of Massachusetts, Amherst found that the energy it takes to train a single AI model “emitted as much carbon as five cars in their lifetimes.” This was the minimum energy it would take to train a single model for paper-worthiness. Scaling this baseline up to real-world AI training and tuning would mean massive increases in emissions.

Of course, AI has helped some data storage firms increase their energy efficiency. But demand for new digital goods and services using AI remains responsible for a growing share of carbon emissions, which some research indicates will surge mid-decade with increased data traffic, internet-of-things, enhanced computation, video streaming, gaming, augmented reality, 5G, autonomous vehicles, holography, digitalization, and so on.   Storage alone is a highly questionable practice. By one estimate, only six percent of all data ever created is in use. The rest is a “vast cyber landfill” that wastes electricity and pollutes the atmosphere.

A research analyst at Parnassus Investments admits that “Data is possibly overstated as an advantage for business, and no one’s really asking the question”— what if “a small group of people are the only ones really benefiting from this data revolution, then what are we actually doing, using all of this power?”

We should oppose the intensifying privatization of AI and other key technologies. In its place, we must envision a new public interest model for digital governance, one imbued with eco-ethical values and goals.