Since November, the internet has gone into a frenzy over developments in artificial intelligence. Generative AI systems continually surprise their own creators with powers that are improving at exponential rates. They write term papers, legal briefs, diet plans, political campaigns and screenplays. They pass college entrance tests and bar exams at the 90th percentile of human scores. People without any coding skills can provide prompts that generate compelling visual images to order. People in need of therapy can turn to AI for interaction. It is all very exciting – and disturbing.
Programmers do not actually understand the software they have created the way mechanics, for instance, understand the machines they build. These new systems are designed to mostly program themselves using methods modeled, loosely, on evolution and animal learning through reinforcement. The programmers function more like dog breeders and trainers who can guide the process without actually knowing anything about the DNA of evolving breeds or the neural brain systems of creatures they train. The evolving black boxes of code include billions of parameters and formulas that are beyond any human’s power to grasp.
Despite ignorance about these black boxes’ workings and these systems’ side effects, Microsoft and others are plunging ahead in marketing them to millions. They plan to wrap them into search engines like Bing and commonly used apps like Word. And they plan to pay for them, at least in part, with advertising. If funded by advertising, the AI systems that now exist will prove incredibly effective at turning citizens and their deliberative powers into products any businesses and politicians can purchase.
Consider an analogy. Suppose Purdue invents a drug that raises everyone’s IQ by 40 points. It increases worker productivity dramatically, causing massive layoffs. It creates super-empowered criminals and terrorists. Why it’s effective as well as its side effects (like addiction and personality disorders) are not well understood. But they can make customers have desires for specific foods, drugs and beverages as well as susceptibilities to specific kinds of messaging. Investors and customers are excited. The drug can be distributed for pennies, so Purdue decides to give it away. The plan is to create exponential growth in the base of users who can then, themselves, be marketed as products on the back end. Purdue’s income will flow from companies that market the beverages and messages that get sold to the drug’s users. Other companies with similar drugs feel strong pressure to jump into the market before Purdue can dominate it.
Competition to build user bases for somewhat analogous products is creating chaos in the AI industry. Conscientious researchers and entrepreneurs are deeply worried. They would welcome initiatives to slow things down and bring predictable structure to the chaos. They need to figure out ethical guidelines, testing procedures and security systems to promote friendly, beneficial AI. Might government help with this?
Here are possible government initiatives worth exploring: 1. Ban funding AI apps with advertising; 2. Require the equivalent of environmental impact assessments for public review and systematic assessment of major new AI systems before marketing; 3. Moratoria on wrapping AI into existing apps like Word and Google Docs until appropriate testing and oversight can be completed; 4. Create an industrial board for collaborative research and oversight that will empower companies themselves to set and enforce industry standards for ethical and socially beneficial AI, and 5. Create a well-staffed and -funded national security agency for providing policy research and implementation to ensure that legislation is developed in sensible, cutting-edge ways and that it is implemented effectively.
The AI industry is at a pivotal moment. Regardless of whether the machines ever become conscious or duplicate human intelligence, they are about to dramatically transform the way we run schools, clinics, businesses, police departments, law offices, research labs, factories, forests and food systems.
The Maine congressional delegation is noted for working across aisles to get important things done when time is of the essence. It is essential at this time that they initiate hearings to ensure coming “smart” systems are wise and ethical.
Send questions/comments to the editors.
