Gary Conklin: How to prevent an AI race to recklessness
About the writer: Gary Conkling started writing stories as a child and publishing them on his own hand-cranked printing press. Little did he know digital technology would make it possible to repeat the task as an adult by publishing his own blog, Life Notes. He is a journalist by trade who has worked in the trenches of public affairs at the federal, state, regional and local levels. But he also is an observer of life occurring around him. This piece is from his blog, found at https://garyconklinglifenotes.wordpress.com.
The barons of artificial intelligence say they want robots to perform all forms of manual labor. Critics say they really want to monopolize economic work, making most human labor obsolete.
The promise of using AI to perform repetitive tasks faster and more efficiently may be slipping into a nightmare where humans are sidelined from most tasks where they earn money. AI already is revealing its wolf-in-sheep’s-clothing by performing blue- and white-collar jobs under the cover of helping humans escape “dull, dirty or dangerous work.”
On the other hand, humanistic AI seeks to augment human intelligence so people can perform tasks better and faster, according to proponent Tom Gruber, who was the head designer of SIRI.
Robots are deployed in automotive and electronics manufacturing, materials handling, order fulfillment, sorting and packing, sanitation and surgical assistance. They assist stroke patients and the elderly with mobility issues, greet customers at banks and hotels and flip burgers and fry chips in fast-food restaurants.
There are autonomous tractors and drones that can plant, weed and spray insecticides. Some specialized robots assist in rescues, inspect infrastructure and defuse bombs.
However, the next wave will be humanoid robots with pre-programmed routines and AI agency capable of sensing and adapting to different environments. This generation of robots will feature flexible materials that can safely interact with humans and handle delicate objects. And these next-generation robots could replace even more Americans in the workplace.
Tristan Harris, a former tech worker who runs the Center for Human Technology, told NPR, “The only way they can justify the amount of money they’ve invested is if they race to replace economic labor in the economy. That’s the stated mission statement of OpenAI - to do everything that a human laborer can do.”
“They’re not racing to augment and support human workers. They’re racing to replace them,” Harris insists. “And what that will lead to is unprecedented levels of concentration of wealth and power because all the money in the economy - instead of being paid to individual laborers - will go to five to 10 AI companies with robots that do all of the work.”
“There is no question we have reached a new level of competence in artificial intelligence,” Gruber says. “We read breathless celebrations of how AI programs have beaten the masters of humanity’s most challenging games, or how it can recognize a human face among billions of images.
“We are alarmed by warnings that AI will displace our jobs, or even threaten our existence as it reaches superhuman levels of cognition. Yet we hope that it will be used towards more useful and benevolent ends, like driving our cars and curing diseases.”
Gruber believes humans still can determine AI’s ultimate direction.
“Both futures are possible,” he says. “Are they both inevitable? That’s up to us. How we frame the problem is vital.
“Must we live in a world where machines are set against humans at every turn, surveilling us, shaping our beliefs, manipulating our attention and competing with us for jobs, all driven by a zero-sum race to superintelligence that leaves the winners to take all? What can we do to ensure that applications of AI that are beneficial to humanity receive the necessary resources, priority and policies?”
According to Harris, the pursuit of AI dominance is a “race to recklessness” that outstrips current uncoordinated efforts to regulate the AI industry.
He cites the “wisdom gap” between exponential growth in the AI industry and resources devoted to regulatory, societal and ethical governance. He estimates a 30-to-1 ratio between investments on AI capability versus AI safety.
Harris advises slowing down AI deployment to provide time to establish “robust guardrails, AI auditing and fairness and safety systems.” The ultimate goal, he believes, should be “humane AI design aligned with human well-being, not just profit.”
President Trump has warned off state-level AI regulation, claiming a patchwork of regulations would hobble the ability of America’s AI industry to innovate and win the “global AI race.”
Some states are moving ahead anyway in the absence of meaningful federal regulation. California Governor Gavin Newsom issued an executive order requiring safety and privacy guardrails for AI companies contracting with the state.
Policymakers in Utah, Florida and Colorado eyed some form of AI regulation, but backed off under pressure from the Trump administration and AI industry leaders.
Republican Governor Sarah Huckabee Sanders of Arkansas, a former top Trump staff member, told The Washington Post, “Americans are at risk from bad actors in the AI industry until lawmakers are allowed to establish basic rules and fairness - and 10 years is too long to wait.”
“AI is like the Super Bowl of public policy because it is at once hyperlocal, national and international in significance,” explains Kevin Frazier, an adjunct research fellow at the Cato Institute. “Everyone feels like they have the right to do something about it.”
The Trump White House issued a statement reinforcing its opposition to state regulations and willingness to impose federal rules to protect children, force AI companies to pay for data center energy costs and ensure copyright protection. David Sacks, Trump’s top AI adviser, predicted Congress could consider such legislation on the federal level in the near future, pre-empting the states.
Who regulates AI and how has repercussions ranging from economic growth to child safety, from national security to the ability of Americans to find jobs. The tussle over AI regulation may have a new dimension as public concern grows over the rapid expansion of sprawling, energy-consuming data centers.
Harris raises another concern as well - what happens if advanced AI systems go haywire. He cited a university study that used all the leading AI models in a series of war-game scenarios.
“They generated 780,000 words of reasoning instantly, and they’re going back and forth in a turn-by-turn scenario,” he said. “And in this scenario, AI escalated to the use of nuclear weapons 95 percent of the time. What this says is that AI is reasoning in a way we don’t understand - and we don’t control what it’s going to do.”
Harris isn’t alone in foreseeing dire consequences. Rudolf Laine and Luke Drago wrote “The Intelligence Curse” that describes cascading adverse outcomes that “disempower the vast majority of humanity” by starting with automation, then making human talent obsolete, prioritizing capitalization and breaking the social contract.
Laine and Drago have professional experience working with AI and machine learning. Their antidote to avert catastrophe is to diffuse control of AI technology.
“Get AI in the hands of regular people,” they counsel. “In the short-term, build AI that augments human capabilities. In the long-term, align AI directly to individual users and give everyone control in the AI economy.”
Laine and Drago also advise “democratizing institutions now, making them more anchored to the needs of humans even as they are buffeted by the changing incentive landscape and fast-moving events of the AGI transition.”
A blaring warning sign came when the Anthropic CEO Dario Amodei refused a Department of Defense demand to remove guardrails from its Claude AI. The Pentagon insisted the AI system should be unimpeded in all tasks including autonomous weapons and surveillance.
When Amodei said “in good conscience” he could not agree with those demands, the Pentagon banned Anthropic products in all federal agencies, labeling Claude AI as a “supply chain risk.”
In rejecting the Pentagon’s demand, Amodei said current AI models are too unreliable for life-or-death targeting and the risk of enabling unconstitutional surveillance.
Gary Conkling started writing stories as a child and publishing them on his own hand-cranked printing press. Little did he know digital technology would make it possible to repeat the task as an adult by publishing his own blog, Life Notes. He is a journalist by trade who has worked in the trenches of public affairs at the federal, state, regional and local levels. But he also is an observer of life occurring around him. This piece is from his blog, found at https://garyconklinglifenotes.wordpress.com.
Comments