Artificial Intelligence Regulation – Three Months’ Worth

Controlling AI has been more fun for some people than using it or thinking about what it can do.  What’s the story been?

Before the current regime, we watched as the “Biden Administration Adopts Rules to Guide A.I.’s Global Spread” (Ana Swanson, The New York Times, January 13^th).  The “sweeping rules… governing how A.I. chips and models can be shared with foreign countries,” included various limitations on the number of A.I. chips that companies can send to different countries, with no bounds on those going domestically or to “18 of (our) closest partners,” with those “already subject to U.S. arms embargoes” barred, and all others “subject to caps restricting the number of A.I. chips that can be imported.”  There were also rules governing how much American companies can sell chips they have acquired elsewhere.

Is it true that “The Rush to A.I. Threatens National Security” (Heidy Khlaaf and Sarah Myers West, January 27^th, The New York Times)?  The authors claimed that “now that Donald Trump is taken office, the tech industry is moving full steam ahead in its push to integrate A.I. products across the defense establishment, which could make a dangerous situation even more perilous.”  Companies involved in the “slew of new partnerships and initiatives to integrate A.I. technology into deadly weaponry” included OpenAI, Anduril, Palantir, Meta, and Scale AI.  Potential problems include hallucinations, “cybersecurity vulnerabilities,” and data that could manipulate software, issues that the authors did not think could be solved.

What was “The Dangerous A.I. Nonsense That Trump and Biden Fell For” (Zeynep Tufekci, The New York Times, February 5^th)?  “America’s approach to A.I. safety and regulations,” which “was largely nonsense,” as “it was never going to be possible to contain the spread of this powerful emergent technology, and certainly not just by placing trade restrictions on components like graphics chips.”  “Instead… the government and the industry should be preparing our society for the sweeping changes that are soon to come.”  Specifically, “it’s time to harden our networked infrastructure,” “to start thinking clearly about how corporations and governments could use the A.I. that’s available right now to entrench their dominance, erode our rights, worsen inequality,” and determine “what we can do so that this powerful technology with so much potential for good can benefit the public.”  Perhaps regulation, Tufekci seems to be saying, is futile.

Some companies don’t mind that attitude, as “Emboldened by Trump, A.I. Companies Lobby for Fewer Rules” (Cecilia Kang, again The New York Times, March 24^th).  When under Biden “they wanted Washington to regulate them,” citing “the potential to disrupt national security and elections” and the chance to “eventually eliminate millions of jobs,” starting in late January AI companies have made “bold requests of government to stay out of their way,” by saying “it is legal for them to use copyrighted material to train their A.I. models” and “asking for the federal government to pre-empt states from creating A.I. laws.”  The Trump Administration has at least symbolically taken the less-regulation side through executive orders, statements supporting fewer laws, and invoking the value of “America’s global A.I. dominance.”

One form of regulation took effect April 2^nd, as “Deceptive deepfake media now a crime in N.J.” (Associated Press in Advertiser-News North, April 10^th).  New Jersey governor Phil Murphy “signed legislation… making the creation and dissemination of so-called deceptive deepfake media a crime punishable by up to five years in prison,” joining “at least 20 states.”  This version “defines a deepfake as any video or audio recording or image that appears to a reasonable person to realistically depict someone doing something they did not actually do,” and “establishes civil penalties that would permit victims to pursue lawsuits.”  Will the number of violations be small enough to allow enforcement?  Will such laws damage our need to assess veracity ourselves?  Will they turn out to be just adjuncts to others barring child sexual imagery?  I don’t think these questions will be easy to answer.  That makes this example, along with the others here, an artificial intelligence output which we will need to judge, accept, or reject.  It will take a while.