Bad Moon Rising
Bad Moon Rising
Landmark report reveals lack of compliance by major AI companies with forthcoming regulation
I see the bad moon a-risin'
I see trouble on the way
I see earthquakes and lightnin'
I see bad times today
Don't go around tonight
Well it's bound to take your life
There's a bad moon on the rise
In a story more suited for this lunatic day in age, John Fogerty - the mastermind behind Creedence Clearwater Revival, one of the most popular bands in the history of recorded music - was sued for impersonating himself, back in 1985.
The lawsuit was the culmination of many years of acrimony between him and his former label chief, Saul Zaentz.
Fogerty, who departed CCR and started a solo career back in 1972, wrote a song that Zaentz’s label, Fantasy Records, decided sounded a little too much like a previous CCR song…a song which Fogerty also wrote.
Bizarrely, this lawsuit progressed through the court system, leading to a surreal moment where Fogerty, guitar in hand, testified in court with regards to how his uniform way of songwriting can lead to songs musically sounding similar, when the words and other aspects of the song can be radically different.
Predictably, the court eventually ruled in favor of Fogerty. But the story didn’t end there…
Fogerty had accumulated a million dollars in legal fees fighting the lawsuit, which the court did not require Fantasy Records to reimburse even though they lost. Fogerty appealed, and lost again (running up even more legal fees).
Undaunted, Fogerty then took the case to the Supreme Court, where in 1993, Chief Justice William H. Rehnquist led a 9-0 unanimous ruling in favor of Fogerty’s reimbursement of $1.3M in fees by Fantasy Records, instantly and permanently changing the way the judicial system evaluated the costs of legal defense.
With 499 votes in favor, 28 votes against, and 93 abstentions, Europe’s AI Act was passed by European Parliament last week, clearing the way for each individual member-state to ratify it for themselves - likely a formality.
This news was immediately followed by the release of a report by the Harvard Kennedy School, completed in partnership with Stanford’s Institute for Human-Centered Artificial Intelligence (HAI), measuring whether current large language model-based AI systems comply with this law.
Big shock: most are non-compliant, in the ways that matter most.
The authors of this report looked at 12 requirements in the draft law that specifically apply to foundation models and developed rubrics to grade compliance.
Then, they graded ten major providers and their flagship foundation models on each requirement.
Finally, they aggregated providers’ performance on each requirement into a single score as a proxy for compliance, and compared results among providers to gauge trends across the landscape.
There’s many newsletters worth of thoughts that could be written about these findings, but what I’ll focus on for now is the laughable, almost-universal non-compliance with regulations related to data:
AI compliance is about to be an entirely new area of our international legal system.
Hiring skilled lawyers to work in-house for conversational AI companies will be commonplace for every single venture-backed startup by the end of the year, and these people will be equal in importance to the CEO probably for a good while to come.
Also, take another good look at that graphic above. Someone on that list will get sued into oblivion and made a martyr for AI ethics and compliance - just a matter of who. In fact, the amount of AI lawsuits will explode here in the US as soon as this regulation makes its way over here. Every company doing anything even remotely noteworthy will be tested.
Within the startup ecosystem within conversational AI, venture-backed companies should be moving quickly to make AI compliance a competitive advantage. In fact, given the nebulous nature of attempting to quantitatively measure one AI system’s superiority over another, and that it’s probably much easier to demonstrate AI compliance to potential customers, this is where business development and content marketing efforts should also immediately head.
At Project Voice 2023 back in April, we were fortunate to have attendees and speakers alike with numerous connections to both the Harvard Kennedy School and Stanford’s HAI institute present, and we are in the process of connecting the Conversational AI Leadership Council with both of these groups to provide further support.
Welcome to what will be the singular AI focus for the rest of this year.
I encourage you to review the full report here.
