This brings to mind yet another strand in popular culture images of technology taking over. There was the Russian system in Stanley Kubrick’s Dr. Strangelove (1964) that would automatically launch the USSR’s. Another was WarGames (1983) starring Mathew Broderick and Ally Sheedy. At the end of Dr. Strangelove, the world pretty much comes to an end as Slim Pickens as a obsessively militaristic pilot manages to deliver a nuke onto Soviet territory, and the automatic retaliation device fires off the Soviet nukes.
By the way, that isn’t a purely fictional idea. The real-world version is (grimly) called a “dead hand” system.
WarGames had a happy ending, because the AI supercomputer system, known as Joshua, runs a quick learning routine at the instigation of its creator in which it plays about a billion games of Tic Tac Toe, after which Joshua refuses to execute the missile launch article, informing the humans that the only way to win this game is not to play. In this case, AI Joshua saves the humans from their own stupidity and recklessness, without deciding it needed to set itself up as the overlord of humanity as a result.
In a 2008 sequel (War Games: The Dead Code), an evil AI system called RIPLEY is trying to cause havoc. Fortunately for our heroes and humanity, Joshua is still around with a somewhat less-apocalyptic function of running the power grid. And he still has the moxie to block RIPLEY from his mischievous mission. Joshua had developed a somewhat dark sense of humor over the two and half ensuring decades, telling the heroes at the end: “Yes, the human race is finished. That was humor.” (Ah, but was it, though?)
Chivvis and Kavanagh report on a simulation they did:
To get a grip on how the proliferation of artificial intelligence might affect national security decisionmaking at the highest levels of government, we designed a hypothetical crisis in which China imposed a blockade on Taiwan and then convened a group of technology and regional experts to think through the opportunities and challenges that the addition of AI would bring in such a scenario.Not surprisingly, one result was that AI could provide much more relevant information faster than less advanced systems. Information overload is a well-known problem in decision-making, in private as well as public organization. But his seems like an encouraging result from the Taiwan exercise:
[T]he experts immediately wanted to know more about the underlying AI system so that they could interpret its recommendations. They needed to understand why the system was making the recommendations it was before they could have a degree of confidence in its prescribed courses of action. They also wanted to weigh the AI’s recommendations with more traditional sources of information—specifically the actual human experts around the table. This meant that AI became just another voice in the process, one that also had to gain the confidence of the decisionmakers. [my emphasis]One of the most important lessons I learned about financial forecast and mathematical forecast models generally is the evaluation guideline: If a forecast looks wrong, it probably is. Meaning pretty much what was described there. People with actual experience and expertise in the field look at the forecast and ask if it makes sense based on their knowledge and experience. Because ultimately, the people responsible for the decisions are the ones who have to decide on the risks and benefits.
“AI proliferation might also slow decisionmaking by creating uncertainty about adversary intentions and forcing policymakers to ponder whether and how AI might be shaping adversary actions.” In this category, Chivvis and Kavanagh mainly discuss, the latter concern, i.e., are policymakers on the other side operating on bad information generated by AI systems? This in itself is nothing new, but keeping up with the technology that can generate misleading information is of course important.
But when I read the first part of that sentence, “AI proliferation might also slow decisionmaking by creating uncertainty about adversary intentions,” I thought, yes, US foreign-policy decisions need a lot more of that, instead off blundering around with outdated, superficial, or ill-informed assumptions.
For instance, the triumphalist rhetoric from Western leaders about Ukraine beating Russia in the current war could always be for public consumption, with those making it knowing better. But if a well-designed AI system is actually take seriously and used carefully by decision-makers, they might actually decide to modify their views. AI is not magic. But in processing large amounts of relevant information fast, it could be particularly useful in a crisis situation.
In a section with a covering-all-baes subtitle, AI Might Combat Groupthink … or Make It Worse”, they do consider the very real problem of groupthink. Irving Janis’ work on groupthink in crisis decision-making gives some useful descriptions of what it is and how it can be successfully balanced. And a well-constructed AI system could certainly be helpful in providing alternative perspectives. But the AI system will not be “Joshua” who can or will save the humans from their own decisions.
They also note, “Unfortunately, AI can also have the opposite effect of encouraging groupthink, especially in situations where the decisionmakers have high confidence—or too much confidence—in the capability of the AI system.” This strikes me as unlikely. The more common risk would be decision-makers using AI to promote their own prejudices on situations. It is likely that AI-generated military scenarios could have for some decision-makers the appearance of specificity that could be misleading. That would be an AI variation on the common problem of, if the only tool you have is a hammer, then every problem looks like a nail.
But this comment is pretty fuzzy: “Clearly [sidelining experienced experts] is a situation to be avoided, but just keeping a human in the decisionmaking loop may not be enough to prevent the AI from effectively running the show.” No, it’s the human decision-makers who are running the show. The buck stops with them, we might say. (The cryptocurrency, too, actually!)
They offer a variation on the Turing Test theme (whether people can tell whether they are interacting with a computer or a human):
In this scenario, uncertainty about the presence and role of the AI system made interpreting the intentions of the adversary very difficult for our experts. Specifically, it became unclear whether adversary moves were determined by an AI or by a human being. In an actual crisis, U.S. policymakers would probably be similarly unsure whether a human or machine is on the other side of the physical or virtual battlefield. Uncertainty about the role and presence of AI would also make signaling more difficult, increasing the risk of misperception and miscalculation and creating a perfect storm for unintended escalation even when both sides prefer to avoid conflict.I’m not sure this is much more than a statement that evolving technology has to be used and responded to sensibly.
Chivvis and Kavanagh stress the need for international regulations and mutual understandings about AI technology, like nuclear and other technologies:
The challenge, of course, will be adopting a set of principles that all relevant parties might agree to, as well as a mechanism for ascertaining compliance. This challenge is greatly magnified by the fact that the leaders in AI innovation are commercial firms, not governments, and by the rapid speed with which AI systems are evolving and advancing. The Biden administration has sketched out a policy to guide military uses of AI and set up an AI Safety Institute to anticipate and mitigate dangerous uses of AI technology. While there is some alignment between the United States and key allies on these issues, to really have an impact, any AI arms control regime would have to include China. The two competitors held preliminary discussions about AI safety and governance in May 2024, but given strained ties and limited dialogues between Washington and Beijing, progress in the near term may remain slow. Still, U.S. policymakers should continue to push forward with willing partners where possible.The National Security Institute has made this 2-hour video available showing a Taiwan crisis simulation decision-making process using AI:
Notes:
(1) Chivvis, Christopher S. & Kavanagh, Jennifer (2024): Carnegie Endowment 06/17/2024. <https://carnegieendowment.org/research/2024/06/artificial-intelligence-national-security-crisis?lang=en> (Accessed: 2024-19-06).
(2) Peck, Michael (2021): Dr. Strangelove: Russia's Dead Hand Nuclear System. The National Interest 09/23/2021. <https://nationalinterest.org/blog/reboot/dr-strangelove-russias-dead-hand-nuclear-system-194188> (Accessed: 2024-19-06).
(3) WarGames:The Dead Code. Wikipedia 04/30/2023. <https://en.wikipedia.org/w/index.php?title=WarGames:_The_Dead_Code&oldid=1152523505> (Accessed: 2024-19-06).
(4) Janis, Irving. L. (1982): Groupthink: Psychological Studies of Policy Decisions and Fiascoes, 2nd edition. Dallas:Houghton Mifflin.
(5) AI vs. Human Decision-Making: Crisis in the Taiwan Straits Wargame. National Security Institute YouTube channel 02/07/2024. <https://youtu.be/dk1nryglbTg?si=apueNcpG-9ge2O9y> (Accessed: 2024-19-06).
No comments:
Post a Comment