The war on Gaza appears to have drawn the media’s attention away from the big geopolitical drama of the past two years: the Ukraine war. Now is the time to try to understand how AI assesses a still unfolding event that soon may be spoken of in the past tense. Despite Ukraine’s top general Valerii Zaluzhnyi’s heterodox assessment that the conflict is now a stalemate and President Volodymyr Zelenskyy’s violent reaction as he continues to insist the war can be won, some experts believe it has already begun its denouement.

I thought one appropriate approach would be to bring up the case of esteemed US political scientist, John Mearsheimer, known for his historical realism. For the past decade, his analysis of the Russia–Ukraine relationship has been both remarkably consistently and completely at odds with official US policy. Because the latest events appear to validate his analysis, I decided to interrogate ChatGPT4 and asked the following question.

“John Mearsheimer has been saying in public the same thing about the situation in Ukraine since at least 2015. Had anyone in the Biden administration listened to his analysis and attempted to incorporate it into their policy orientations back in 2021, the war would have been avoided and a compromise would have been worked out, one that would probably have been more favorable to Ukraine than anything that is likely to happen in the near or even distant future. Most experts agree that Ukraine cannot win the war and has probably already lost it. What is the best explanation for the reluctance of three different State Departments to benefit from Mearsheimer’s expertise?”

Here is ChatGPT’s reply:

“John Mearsheimer’s analysis of the Ukraine situation has indeed been consistent and provocative. He has argued that the West, particularly the United States and NATO, are to blame for the crisis in Ukraine. He believes that the West’s push for Ukraine’s NATO membership was seen as an existential threat by Russia, leading to the current conflict.

However, there are several reasons why his analysis might not have been incorporated into the policy orientations of the Biden administration or previous administrations:

1. Different Interpretations of the Situation. Mearsheimer’s analysis is based on a realist perspective of international relations, which emphasizes power politics and national interests. However, not all policymakers or scholars agree with this perspective. Some might argue that values, norms, and international law should also play a significant role in shaping foreign policy.

2.  Disagreement with Mearsheimer’s Assumptions. Some critics argue that Mearsheimer’s analysis oversimplifies the situation and ignores important factors. For example, he has been criticized for not adequately considering the security imperatives of Russia’s neighbors, the increasing authoritarianism of the Russian state, and the impact of Russia’s actions on Ukraine.

3. Policy Constraints. Even if policymakers agreed with Mearsheimer’s analysis, they might face constraints that prevent them from acting on his recommendations. These could include domestic political considerations, commitments to allies, or legal and normative constraints.

4. Differing Views on NATO Expansion. Mearsheimer has criticized NATO expansion as a cause of the conflict. However, many in the West see NATO expansion as a way to secure peace and stability in Europe. They argue that countries have the right to choose their own security arrangements and that denying them the opportunity to join NATO would be a concession to Russia’s aggressive behavior.

It’s important to note that foreign policy decisions are complex and involve a range of considerations.

Assessing ChatGPT’s conversational skills

This answer reveals several interesting things about how ChatGPT supposedly reasons.

The first is that, though it passes, at least superficially, the Turing test, it does not directly address the question I asked. My question contained an important hypothetical concerning the outcome of the conflict when I compared the situation pre-war and post-war. ChatGPT “chose” to ignore my essential concern. What strategy dictated that choice?

ChatGPT’s third point is the closest to what you might call an engagement with my question, which sought to understand the reluctance to take Mearsheimer’s advice on board. ChatGPT spends most of the time citing reasons for doubting Mearsheimer’s prescience. In my question, “reluctance” refers to an attitude, not a line of reasoning. I then continued my questioning.

“I didn’t ask you about why people disagree with Mearsheimer. I asked about reluctance to listen to him. In point 3 you cite “constraints that prevent them from acting on his recommendations.” That is what my question was about. So what can you say about those constraints in real historical terms? For example, what interests are at play that determine or influence those constraints.”

ChatGPT offered five reasons. I’ll cite only the first one in full because the others are purely theoretical.

1. Domestic Political Considerations. Policymakers often have to consider the domestic political implications of their decisions. For example, taking a stance that appears to be ‘soft’ on Russia could be politically risky, especially if it is perceived as compromising U.S. or allied security.

2. Commitments to Allies.

3. Legal and Normative Constraints.

4. Economic Interests.

5. Security Obligations.

I followed this up by asking the following question.

“Your first point that politicians in Washington don’t want to appear ‘soft’ on Russia is germane to my initial question, which was about how to avoid the horrors of war. Are you suggesting that negotiations fall into the category of being soft? Can that explain why experts in Washington would prefer war to diplomacy? Isn’t that a frightening prospect, if true?”

As usual, ChatGPT avoids the core of the question, which is my very human concern about the risk and the cost of war. Here is the first paragraph of ChatGPT’s response.

The perception of being “soft” on Russia does not necessarily equate to a preference for war over diplomacy. In fact, diplomacy is often the preferred method of resolving international disputes, including those involving Russia. The U.S. and its allies have engaged in numerous diplomatic efforts to resolve the situation in Ukraine, including sanctions, negotiations, and peace agreements.

Our AI friend is clearly evading the question and hedging his bets. Isn’t it capable of taking on board what Mearsheimer and others have noticed is a repeated pattern of refusing offers of diplomacy? Mearsheimer provides plenty of evidence of such offers as well as warnings about the consequences of not engaging. The feeble final sentence of the above paragraph, which is vacuous in meaning, makes the absurd assertion that “sanctions” are a form of diplomacy. Most commentators admit sanctions are “economic warfare,” of a kind that punishes populations rather than governments.

ChatGPT then spends three paragraphs in its patented theoretical mode making points like the following:

— “Balancing the need for effective diplomacy with the need to deter aggressive behavior” to avoid potentially encouraging further aggression.”

— “The complex and often difficult choices that policymakers must make when dealing with international conflicts.”

As it wends its way towards a conclusion, it reaches the height of absurdity when it tells me: “It’s also important to note that the goal of any policy should be to promote peace, stability, and respect for international law.”

My question was about a very real war that has sacrificed hundreds of thousands of lives, the integrity of a nation and compromised the global economy. The point of my initial question was that some wise people have understood the stakes and the probable outcome. I could have cited the cases of Afghanistan or Libya, to name only those two, to clarify my meaning. ChatGPT concludes with a sentence any historical realist can only laugh at.

Conclusions

From this exchange, I cannot avoid concluding that our best current version of AI still has some serious problems passing the Turing test. To start with, it has trouble distinguishing between attitude and reasoning. Next, it lacks the basic human conversational ability to form an idea of the interlocutor’s strategic intentions expressed in a straightforward dialogue.

But that isn’t all. It reflects a defensive attitude about historical events themselves. It strives to find theoretical reasons to justify even such horrors as wars based on mistaken assumptions or perverted intentions.

This is intriguing. I intend to pursue the experiment further next week by developing a contrasting case. My takeaway from this line of questioning is that with AI we need to determine the level of trust or mistrust we can have in its ability to deal with complex human issues. For the moment, I give the advantage to mistrust, simply because the persona I was talking to never truly enters into the conversation.

*[Artificial Intelligence is rapidly becoming a feature of everyone’s daily life. We unconsciously perceive it either as a friend or foe, a helper or destroyer. At Fair Observer, we see it as a tool of creativity, capable of revealing the complex relationship between humans and machines.]

The views expressed in this article are the author’s own and do not necessarily reflect Fair Observer’s editorial policy.

Is the Head Of the CIA a Useful Idiot, or a Peacemaker?

Outside Ukraine, Russia Is Now Losing Its Grip

How Does Brutal Fighting in Ukraine Affect A Soldier’s Mind?