AI as an arbitrator
The solution to the rail strike could be so simple – you just have to listen to ChatGPT
The train drivers’ union GDL and the railway have been arguing for weeks and are literally bringing the country to a standstill. The solution to the rail strike would actually be quite simple – if only it weren’t for the stubborn people.
It’s a tempting thesis: If there is another dispute like the current collective bargaining and the resulting one If a rail strike is imminent, why not let it be negotiated by an authority that is truly neutral – namely an artificial intelligence? This suggestion is currently increasingly circulating on social media. We simply tested it. And used ChatGPT as an arbitrator.
Of course, this is not an ideal line-up: OpenAI’s language model is not actually intended to moderate negotiations, but rather to write texts. However, AI can especially shine when it has been optimized for a specific purpose. But: Negotiating is also about processing arguments and then evaluating them. A voice AI can do that quite well. Unlike human arbitrators, the chatbot cannot be accused of lacking neutrality.
Rail strike: According to AI, this is the solution
The experimental setup is simple: We first feed ChatGPT with all the positions of the two parties, in the current wage dispute that is Deutsche Bahn and the train drivers’ union GDL. The train drivers want more money for fewer hours per week in order to dampen the effects of inflation and the strain on health. The railway wants to prevent this – and argues with cost pressure and problems with shift staffing. Of course, you have to explain all of this to the bot in much greater detail than is necessary here. Then comes the big question: Is a compromise possible?
After a bit of chattering, ChatGPT – the current model PGT-4 was used – actually spits out a solution. Noteworthy: The AI weighs up the positions and demands tough concessions from both. She also emphasizes that this is only a possible suggestion. But it also shows that a neutral compromise that takes both sides into account is possible. You can find the compromise proposal here:
The human factor
What the nice compromise unfortunately doesn’t take into account – and after the first request it shouldn’t – is the human factor. Unfortunately, it is not the case that the negotiations are conducted by neutral algorithms. The railway negotiators and the union boss Claus Weselsky are just people. Of course, they also know that a compromise would quickly end the dispute. It’s just that neither side is currently ready to meet in the middle. So what if you bring this factor into play?
Then ChatGPT also sees problems. “If both sides insist hard on their positions, it will be more difficult to reach an amicable agreement,” the bot muses. “But there are still scenarios for how a compromise can be found.”
These are pious ideas. Even ChatGPT notices this. He adds: “Given the extreme positions of both sides, none of these approaches can guarantee a quick solution. But such negotiations are often a process and it can take some time to reach an agreement. It is important that both parties have the will show that we can find a solution that is in the best interests of everyone involved.”
It gets easier without feeling
Now that could be a problem. Weselsky is retiring in the fall; these are his last major negotiations with the railway. The 35-hour week could be his big goodbye. At the railway, on the other hand, there are supposed to be people who do not begrudge the trade unionist, who is not very popular there, this success. So how do you deal with this, dear AI?
But that’s exactly what’s so difficult.
What makes negotiation AI so difficult?
Creating a truly neutral negotiation AI would be almost impossible anyway. Because that would fulfill several huge requirements: The AI would not only have to know the positions of the parties down to the smallest detail – but also the hidden motivations of the respective negotiators. In addition, both parties would have to feed the artificial intelligence with every aspect of their position and the motivation behind it and be honest throughout – with the AI, but above all with themselves. If that were the case, you would hardly need any AI more. ChatGPT already recognizes this fundamental dilemma of the human factor:
If only it were that easy.