#TradeXpresso Lungo: will ChatGPT be a good trade negotiator?
Lucian Cernat
Head of Global Regulatory Cooperation and International Procurement Negotiations at European Commission
It’s a quiet, pleasant Sunday. A perfect day to scroll through random things online, without any particular purpose. One thing that made me pause from my erratic gazing was Bill Gates’s prediction that artificial intelligence will be the hottest topic in 2023. What made me pause was not the prediction itself, which wasn’t that surprising, but the sudden thought that AI might become part of my profession.
Could ChatGPT become a trade negotiator? After all, ChatGPT has recently managed to pass difficult MBA or law school exams , and even medical licensing exams . This is quite impressive, given that human language is a very imperfect means of communication. Words have multiple meanings, often unrelated. People use the same words to express different thoughts. Two people can read the same phrase and understand different things. And then there is irony, metaphors, poetry and ambiguity.
Old-fashioned negotiators might argue that ChatGPT would fare poorly as a trade negotiator.?Trade negotiations may not be famous for their poetic nature but they are renowned for their constructive ambiguity. Sometimes deliberate, sometimes accidental. Trade negotiations also require a fair amount of posturing, lots of personal interactions during negotiating rounds, and soft skills to reach a final deal. An experienced negotiator once said that one of the best ways to persuade your counterparts is by listening to them. You also need to be able to bluff, compromise and let go of less important stuff.
Yet, you might be in for a surprise. Trade policy has been already casually flirting with artificial intelligence since 2018. The IBM “garage experiment” didn’t come out of the blue. There is an emerging and promising “law as code” school of thought. Words, not just numbers, are data. Big, legal words lead to big data. By my quick and rough estimate, the total number of words in the text of all existing FTAs around the world surpasses 100 million. Then add the trade law jurisprudence, and all the textbooks available online and you get close to the kind of data AI needs to become legally smart. Then, even the public sector (not usually known for their ability to generate disruptive innovation) tried to take advantage of "Trade Policy 2.0" digital tools and algorithms to facilitate trade policy implementation.
However, despite these trends towards greater use of AI in trade policy making and all the hype about ChatGPT, I wasn’t quite sure how good it is at handling specific trade policy questions. So, I decided to test ChatGPT with several “geeky” questions from two trade policy areas I am currently responsible for: technical barriers to trade, and public procurement.
It did fairly well in some respects, and it performed rather poorly in others. I asked ChatGPT two types of questions. The first kind were to compare certain TBT articles from two EU FTAs to find the legal differences and decide which FTA is more ambitious. It did manage to identify correctly the more ambitious provisions and briefly explain why. This indicates that perhaps there is some scope for using ChatGPT-like AI tools to assist in more routine legal trade analyses, e.g. in the so-called "legal scrubbing" of FTAs, a rather tedious and time-consuming task.
领英推荐
The second type of questions were related to FTA implementation. I asked ChatGPT to establish if the public procurement chapter of two EU FTAs offered non-discriminatory and national treatment access for EU bidders in certain specific sectors. Here, ChatGPT failed this test. It simply summarised the general objectives of the FTAs in the area of public procurement and admitted that the answer to such specific questions depends on a complex interpretation of several FTA market access schedules, and invited me to do a further analysis to find the answer.
This inability to tackle a detailed public procurement question was somewhat reassuring. It meant that the recent “Access2Procurement” algorithm developed by DG Trade for the benefit of tens of thousands of EU prospective bidders for procurement opportunties abroad is (still) superior to one of the most advanced AI tools.?
It also means that, probably, I will reach retirement age before being made redundant by AI. In fact, ChatGPT did explicitly admit in one of its answers that it has no ambition to replace the human expertise required for successful trade negotiations. I hope it doesn't change its mind.
Notice: No parts of this text have been generated by AI. Just old-fashioned, traditional writing methods.
Associate Professor in Applied Economics at Universidad Complutense de Madrid
1 年It is really worth it to read!
Trade negotiator, trainer and analyst
1 年Very interesting, and reassuring. Thanks for sharing your thoughts on the issue