In his regular monthly column, Pierre Aury gives readers the pluses and minuses of artificial intelligence for our industry.
Hi Pierre
Thank you for reaching out. I would be glad to discuss the question and provide clarity. I will be happy to connect and talk more. Please let me know if you would like to schedule a meeting or a call.
Regards
The above is a ChatGPT-generated answer to an email. These days AI is used in e-commerce, banking, finance, healthcare, social media, and lifestyle to mention only a few sectors and AI has been mentioned for some time by a number of Software as a Service or SaaS vendors for shipping applications as well but now it has even made it to the field of emails with the use of this ChatGPT plugin by a number of brokerage firms and principals. If enabled the plugin proposes answers to incoming messages with the option to accept them or not. So believe it or not you are going to be exposed daily to AI in shipping in activities as trivial as exchanging emails. Strange enough although AI is mentioned all the time there is no commonly shared definition of AI but the Council of Europe proposes a long but good one which can be boiled down to:
AI refers to systems that are pure science fiction (so-called “strong” AIs with a self-aware form) and systems that are already operational and capable of performing very complex tasks (face or voice recognition, vehicle driving – these systems are described as “weak” or “moderate” AIs).
…essentially using machine learning…
Machine learning is … an inductive approach: …letting computers discover rules by correlation and classification, on the basis of a massive amount of data…the objective of machine learning is not really to acquire already formalized knowledge but to understand the data structure and integrate it into models, in particular, to automate tasks.
The question is: are there situations in shipping that qualify as being so complex that AI will do much better than a normal human brain? And the answer is yes … and no!
Email Traffic
Let us take emails as an example. Email traffic has grown steadily for many years and with an average of 165bn mails being sent daily in 2022, inboxes can get hundreds or thousands of mails a day depending on whether you are a small shipping company or big brokerage firm. A solution which predated the widespread use of AI was simple email parsing. Parsing is the process of using software to identify and extract specific data in usually formatted incoming emails. AI parsing is becoming more sophisticated recognising and sorting all the terms of proposed shipping spot transactions or the specifications and the next open positions of various ships. In a second step, the information is made available to other programs through a defined interface killing the good old copy and paste.
If we look at real numbers like the dry bulk panamax fleet at say 2,400 vessels with 60% of them trading spot with an average number of trips performed every year of six we have an average of 166 vessels open and available for a cargo every week worldwide. Even splitting this number into the two main basins, Atlantic and Pacific, this leaves a considerable number of ships to track with all the associated email traffic making a good case for some technological help be it AI or something else.
Another front on which AI is a must is on the sanctions front as AI applied to AIS can flag potential breaches that a human brain will not be able to detect. Where the case is not so clear is on the pre-trade front. Pre-trade as it is called is defined as all the operations taking place before doing a trade, so basically all the work done to decide the timing of the trade and its price level and to select the right counterparty with the right ship.
We can’t help but think that SaaS being proposed on the market, most now with an AI component, are a bit of an overkill. Shipping is essentially an immensely simple activity with the market going down if there are two ships for one cargo and going up when there are two cargoes for one ship. The outcome of any research pre-trade will be binary: either the spot is going up, in which case the owner will want to wait to fix or it is going down in which case the owner will want to hurry to fix. These systems are very powerful and some are very sleek as well as meaning that crunching a big number of time series becomes an easy task.
This can develop the perception that the answer to the old problem of making good deals lies in crunching more data when perhaps another answer is to sit back and think about what sort of data is relevant. A case of using data because it is relevant not because it is available. Using data because it is available can as well lead to severe cases of analysis paralysis: always waiting for more data to finally make a good decision. This begs another question: what is a right decision?
In the end the profit of a company is not made of the addition of only profitable deals but in real life is yielded by a portfolio of good and bad deals. The point is not to get every deal right but to get more good deals than bad deals while maximising profits on the good deals and cutting losses on the bad deals. Being a student of the market using these AI-assisted SaaS… yes. Relying on them to produce the perfect trading decision…no! Occasional imperfect decisions can efficiently be dealt with by a robust risk management system.
Will shipping suffer from departing from the Kiss principle by applying/using AI/big data…to/for what is essentially a very simple market? Only time will tell.
This column has not been generated by ChatGPT.
